paper_id,title,keywords,abstract,meta_review
1,"""Question Generation using a Scratchpad Encoder""","['Question Generation', 'Natural Language Generation', 'Scratchpad Encoder', 'Sequence to Sequence']","""In this paper we introduce the Scratchpad Encoder, a novel addition to the sequence to sequence (seq2seq) framework and explore its effectiveness in generating natural language questions from a given logical form. The Scratchpad encoder enables the decoder at each time step to modify all the encoder outputs, thus using the encoder as a ""scratchpad"" memory to keep track of what has been generated so far and to guide future generation. Experiments on a knowledge based question generation dataset show that our approach generates more fluent and expressive questions according to quantitative metrics and human judgments.""","""This paper introduces a ""scratchpad"" extension to seq2seq models whereby the encoder outputs, typically ""read-only"" during decoding, are editable by the decoder. In practice, this bears quite a lot of similarityif not in the general concept, then in the the implementationto a variety of models proposed in the NLP community (see reviews for details). As the technical novelty of the paper is quite limited, and there are issues with the clarity both in the technical contribution and in presenting what exactly is the main contribution of the paper, I must concur with the reviewers and recommend rejection."""
2,"""DEEP ADVERSARIAL FORWARD MODEL""","['forward model', 'adversarial learning']","""Learning world dynamics has recently been investigated as a way to make reinforcement learning (RL) algorithms to be more sample efficient and interpretable. In this paper, we propose to capture an environment dynamics with a novel forward model that leverages recent works on adversarial learning and visual control. Such a model estimates future observations conditioned on the current ones and other input variables such as actions taken by an RL-agent. We focus on image generation which is a particularly challenging topic but our method can be adapted to other modalities. More precisely, our forward model is trained to produce realistic observations of the future while a discriminator model is trained to distinguish between real images and the models prediction of the future. This approach overcomes the need to define an explicit loss function for the forward model which is currently used for solving such a class of problem. As a consequence, our learning protocol does not have to rely on an explicit distance such as Euclidean distance which tends to produce unsatisfactory predictions. To illustrate our method, empirical qualitative and quantitative results are presented on a real driving scenario, along with qualitative results on Atari game Frostbite.""","""The paper presents an action conditioned video prediction method that combines previous losses in the literature, such as, perceptual, adversarial and infogan type of losses. The reviewers point out the lack of novelty in the formulation, as well as the lack of experiments that would verify its usefulness in model based RL. There is no rebuttal thus no ground for discussion or acceptance."""
3,"""On Random Deep Weight-Tied Autoencoders: Exact Asymptotic Analysis, Phase Transitions, and Implications to Training""","['Random Deep Autoencoders', 'Exact Asymptotic Analysis', 'Phase Transitions']","""We study the behavior of weight-tied multilayer vanilla autoencoders under the assumption of random weights. Via an exact characterization in the limit of large dimensions, our analysis reveals interesting phase transition phenomena when the depth becomes large. This, in particular, provides quantitative answers and insights to three questions that were yet fully understood in the literature. Firstly, we provide a precise answer on how the random deep weight-tied autoencoder model performs approximate inference as posed by Scellier et al. (2018), and its connection to reversibility considered by several theoretical studies. Secondly, we show that deep autoencoders display a higher degree of sensitivity to perturbations in the parameters, distinct from the shallow counterparts. Thirdly, we obtain insights on pitfalls in training initialization practice, and demonstrate experimentally that it is possible to train a deep autoencoder, even with the tanh activation and a depth as large as 200 layers, without resorting to techniques such as layer-wise pre-training or batch normalization. Our analysis is not specific to any depths or any Lipschitz activations, and our analytical techniques may have broader applicability.""","""This paper analyzes random auto encoders in the infinite dimension limit with an assumption that the weights are tied in the encoder and decoder. In the limit the paper is able to show the random auto encoder transformation as doing an approximate inference on data. The paper is able to obtain principled initialization strategies for training deep autoencoders using this analysis, showing the usefulness of their analysis. Even though there are limitations of paper such as studying only random models, and characterizing them only in the limit, all the reviewers agree that the analysis is novel and gives insights on an interesting problem. """
4,"""Graph Spectral Regularization For Neural Network Interpretability""","['autoencoder', 'interpretable', 'graph signal processing', 'graph spectrum', 'graph filter', 'capsule']","""Deep neural networks can learn meaningful representations of data. However, these representations are hard to interpret. For example, visualizing a latent layer is generally only possible for at most three dimensions. Neural networks are able to learn and benefit from much higher dimensional representations but these are not visually interpretable because nodes have arbitrary ordering within a layer. Here, we utilize the ability of the human observer to identify patterns in structured representations to visualize higher dimensions. To do so, we propose a class of regularizations we call \textit{Graph Spectral Regularizations} that impose graph-structure on latent layers. This is achieved by treating activations as signals on a predefined graph and constraining those activations using graph filters, such as low pass and wavelet-like filters. This framework allows for any kind of graph as well as filter to achieve a wide range of structured regularizations depending on the inference needs of the data. First, we show a synthetic example that the graph-structured layer can reveal topological features of the data. Next, we show that a smoothing regularization can impose semantically consistent ordering of nodes when applied to capsule nets. Further, we show that the graph-structured layer, using wavelet-like spatially localized filters, can form localized receptive fields for improved image and biomedical data interpretation. In other words, the mapping between latent layer, neurons and the output space becomes clear due to the localization of the activations. Finally, we show that when structured as a grid, the representations create coherent images that allow for image-processing techniques such as convolutions.""","""The work presents a method of imposing harmonic structural regularizations to layers of a neural network. While the idea is interesting, the reviewers point out multiple issues. Pros: + Interesting method + Hidden layer coherence tends to improve Cons: - Deficient comparisons to baselines or context with other works. - Insufficient assessment of impact to model performance. - Lack of strategy to select regularizers - Lack of evaluation on more realistic datasets"""
5,"""On the Universal Approximability and Complexity Bounds of Quantized ReLU Neural Networks""","['Quantized Neural Networks', 'Universial Approximability', 'Complexity Bounds', 'Optimal Bit-width']","""Compression is a key step to deploy large neural networks on resource-constrained platforms. As a popular compression technique, quantization constrains the number of distinct weight values and thus reducing the number of bits required to represent and store each weight. In this paper, we study the representation power of quantized neural networks. First, we prove the universal approximability of quantized ReLU networks on a wide class of functions. Then we provide upper bounds on the number of weights and the memory size for a given approximation error bound and the bit-width of weights for function-independent and function-dependent structures. Our results reveal that, to attain an approximation error bound of pseudo-formula , the number of weights needed by a quantized network is no more than pseudo-formula times that of an unquantized network. This overhead is of much lower order than the lower bound of the number of weights needed for the error bound, supporting the empirical success of various quantization techniques. To the best of our knowledge, this is the first in-depth study on the complexity bounds of quantized neural networks.""","""This paper addresses a well motivated problem and provides new insight on the theoretical analysis of representational power in quantized networks. The results contribute towards a better understanding of quantized networks in a way that has not been treated in the past. The most moderate rating (marginally above acceptance threshold) explains that while the paper is technically quite simple, it gives an interesting study and blends well into recent literature on an important topic. A criticism is that the approach uses modules to approximate the basic operations of non quantized networks. As such it not compatible with quantizing the weights of a given network structure, but rather with choosing the network structure under a given level of quantization. However, reviewers consider that this issue is discussed directly and clearly in the paper. The reviewers report to be only fairly confident about their assessment, but they all give a positive or very positive evaluation of the paper. """
6,"""Variational Autoencoders with Jointly Optimized Latent Dependency Structure""","['deep generative models', 'structure learning']","""We propose a method for learning the dependency structure between latent variables in deep latent variable models. Our general modeling and inference framework combines the complementary strengths of deep generative models and probabilistic graphical models. In particular, we express the latent variable space of a variational autoencoder (VAE) in terms of a Bayesian network with a learned, flexible dependency structure. The network parameters, variational parameters as well as the latent topology are optimized simultaneously with a single objective. Inference is formulated via a sampling procedure that produces expectations over latent variable structures and incorporates top-down and bottom-up reasoning over latent variable values. We validate our framework in extensive experiments on MNIST, Omniglot, and CIFAR-10. Comparisons to state-of-the-art structured variational autoencoder baselines show improvements in terms of the expressiveness of the learned model.""","""Strengths: This paper develops a method for learning the structure of discrete latent variables in a VAE. The overall approach is well-explained and reasonable. Weaknesses: Ultimately, this is done using the usual style of discrete relaxations, which come with tradeoffs and inconsistencies. Consensus: The reviewers all agreed that the paper is above the bar."""
7,"""Learning State Representations in Complex Systems with Multimodal Data""","['deep learning', 'representation learning', 'state representation', 'disentangled representation', 'dataset', 'autonomous system', 'temporal multimodal data']","""Representation learning becomes especially important for complex systems with multimodal data sources such as cameras or sensors. Recent advances in reinforcement learning and optimal control make it possible to design control algorithms on these latent representations, but the field still lacks a large-scale standard dataset for unified comparison. In this work, we present a large-scale dataset and evaluation framework for representation learning for the complex task of landing an airplane. We implement and compare several approaches to representation learning on this dataset in terms of the quality of simple supervised learning tasks and disentanglement scores. The resulting representations can be used for further tasks such as anomaly detection, optimal control, model-based reinforcement learning, and other applications.""","""The paper introduces a new dataset that contains multiple landings from the X plane simulator, and each includes readings from multiple sensors for aircraft landing. The paper also trains a set of self-supervised methods presented in previous works in order to learn sensory representations, and evaluates the learnt representations in terms of disentanglement and re-purposing to a discriminative task. Though the evaluations presented are interesting, they are not convincingly useful, as noted by the reviewers. Overall, it is not clear why this dataset is particularly well suited for representation learning. Furthermore, it is difficult to evaluate representation learning methods without relating them to an end-task, e.g., that of landing the aircraft. The paper writing would also benefit from restructuring and improving on English expressions. In particular, the conclusion section contains half-finished sentences. """
8,"""A NON-LINEAR THEORY FOR SENTENCE EMBEDDING""","['sentence embedding', 'generative models']","""This paper revisits the Random Walk model for sentence embedding in the context of non-extensive statistics. We propose a non-extensive algebra to compute the discourse vector. We argue that by doing so we are taking into account high non-linearity in the semantic space. Furthermore, we show that by considering a non-extensive algebra, the compounding effect of the vector length is mitigated. Overall, we show that the proposed model leads to good sentence embedding. We evaluate the embedding method on textual similarity tasks.""","""The paper is poorly written and below the bar of ICLR. The paper could be improved with better exposition and stronger experiment results (or clearer exposition of the experimental results.) """
9,"""Reinforcement Learning with Perturbed Rewards""","['robust reinforcement learning', 'noisy reward', 'sample complexity']","""Recent studies have shown the vulnerability of reinforcement learning (RL) models in noisy settings. The sources of noises differ across scenarios. For instance, in practice, the observed reward channel is often subject to noise (e.g., when observed rewards are collected through sensors), and thus observed rewards may not be credible as a result. Also, in applications such as robotics, a deep reinforcement learning (DRL) algorithm can be manipulated to produce arbitrary errors. In this paper, we consider noisy RL problems where observed rewards by RL agents are generated with a reward confusion matrix. We call such observed rewards as perturbed rewards. We develop an unbiased reward estimator aided robust RL framework that enables RL agents to learn in noisy environments while observing only perturbed rewards. Our framework draws upon approaches for supervised learning with noisy data. The core ideas of our solution include estimating a reward confusion matrix and defining a set of unbiased surrogate rewards. We prove the convergence and sample complexity of our approach. Extensive experiments on different DRL platforms show that policies based on our estimated surrogate reward can achieve higher expected rewards, and converge faster than existing baselines. For instance, the state-of-the-art PPO algorithm is able to obtain 67.5% and 46.7% improvements in average on five Atari games, when the error rates are 10% and 30% respectively. ""","""This paper studies RL with perturbed rewards, where a technical challenge is to revert the perturbation process so that the right policy is learned. Some experiments are used to support the algorithm, which involves learning the reward perturbation process (the confusion matrix) using existing techniques from the supervised learning (and crowdsourcing) literature. Reviewers found the problem setting new and worth investigating, but had concerns over the scope/significance of this work, mostly about how the confusion matrix is learned. If this matrix is known, correcting reward perturbation is easy, and standard RL can be applied to the corrected rewards. Specifically, the work seems to be limited in two substantial ways, both related to how the confusion matrix is learned. * The reward function needs to be deterministic. * Majority voting requires the number of states to be finite. The significance of this work is therefore mostly limited to finite-state problems with deterministic reward, which is quite restricted. As the authors pointed out, the paper uses discretization to turn a continuous state space into a finite one, which is how the experiment was done. But discretization is likely not robust or efficient in many high-dimensional problems. It should be noted that the setting studied here, together with a thorough treatment of an (even restricted) case, could make an interesting paper that inspires future work. However, the exact problem setting is not completely clear in the paper, and the limitations of the technical contributions is also somewhat unclear. The authors are strongly advised to revise the paper accordingly to make their contributions clearer. Minor questions: - In lemma 2, what if C is not invertible. - The sampling oracle assumed in def 1 is not very practical, as opposed to what the paper claims. - There are more recent work at NIPS and STOC on attacking RL (including bandits) algorithms by manipulating the reward signals. The authors may want to cite and discuss."""
10,"""FROM DEEP LEARNING TO DEEP DEDUCING: AUTOMATICALLY TRACKING DOWN NASH EQUILIBRIUM THROUGH AUTONOMOUS NEURAL AGENT, A POSSIBLE MISSING STEP TOWARD GENERAL A.I.""","['Reinforcement Learning', 'Deep Feed-forward Neural Network', 'Recurrent Neural Network', 'Game Theory', 'Control Theory', 'Nash Equilibrium', 'Optimization']","""Contrary to most reinforcement learning studies, which emphasize on training a deep neural network to approximate its output layer to certain strategies, this paper proposes a reversed method for reinforcement learning. We call this Deep Deducing. In short, after adequately training a deep neural network according to a strategy-environment-to-payoff table, then we initialize randomized strategy input and propagate the error between the actual output and the desired output back to the initially-randomized strategy input in the input layer of the trained deep neural network gradually to perform a task similar to human deduction. And we view the final strategy input in the input layer as the fittest strategy for a neural network when confronting the observed environment input from the world outside.""","""The paper presents ""deep deducing"", which means learning the state-action value function of 2 player games from a payoff table, and using the value function by maximizing over the (actionable) inputs at test time. The paper lacks clarity overall. The method does not contain any new model nor algorithm. The experiments are too weak (easy environments, few/no comparisons) to support the claims. The paper is not ready for publication at this time."""
11,"""Music Transformer: Generating Music with Long-Term Structure""",['music generation'],"""Music relies heavily on repetition to build structure and meaning. Self-reference occurs on multiple timescales, from motifs to phrases to reusing of entire sections of music, such as in pieces with ABA structure. The Transformer (Vaswani et al., 2017), a sequence model based on self-attention, has achieved compelling results in many generation tasks that require maintaining long-range coherence. This suggests that self-attention might also be well-suited to modeling music. In musical composition and performance, however, relative timing is critically important. Existing approaches for representing relative positional information in the Transformer modulate attention based on pairwise distance (Shaw et al., 2018). This is impractical for long sequences such as musical compositions since their memory complexity is quadratic in the sequence length. We propose an algorithm that reduces the intermediate memory requirements to linear in the sequence length. This enables us to demonstrate that a Transformer with our modified relative attention mechanism can generate minute-long (thousands of steps) compositions with compelling structure, generate continuations that coherently elaborate on a given motif, and in a seq2seq setup generate accompaniments conditioned on melodies. We evaluate the Transformer with our relative attention mechanism on two datasets, JSB Chorales and Piano-e-competition, and obtain state-of-the-art results on the latter.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - improvements to a transformer model originally designed for machine translation - application of this model to a different task: music generation - compelling generated samples and user study. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - lack of clarity at times (much improved in the revised version) 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. The main contention was novelty. Some reviewers felt that adapting an existing transformer model to music generation and achieving SOTA results and minute-long music sequences was not sufficient novelty. The final decision aligns with the reviewers who felt that the novelty was sufficient. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. A consensus was not reached. The final decision is aligned with the positive reviews for the reason mentioned above. """
12,"""Three Mechanisms of Weight Decay Regularization""","['Generalization', 'Regularization', 'Optimization']","""Weight decay is one of the standard tricks in the neural network toolbox, but the reasons for its regularization effect are poorly understood, and recent results have cast doubt on the traditional interpretation in terms of pseudo-formula regularization. Literal weight decay has been shown to outperform pseudo-formula regularization for optimizers for which they differ. We empirically investigate weight decay for three optimization algorithms (SGD, Adam, and K-FAC) and a variety of network architectures. We identify three distinct mechanisms by which weight decay exerts a regularization effect, depending on the particular optimization algorithm and architecture: (1) increasing the effective learning rate, (2) approximately regularizing the input-output Jacobian norm, and (3) reducing the effective damping coefficient for second-order optimization. Our results provide insight into how to improve the regularization of neural networks.""","""Reviewers are in a consensus and recommended to accept after engaging with the authors. Please take reviewers' comments into consideration to improve your submission for the camera ready. """
13,"""Do Language Models Have Common Sense?""",[],"""It has been argued that current machine learning models do not have commonsense, and therefore must be hard-coded with prior knowledge (Marcus, 2018). Here we show surprising evidence that language models can already learn to capture certain common sense knowledge. Our key observation is that a language model can compute the probability of any statement, and this probability can be used to evaluate the truthfulness of that statement. On the Winograd Schema Challenge (Levesque et al., 2011), language models are 11% higher in accuracy than previous state-of-the-art supervised methods. Language models can also be fine-tuned for the task of Mining Commonsense Knowledge on ConceptNet to achieve an F1 score of 0.912 and 0.824, outperforming previous best results (Jastrzebskiet al., 2018). Further analysis demonstrates that language models can discover unique features of Winograd Schema contexts that decide the correct answers without explicit supervision.""","""This paper adapts language models (LMs), recurrent models trained on large corpus to produce the next word in English, to two commonsense reasoning tasks: the Winograd schema challenge and commonsense knowledge extraction. For the former, the language model score itself is used to obtain substantial gains over existing approaches for this challenging task, while a slightly more involved training procedure adapts the LMs to commonsense extraction. The reviewers appreciated the simplicity of the changes to existing LMs and the impressive results (especially on the WSC). The reviewers point out the following potential weaknesses: (1) clarity issues in the writing and the presentation, (2) a lack of novelty in the proposed approach, given a number of recent work has shown the ability of language models to perform commonsense reasoning, and (3) critical methodological issues in the evaluation that raise questions about the significance of the results. A lack of response from the authors meant that there was no further discussion needed, and the reviewers encourage the authors to take the feedback to improve further versions of the paper."""
14,"""ATTENTIVE EXPLAINABILITY FOR PATIENT TEMPORAL EMBEDDING""",[],"""Learning explainable patient temporal embeddings from observational data has mostly ignored the use of RNN architecture that excel in capturing temporal data dependencies but at the expense of explainability. This paper addresses this problem by introducing and applying an information theoretic approach to estimate the degree of explainability of such architectures. Using a communication paradigm, we formalize metrics of explainability by estimating the amount of information that an AI model needs to convey to a human end user to explain and rationalize its outputs. A key aspect of this work is to model human prior knowledge at the receiving end and measure the lack of explainability as a deviation from human prior knowledge. We apply this paradigm to medical concept representation problems by regularizing loss functions of temporal autoencoders according to the derived explainability metrics to guide the learning process towards models producing explainable outputs. We illustrate the approach with convincing experimental results for the generation of explainable temporal embeddings for critical care patient data.""","""The paper proposes an approach to define an ""interpretable representation"", in particular for the case of patient condition monitoring. Reviewers point to several concerns, including even the definition of explainability and limited significance. The authors tried to address the concerns but reviewers think the paper is not ready for acceptance. I concur with them in rejecting it."""
15,"""Double Neural Counterfactual Regret Minimization""","['Counterfactual Regret Minimization', 'Imperfect Information game']","""Counterfactual regret minimization (CRF) is a fundamental and effective technique for solving imperfect information games. However, the original CRF algorithm only works for discrete state and action spaces, and the resulting strategy is maintained as a tabular representation. Such tabular representation limits the method from being directly applied to large games and continuing to improve from a poor strategy profile. In this paper, we propose a double neural representation for the Imperfect Information Games, where one neural network represents the cumulative regret, and the other represents the average strategy. Furthermore, we adopt the counterfactual regret minimization algorithm to optimize this double neural representation. To make neural learning efficient, we also developed several novel techniques including a robust sampling method, mini-batch Monte Carlo counterfactual regret minimization (MCCFR) and Monte Carlo counterfactual regret minimization plus (MCCFR+) which may be of independent interests. Experimentally, we demonstrate that the proposed double neural algorithm converges significantly better than the reinforcement learning counterpart. ""","""The reviewers agreed that there are some promising ideas in this work, and useful empirical analysis to motivate the approach. The main concern is in the soundness of the approach (for example, comments about cumulative learning and negative samples). The authors provided some justification about using previous networks as initialization, but this is an insufficient discussion to understand the soundness of the strategy. The paper should better discuss this more, even if it is not possible to provide theory. The paper could also be improved with the addition of a baseline (though not necessarily something like DeepStack, which is not publicly available and potentially onerous to reimplement). """
16,"""Interactive Parallel Exploration for Reinforcement Learning in Continuous Action Spaces""","['reinforcement learning', 'continuous action space RL']","""In this paper, a new interactive parallel learning scheme is proposed to enhance the performance of off-policy continuous-action reinforcement learning. In the proposed interactive parallel learning scheme, multiple identical learners with their own value-functions and policies share a common experience replay buffer, and search a good policy in collaboration with the guidance of the best policy information. The information of the best policy is fused in a soft manner by constructing an augmented loss function for policy update to enlarge the overall search space by the multiple learners. The guidance by the previous best policy and the enlarged search space by the proposed interactive parallel learning scheme enable faster and better policy search in the policy parameter space. Working algorithms are constructed by applying the proposed interactive parallel learning scheme to several off-policy reinforcement learning algorithms such as the twin delayed deep deterministic (TD3) policy gradient algorithm and the soft actor-critic (SAC) algorithm, and numerical results show that the constructed IPE-enhanced algorithms outperform most of the current state-of-the-art reinforcement learning algorithms for continuous action control.""","""The authors present a new method for leveraging multiple parallel agents to speed RL in continuous action spaces. By monitoring the best performers, that information can be shared in a soft way to speed policy search. The problem space is interesting and faster learning is important. However, multiple reviewers [R2, R1] had significant concerns with how the work is framed with respect to the wider literature (even after the revisions), and some concerns over the significance of the performance improvements which seem primarily to come from early boosts. There is also additional related work on concurrent RL (Guo and Brunskill 2015; Dimakopoulou, Van Roy 2018 ; Dimakopoulou, Osband, Van Roy 2018) which provides some more formal considerations of the setting the authors consider, which would be good to reference. """
17,"""Unsupervised Document Representation using Partition Word-Vectors Averaging""","['Unsupervised Learning', 'Natural Language Processing', 'Representation Learning', 'Document Embedding']","""Learning effective document-level representation is essential in many important NLP tasks such as document classification, summarization, etc. Recent research has shown that simple weighted averaging of word vectors is an effective way to represent sentences, often outperforming complicated seq2seq neural models in many tasks. While it is desirable to use the same method to represent documents as well, unfortunately, the effectiveness is lost when representing long documents involving multiple sentences. One reason for this degradation is due to the fact that a longer document is likely to contain words from many different themes (or topics), and hence creating a single vector while ignoring all the thematic structure is unlikely to yield an effective representation of the document. This problem is less acute in single sentences and other short text fragments where presence of a single theme/topic is most likely. To overcome this problem, in this paper we present PSIF, a partitioned word averaging model to represent long documents. P-SIF retains the simplicity of simple weighted word averaging while taking a document's thematic structure into account. In particular, P-SIF learns topic-specific vectors from a document and finally concatenates them all to represent the overall document. Through our experiments over multiple real-world datasets and tasks, we demonstrate PSIF's effectiveness compared to simple weighted averaging and many other state-of-the-art baselines. We also show that PSIF is particularly effective in representing long multi-sentence documents. We will release PSIF's embedding source code and data-sets for reproducing results.""","""This paper proposes a document classification algorithm based on partitioned word vector averaging. I agree with even the most positive reviewer. More experiments would be good. This is a very developed old area."""
18,"""Approximation capability of neural networks on sets of probability measures and tree-structured data""","['multi-instance learning', 'hierarchical models', 'universal approximation theorem']","""This paper extends the proof of density of neural networks in the space of continuous (or even measurable) functions on Euclidean spaces to functions on compact sets of probability measures. By doing so the work parallels a more then a decade old results on mean-map embedding of probability measures in reproducing kernel Hilbert spaces. The work has wide practical consequences for multi-instance learning, where it theoretically justifies some recently proposed constructions. The result is then extended to Cartesian products, yielding universal approximation theorem for tree-structured domains, which naturally occur in data-exchange formats like JSON, XML, YAML, AVRO, and ProtoBuffer. This has important practical implications, as it enables to automatically create an architecture of neural networks for processing structured data (AutoML paradigms), as demonstrated by an accompanied library for JSON format.""","""Several reviewers thought the results were not surprising in light of existing universality results, and thought the results were of limited relevance, given that the formalization is not quite in line with real-world networks for MIL. The authors draw out some further justifications in the rebuttal. These should be reintegrated. I agree with the general criticisms regarding relevance to ICLR. Ultimately, this work may belong in a journal."""
19,"""ACE: Artificial Checkerboard Enhancer to Induce and Evade Adversarial Attacks""","['Adversarial Examples', 'Neural Network Security', 'Deep Neural Network', 'Checkerboard Artifact']","""The checkerboard phenomenon is one of the well-known visual artifacts in the computer vision field. The origins and solutions of checkerboard artifacts in the pixel space have been studied for a long time, but their effects on the gradient space have rarely been investigated. In this paper, we revisit the checkerboard artifacts in the gradient space which turn out to be the weak point of a network architecture. We explore image-agnostic property of gradient checkerboard artifacts and propose a simple yet effective defense method by utilizing the artifacts. We introduce our defense module, dubbed Artificial Checkerboard Enhancer (ACE), which induces adversarial attacks on designated pixels. This enables the model to deflect attacks by shifting only a single pixel in the image with a remarkable defense rate. We provide extensive experiments to support the effectiveness of our work for various attack scenarios using state-of-the-art attack methods. Furthermore, we show that ACE is even applicable to large-scale datasets including ImageNet dataset and can be easily transferred to various pretrained networks.""","""The reviewers have agreed this work is not ready for publication at ICLR."""
20,"""DyRep: Learning Representations over Dynamic Graphs""","['Dynamic Graphs', 'Representation Learning', 'Dynamic Processes', 'Temporal Point Process', 'Attention', 'Latent Representation']","""Representation Learning over graph structured data has received significant attention recently due to its ubiquitous applicability. However, most advancements have been made in static graph settings while efforts for jointly learning dynamic of the graph and dynamic on the graph are still in an infant stage. Two fundamental questions arise in learning over dynamic graphs: (i) How to elegantly model dynamical processes over graphs? (ii) How to leverage such a model to effectively encode evolving graph information into low-dimensional representations? We present DyRep - a novel modeling framework for dynamic graphs that posits representation learning as a latent mediation process bridging two observed processes namely -- dynamics of the network (realized as topological evolution) and dynamics on the network (realized as activities between nodes). Concretely, we propose a two-time scale deep temporal point process model that captures the interleaved dynamics of the observed processes. This model is further parameterized by a temporal-attentive representation network that encodes temporally evolving structural information into node representations which in turn drives the nonlinear evolution of the observed graph dynamics. Our unified framework is trained using an efficient unsupervised procedure and has capability to generalize over unseen nodes. We demonstrate that DyRep outperforms state-of-the-art baselines for dynamic link prediction and time prediction tasks and present extensive qualitative insights into our framework.""","""After discussion, all reviewers agree to accept this paper. Congratulations!!"""
21,"""Per-Tensor Fixed-Point Quantization of the Back-Propagation Algorithm""","['deep learning', 'reduced precision', 'fixed-point', 'quantization', 'back-propagation algorithm']","""The high computational and parameter complexity of neural networks makes their training very slow and difficult to deploy on energy and storage-constrained comput- ing systems. Many network complexity reduction techniques have been proposed including fixed-point implementation. However, a systematic approach for design- ing full fixed-point training and inference of deep neural networks remains elusive. We describe a precision assignment methodology for neural network training in which all network parameters, i.e., activations and weights in the feedforward path, gradients and weight accumulators in the feedback path, are assigned close to minimal precision. The precision assignment is derived analytically and enables tracking the convergence behavior of the full precision training, known to converge a priori. Thus, our work leads to a systematic methodology of determining suit- able precision for fixed-point training. The near optimality (minimality) of the resulting precision assignment is validated empirically for four networks on the CIFAR-10, CIFAR-100, and SVHN datasets. The complexity reduction arising from our approach is compared with other fixed-point neural network designs.""","""The paper investigates a detailed analysis of reduced precision training for a feedforward network, that accounts for both the forward and backward passes in detail. It is shown that precision can be greatly reduced throughout the network computations while largely preserving training quality. The analysis is thorough and carefully executed. The technical presentation, including the motivation for some of the specific choices should be made clearer. Also, the requirement that the network first be trained to convergence at full 32 bit precision is a significant limitation of the proposed approach (a weakness that is shared with other work in this area). It would be highly desirable to find ways to bypass or at least mitigate this requirement, which would provide a real breakthrough rather than merely a solid improvement over competing work. The reviewer disagreement revolves primarily around the clarity of the main technical exposition: there appears to be consensus that the paper is sound and provides a serious contribution to this area. Although the persistent reviewer disagreement left this paper rated at the borderline, I am recommending acceptance, with the understanding that the authors will not disregard the dissenting review and strive to further improve the clarity of the presentation."""
22,"""Decoupling feature extraction from policy learning: assessing benefits of state representation learning in goal based robotics""","['reinforcement learning', 'state representation learning', 'feature extraction', 'robotics', 'deep learning']","""Scaling end-to-end reinforcement learning to control real robots from vision presents a series of challenges, in particular in terms of sample efficiency. Against end-to-end learning, state representation learning can help learn a compact, efficient and relevant representation of states that speeds up policy learning, reducing the number of samples needed, and that is easier to interpret. We evaluate several state representation learning methods on goal based robotics tasks and propose a new unsupervised model that stacks representations and combines strengths of several of these approaches. This method encodes all the relevant features, performs on par or better than end-to-end learning, and is robust to hyper-parameters change.""","""This paper proposes a new method for combining previous state representation learning methods and compares to end-to-end learning without without separately learning a state representation. The topic is important, and the authors have made an extensive effort to address the reviewer's concerns, particularly regarding clarity, related work, and accuracy of the drawn conclusions. The reviewers found that the main weakness of the paper was the experiments not being sufficiently convincing that the proposed approach is better than the alternatives. Hence, it does not currently meet the bar for publication."""
23,"""Unseen Action Recognition with Unpaired Adversarial Multimodal Learning""",[],"""In this paper, we present a method to learn a joint multimodal representation space that allows for the recognition of unseen activities in videos. We compare the effect of placing various constraints on the embedding space using paired text and video data. Additionally, we propose a method to improve the joint embedding space using an adversarial formulation with unpaired text and video data. In addition to testing on publicly available datasets, we introduce a new, large-scale text/video dataset. We experimentally confirm that learning such shared embedding space benefits three difficult tasks (i) zero-shot activity classification, (ii) unsupervised activity discovery, and (iii) unseen activity captioning. ""","""The paper received mixed reviews. The proposed ideas are reasonable and it shows that unpaired data can improve the performance of unseen video (action) classification tasks and other related tasks. The authors rightfully argue that the main contribution is the use of unpaired, multimodal data for learning a joint embedding (that generalizes to unseen actions) with positive results, but not the use of attentional pooling mechanism. Despite this, as the Reviewer3 points out, technical novelty seems minor as there are quite many papers on learning joint embedding for multimodal data. Many of these works were evaluated for fine-grained image classification setting, but there is no reason that such methods cannot be used here. The revision only compares against methods published in 2017 or before. So more comprehensive evaluation would be needed to fully justify the proposed method. In addition, it seems that the proposed method has fairly marginal gain for the generalized zero-shot learning setting. Overall, the paper can be viewed as an application paper on unseen action recognition tasks but the technical novelty and more rigorous comparisons against recent related work are somewhat lacking. I recommend rejection due to several concerns raised here and by the reviewers. """
24,"""Look Ma, No GANs! Image Transformation with ModifAE""","['Computer Vision', 'Deep Learning', 'Autoencoder', 'GAN', 'Image Modification', 'Social Traits', 'Social Psychology']","""Existing methods of image to image translation require multiple steps in the training or modification process, and suffer from either an inability to generalize, or long training times. These methods also focus on binary trait modification, ignoring continuous traits. To address these problems, we propose ModifAE: a novel standalone neural network, trained exclusively on an autoencoding task, that implicitly learns to make continuous trait image modifications. As a standalone image modification network, ModifAE requires fewer parameters and less time to train than existing models. We empirically show that ModifAE produces significantly more convincing and more consistent continuous face trait modifications than the previous state-of-the-art model.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The paper tackles an interesting and relevant problem for ICLR: guided image modification of images (in this case of facial attributes). - The proposed method is in general well-explained (although some details are lacking) 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The training set of faces and associated attributes were annotated using a pre-trained model which introduced a bias into the annotations used for training the method. - The experimental results weren't convincing. The qualitative results showed no clear advantage of the proposed method and the quantitative comparison to StarGAN only considered two attribute manipulations and only found a statistically significant different in performance for one of those. The second weakness was the key determining factor in the AC's final recommendation. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. There were no major points of contention and no author feedback. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be rejected. """
25,"""Fast Binary Functional Search on Graph""","['Binary Functional Search', 'Large-scale Search', 'Approximate Nearest Neighbor Search']","""The large-scale search is an essential task in modern information systems. Numerous learning based models are proposed to capture semantic level similarity measures for searching or ranking. However, these measures are usually complicated and beyond metric distances. As Approximate Nearest Neighbor Search (ANNS) techniques have specifications on metric distances, efficient searching by advanced measures is still an open question. In this paper, we formulate large-scale search as a general task, Optimal Binary Functional Search (OBFS), which contains ANNS as special cases. We analyze existing OBFS methods' limitations and explain they are not applicable for complicated searching measures. We propose a flexible graph-based solution for OBFS, Search on L2 Graph (SL2G). SL2G approximates gradient decent in Euclidean space, with accessible conditions. Experiments demonstrate SL2G's efficiency in searching by advanced matching measures (i.e., Neural Network based measures).""","""This paper proposes an Optimal Binary Functional Search (OBFS) algorithm for searching with general score functions, which generalizes the standard similarity measures based on Euclidean distances. This yields an extension of the classical approximate nearest neighbor search (ANNS). As observed by the reviewers, this work targets an important research direction. Unfortunately, the reviewers raised several concerns regarding the clarity and significance of the work. The authors provided a good rebuttal and addressed some concerns, but not to the degree that reviewers think it passes the bar of ICLR. We encourage the authors to further improve the work to address the key concerns. """
26,"""Exploring and Enhancing the Transferability of Adversarial Examples""","['Deep learning', 'Adversarial example', 'Transferability', 'Smoothed gradient']",""" State-of-the-art deep neural networks are vulnerable to adversarial examples, formed by applying small but malicious perturbations to the original inputs. Moreover, the perturbations can \textit{transfer across models}: adversarial examples generated for a specific model will often mislead other unseen models. Consequently the adversary can leverage it to attack deployed systems without any query, which severely hinders the application of deep learning, especially in the safety-critical areas. In this work, we empirically study how two classes of factors those might influence the transferability of adversarial examples. One is about model-specific factors, including network architecture, model capacity and test accuracy. The other is the local smoothness of loss surface for constructing adversarial examples. Inspired by these understandings on the transferability of adversarial examples, we then propose a simple but effective strategy to enhance the transferability, whose effectiveness is confirmed by a variety of experiments on both CIFAR-10 and ImageNet datasets.""","""While the paper contains significant information, most insights have already been revealed in previous work as noted by R1. The empirical novelty is therefore limited and the authors do not provide theoretical analysis to complement this."""
27,"""Manifold Alignment via Feature Correspondence""","['graph signal processing', 'graph alignment', 'manifold alignment', 'spectral graph wavelet transform', 'diffusion geometry', 'harmonic analysis']","""We propose a novel framework for combining datasets via alignment of their associated intrinsic dimensions. Our approach assumes that the two datasets are sampled from a common latent space, i.e., they measure equivalent systems. Thus, we expect there to exist a natural (albeit unknown) alignment of the data manifolds associated with the intrinsic geometry of these datasets, which are perturbed by measurement artifacts in the sampling process. Importantly, we do not assume any individual correspondence (partial or complete) between data points. Instead, we rely on our assumption that a subset of data features have correspondence across datasets. We leverage this assumption to estimate relations between intrinsic manifold dimensions, which are given by diffusion map coordinates over each of the datasets. We compute a correlation matrix between diffusion coordinates of the datasets by considering graph (or manifold) Fourier coefficients of corresponding data features. We then orthogonalize this correlation matrix to form an isometric transformation between the diffusion maps of the datasets. Finally, we apply this transformation to the diffusion coordinates and construct a unified diffusion geometry of the datasets together. We show that this approach successfully corrects misalignment artifacts, and allows for integrated data.""","""The diffusion maps framework is used to embed a given collection of datasets into diffusion coordinates that capture intrinsic geometry. Then a correspondence map is constructed between datasets by finding rotations that align these coordinates. The approach is interesting. The reviewers, however, found the empirical analysis somewhat simplistic with inadequate comparisons to other correspondence construction methods in the literature. """
28,"""Guaranteed Recovery of One-Hidden-Layer Neural Networks via Cross Entropy""","['cross entropy', 'neural networks', 'parameter recovery']","""We study model recovery for data classification, where the training labels are generated from a one-hidden-layer fully -connected neural network with sigmoid activations, and the goal is to recover the weight vectors of the neural network. We prove that under Gaussian inputs, the empirical risk function using cross entropy exhibits strong convexity and smoothness uniformly in a local neighborhood of the ground truth, as soon as the sample complexity is sufficiently large. This implies that if initialized in this neighborhood, which can be achieved via the tensor method, gradient descent converges linearly to a critical point that is provably close to the ground truth without requiring a fresh set of samples at each iteration. To the best of our knowledge, this is the first global convergence guarantee established for the empirical risk minimization using cross entropy via gradient descent for learning one-hidden-layer neural networks, at the near-optimal sample and computational complexity with respect to the network input dimension.""","""This paper shows local convergence results for gradient descent on one hidden layer network with Gaussian inputs and sigmoid activations. Later it shows global convergence by using spectral initialization. All the reviewers agree that the results are similar to existing work in the literature with little novelty. There are also some concerns about the correctness of the statements expressed by some reviewers. """
29,"""Second-Order Adversarial Attack and Certifiable Robustness""",[],"""Adversarial training has been recognized as a strong defense against adversarial attacks. In this paper, we propose a powerful second-order attack method that reduces the accuracy of the defense model by Madry et al. (2017). We demonstrate that adversarial training overfits to the choice of the norm in the sense that it is only robust to the attack used for adversarial training, thus suggesting it has not achieved universal robustness. The effectiveness of our attack method motivates an investigation of provable robustness of a defense model. To this end, we introduce a framework that allows one to obtain a certifiable lower bound on the prediction accuracy against adversarial examples. We conduct experiments to show the effectiveness of our attack method. At the same time, our defense model achieves significant improvements compared to previous works under our proposed attack.""","""The reviewers have agreed this work is not ready for publication at ICLR."""
30,"""Precision Highway for Ultra Low-precision Quantization""","['neural network', 'quantization', 'optimization', 'low-precision', 'convolutional network', 'recurrent network']","""Quantization of a neural network has an inherent problem called accumulated quantization error, which is the key obstacle towards ultra-low precision, e.g., 2- or 3-bit precision. To resolve this problem, we propose precision highway, which forms an end-to-end high-precision information flow while performing the ultra-low-precision computation. First, we describe how the precision highway reduce the accumulated quantization error in both convolutional and recurrent neural networks. We also provide the quantitative analysis of the benefit of precision highway and evaluate the overhead on the state-of-the-art hardware accelerator. In the experiments, our proposed method outperforms the best existing quantization methods while offering 3-bit weight/activation quantization with no accuracy loss and 2-bit quantization with a 2.45 % top-1 accuracy loss in ResNet-50. We also report that the proposed method significantly outperforms the existing method in the 2-bit quantization of an LSTM for language modeling.""","""The submission proposes a strategy for quantization of neural networks with skip connections that quantizes only the convolution paths, while leaving the skip paths at full precision. The approach can save computation through compressing the convolution kernels, while spending more on the skip connections. Empirical results show improved performance at 2-bit quantization compared to a handful of competing methods. Figure 5 provides some interpretation of why the method might be working in terms of ""smoothness"" of the loss surface (term not used in the traditional mathematical sense). The paper seems to focus too much on selling the name ""precision highway"" rather than providing proper definitions of their strategy (a definition block would be a good first step), and there is little mathematical analysis of the consequences of the chosen approach. There are concerns about the novelty of the method, specifically compared to Liu et al. (2018) and Choi et al. (2018b), which propose approximately the same strategy. Footnote 1 claims that these works were conducted in parallel with the current submission, but it is unambiguously the case that Choi et al appeared on arXiv in May, and Liu et al. appeared in ECCV 2018 and on arXiv more than 30 days before the ICLR deadline, and can fairly be considered prior work pseudo-url The reviewer scores were on aggregate borderline for the ICLR acceptance threshold. On the balance, the paper seems to fall under the threshold due to insufficient novelty and analysis of the method. """
31,"""Unsupervised Disentangling Structure and Appearance""","['disentangled representations', 'VAE', 'generative models', 'unsupervised learning']","""It is challenging to disentangle an object into two orthogonal spaces of structure and appearance since each can influence the visual observation in a different and unpredictable way. It is rare for one to have access to a large number of data to help separate the influences. In this paper, we present a novel framework to learn this disentangled representation in a completely unsupervised manner. We address this problem in a two-branch Variational Autoencoder framework. For the structure branch, we project the latent factor into a soft structured point tensor and constrain it with losses derived from prior knowledge. This encourages the branch to distill geometry information. Another branch learns the complementary appearance information. The two branches form an effective framework that can disentangle object's structure-appearance representation without any human annotation. We evaluate our approach on four image datasets, on which we demonstrate the superior disentanglement and visual analogy quality both in synthesis and real-world data. We are able to generate photo-realistic images with 256*256 resolution that are clearly disentangled in structure and appearance.""","""With an average review score of 4.67 and a short review for the one positive review it is just not possible to accept the paper."""
32,"""RelGAN: Relational Generative Adversarial Networks for Text Generation""","['RelGAN', 'text generation', 'relational memory', 'Gumbel-Softmax relaxation', 'multiple embedded representations']","""Generative adversarial networks (GANs) have achieved great success at generating realistic images. However, the text generation still remains a challenging task for modern GAN architectures. In this work, we propose RelGAN, a new GAN architecture for text generation, consisting of three main components: a relational memory based generator for the long-distance dependency modeling, the Gumbel-Softmax relaxation for training GANs on discrete data, and multiple embedded representations in the discriminator to provide a more informative signal for the generator updates. Our experiments show that RelGAN outperforms current state-of-the-art models in terms of sample quality and diversity, and we also reveal via ablation studies that each component of RelGAN contributes critically to its performance improvements. Moreover, a key advantage of our method, that distinguishes it from other GANs, is the ability to control the trade-off between sample quality and diversity via the use of a single adjustable parameter. Finally, RelGAN is the first architecture that makes GANs with Gumbel-Softmax relaxation succeed in generating realistic text.""",""" pros: - well-written and clear - good evaluation with convincing ablations - moderately novel cons: - Reviewers 1 and 3 feel the paper is somewhat incremental over previous work, combining previously proposed ideas. (Reviewer 2 originally had concerns about the testing methodology but feels that the paper has improved in revision) (Reviewer 3 suggests an additional comparison to related work which was addressed in revision) I appreciate the authors' revisions and engagement during the discussion period. Overall the paper is good and I'm recommending acceptance."""
33,"""Object-Oriented Model Learning through Multi-Level Abstraction""","['action-conditioned dynamics learning', 'deep learning', 'generalization', 'interpretability', 'sample efficiency']","""Object-based approaches for learning action-conditioned dynamics has demonstrated promise for generalization and interpretability. However, existing approaches suffer from structural limitations and optimization difficulties for common environments with multiple dynamic objects. In this paper, we present a novel self-supervised learning framework, called Multi-level Abstraction Object-oriented Predictor (MAOP), for learning object-based dynamics models from raw visual observations. MAOP employs a three-level learning architecture that enables efficient dynamics learning for complex environments with a dynamic background. We also design a spatial-temporal relational reasoning mechanism to support instance-level dynamics learning and handle partial observability. Empirical results show that MAOP significantly outperforms previous methods in terms of sample efficiency and generalization over novel environments that have multiple controllable and uncontrollable dynamic objects and different static object layouts. In addition, MAOP learns semantically and visually interpretable disentangled representations.""","""This paper tackles a very valuable problem of learning object detection and object dynamics from video sequences, and builds upon the method of Zhu et al. 2018. The reviewers point out that there is a lot of engineering steps in the object proposal stage, which takes into account background subtraction to propose objects. In its current form, the writing of the paper is not clear enough on the object instantiation part, which is also the novel part over Zhu et al., potentially due to the complexity of using motion to guide object proposals. A limitation of the proposed formulation is that it works for moving cameras but only in 2d environments. Experiments on 3D environments would make this paper a much stronger submission. """
34,"""Learning Programmatically Structured Representations with Perceptor Gradients""","['representation learning', 'structured representations', 'symbols', 'programs']","""We present the perceptor gradients algorithm -- a novel approach to learning symbolic representations based on the idea of decomposing an agent's policy into i) a perceptor network extracting symbols from raw observation data and ii) a task encoding program which maps the input symbols to output actions. We show that the proposed algorithm is able to learn representations that can be directly fed into a Linear-Quadratic Regulator (LQR) or a general purpose A* planner. Our experimental results confirm that the perceptor gradients algorithm is able to efficiently learn transferable symbolic representations as well as generate new observations according to a semantically meaningful specification. ""","""This paper considers the problem of learning symbolic representations from raw data. The reviewers are split on the importance of the paper. The main argument in favor of acceptance is that bridges neural and symbolic approaches in the reinforcement learning problem domain, whereas most previous work that have attempted to bridge this gap have been in inverse graphics or physical dynamics settings. Hence, it makes for a contribution that is relevant to the ICLR community. The main downside is that the paper does not provide particularly surprising insights, and could become much stronger with more complex experimental domains. It seems like the benefits slightly outweigh the weaknesses. Hence, I recommend accept."""
35,"""Recurrent Kalman Networks: Factorized Inference in High-Dimensional Deep Feature Spaces""","['state estimation', 'recurrent neural networks', 'Kalman Filter', 'deep learning']","""In order to integrate uncertainty estimates into deep time-series modelling, Kalman Filters (KFs) (Kalman et al., 1960) have been integrated with deep learning models. Yet, such approaches typically rely on approximate inference techniques such as variational inference which makes learning more complex and often less scalable due to approximation errors. We propose a new deep approach to Kalman filtering which can be learned directly in an end-to-end manner using backpropagation without additional approximations. Our approach uses a high-dimensional factorized latent state representation for which the Kalman updates simplify to scalar operations and thus avoids hard to backpropagate, computationally heavy and potentially unstable matrix inversions. Moreover, we use locally linear dynamic models to efficiently propagate the latent state to the next time step. While our locally linear modelling and factorization assumptions are in general not true for the original low-dimensional state space of the system, the network finds a high-dimensional latent space where these assumptions hold to perform efficient inference. This state representation is learned jointly with the transition and noise models. The resulting network architecture, which we call Recurrent Kalman Network (RKN), can be used for any time-series data, similar to a LSTM (Hochreiter and Schmidhuber, 1997) but uses an explicit representation of uncertainty. As shown by our experiments, the RKN obtains much more accurate uncertainty estimates than an LSTM or Gated Recurrent Units (GRUs) (Cho et al., 2014) while also showing a slightly improved prediction performance and outperforms various recent generative models on an image imputation task.""","""A lot of work has appeared recently on recurrent state space models. So although this paper is in general considered favorable by the reviewers it is unclear exactly how the paper places itself in that (crowded) space. So rejection with a strong encouragement to update and resubmission is encouraged. """
36,"""Spatial-Winograd Pruning Enabling Sparse Winograd Convolution""","['deep learning', 'convolutional neural network', 'pruning', 'Winograd convolution']","""Deep convolutional neural networks (CNNs) are deployed in various applications but demand immense computational requirements. Pruning techniques and Winograd convolution are two typical methods to reduce the CNN computation. However, they cannot be directly combined because Winograd transformation fills in the sparsity resulting from pruning. Li et al. (2017) propose sparse Winograd convolution in which weights are directly pruned in the Winograd domain, but this technique is not very practical because Winograd-domain retraining requires low learning rates and hence significantly longer training time. Besides, Liu et al. (2018) move the ReLU function into the Winograd domain, which can help increase the weight sparsity but requires changes in the network structure. To achieve a high Winograd-domain weight sparsity without changing network structures, we propose a new pruning method, spatial-Winograd pruning. As the first step, spatial-domain weights are pruned in a structured way, which efficiently transfers the spatial-domain sparsity into the Winograd domain and avoids Winograd-domain retraining. For the next step, we also perform pruning and retraining directly in the Winograd domain but propose to use an importance factor matrix to adjust weight importance and weight gradients. This adjustment makes it possible to effectively retrain the pruned Winograd-domain network without changing the network structure. For the three models on the datasets of CIFAR-10, CIFAR-100, and ImageNet, our proposed method can achieve the Winograd-domain sparsities of 63%, 50%, and 74%, respectively.""","""Reviewer scores straddle the decision boundary but overall this does work does not meet the bar yet. Even after discussion with the authors, the reviewers reconfirmed there 'reject' recommendation and the area chair agrees with that assessment."""
37,"""Mode Normalization""","['Deep Learning', 'Expert Models', 'Normalization', 'Computer Vision']","""Normalization methods are a central building block in the deep learning toolbox. They accelerate and stabilize training, while decreasing the dependence on manually tuned learning rate schedules. When learning from multi-modal distributions, the effectiveness of batch normalization (BN), arguably the most prominent normalization method, is reduced. As a remedy, we propose a more flexible approach: by extending the normalization to more than a single mean and variance, we detect modes of data on-the-fly, jointly normalizing samples that share common features. We demonstrate that our method outperforms BN and other widely used normalization techniques in several experiments, including single and multi-task datasets.""","""The paper develops an original extension/generalization of standard batchnorm (and group norm) by employing a mixture-of-experts to separate incoming data into several modes and separately normalizing each mode. The paper is well written and technically correct, and the method yields consistent accuracy improvements over basic batchnorm on standard image classification tasks and models. Reviewers and AC noted the following potential weaknesses: a) while large on artificially mixed data, improvements are relatively small on single standard datasets (<1% on CIFAR10 and CIFAR100) b) the paper could better motivate why multi-modality is important e.g. by showing histograms of node activations c) the important interplay between number of modes and batch size should be more thoroughly discussed d) the closely related approach of Kalayeh & Shah 2018 should be presented and contrasted with in more details in the paper. Also comparing to it in experiments would enrich the work. """
38,"""Zero-shot Learning for Speech Recognition with Universal Phonetic Model""","['zero-shot learning', 'speech recognition', 'acoustic modeling']","""There are more than 7,000 languages in the world, but due to the lack of training sets, only a small number of them have speech recognition systems. Multilingual speech recognition provides a solution if at least some audio training data is available. Often, however, phoneme inventories differ between the training languages and the target language, making this approach infeasible. In this work, we address the problem of building an acoustic model for languages with zero audio resources. Our model is able to recognize unseen phonemes in the target language, if only a small text corpus is available. We adopt the idea of zero-shot learning, and decompose phonemes into corresponding phonetic attributes such as vowel and consonant. Instead of predicting phonemes directly, we first predict distributions over phonetic attributes, and then compute phoneme distributions with a customized acoustic model. We extensively evaluate our English-trained model on 20 unseen languages, and find that on average, it achieves 9.9% better phone error rate over a traditional CTC based acoustic model trained on English.""","""This paper studies the really hard problem of zero-shot learning in acoustic modeling for languages with limited resources, using data from English. Using a novel universal phonetic model, the authors show improvements compared to using an English model for 20 other languages in phone recognition quality. Strengths - Reviewers agree that the problem is an important one, and the presented ideas are novel. - Universal phonetic model to represent phones in any language is interesting. Weaknesses - The results are really weak, to the point that it is unclear how effective or general the techniques are. The work is an interesting first step, but is not developed enough to be accepted at this point. - The universal phonetic model being trained only in English might affect generalizability to languages that do not share phonetic characteristics. The authors agree partly, and argue that the method already addresses some issues since the model can already represent unseen phones. But, coupled with the high phone error rates, it is still unclear how appropriate the technique will be in addressing this issue. - Novelty: Although the idea of mapping phones to attributes, and using those for ASR is not novel (e.g., using articulatory features), application for zero-shot learning is. The work assumes availability of a small text corpus to learn phone-sequence distribution, so is similar to other zero-resource approaches that assume some data (audio, as opposed to text) is available in the new language. This paper presents interesting first steps, but lacks sufficient experimental validation at this point. Therefore, AE recommendation is to reject the paper. I encourage the authors to improve and resubmit in the future."""
39,"""Deep Probabilistic Video Compression""","['variational inference', 'video compression', 'deep generative models']","""We propose a variational inference approach to deep probabilistic video compression. Our model uses advances in variational autoencoders (VAEs) for sequential data and combines it with recent work on neural image compression. The approach jointly learns to transform the original video into a lower-dimensional representation as well as to entropy code this representation according to a temporally-conditioned probabilistic model. We split the latent space into local (per frame) and global (per segment) variables, and show that training the VAE to utilize both representations leads to an improved rate-distortion performance. Evaluation on small videos from public data sets with varying complexity and diversity show that our model yields competitive results when trained on generic video content. Extreme compression performance is achieved for videos with specialized content if the model is trained on similar videos.""","""The proposed method is compressing video sequences with an end-to-end approach, by extending a variational approach from images to videos. The problem setting is interesting and somewhat novel. The main limitation, as exposed by the reviewers, is that evaluation was done on very limited and small domains. It is not at all clear that this method scales well to non-toy domains or that it is possible in fact to get good results with an extension of this method beyond small-scale content. There were some concerns about unfair comparisons to classical codes that were optimized for longer sequences (and I share those concerns, though they are somewhat alleviated in the rebuttal). While the paper presents an interesting line of work, the reviewers did present a number of issues that make it hard to recommend it for acceptance. However, as R1 points out, most of the problems are fixable and I would advise the authors to take the suggested improvements (especially anything related to modeling longer sequences) and once they are incorporated this will be a much stronger submission."""
40,"""Talk The Walk: Navigating Grids in New York City through Grounded Dialogue""","['Dialogue', 'Navigation', 'Grounded Language Learning']","""We introduce `""Talk The Walk"", the first large-scale dialogue dataset grounded in action and perception. The task involves two agents (a 'guide' and a 'tourist') that communicate via natural language in order to achieve a common goal: having the tourist navigate to a given target location. The task and dataset, which are described in detail, are challenging and their full solution is an open problem that we pose to the community. We (i) focus on the task of tourist localization and develop the novel Masked Attention for Spatial Convolutions (MASC) mechanism that allows for grounding tourist utterances into the guide's map, (ii) show it yields significant improvements for both emergent and natural language communication, and (iii) using this method, we establish non-trivial baselines on the full task. ""","""This paper introduces a newly collected dataset of natural language interactions between a tourist and a guide for localization and navigation. The paper also includes baseline experiments with a reasonably novel approach. The task is well motivated (although an open question remains due to GPS, comment by reviewer 1), but the description of the dataset and collection, approach and experiments were not ideal in the first version of the paper. Much of the information was pushed to the appendix and it was hard to follow the paper without going back and forth, and even then some points were missing. Authors rewrote parts of the paper to address these concerns, but there are still some open questions. For example, is it possible to have sub-tasks, given the task is complex and may not be easy to accomplish as a whole? Or could simple LSTM be another baseline (the final review of the third reviewer)? """
41,"""Pooling Is Neither Necessary nor Sufficient for Appropriate Deformation Stability in CNNs""","['Convolutional Neural Networks', 'Deformation Stability', 'Pooling', 'Transformation Invariance']","""Many of our core assumptions about how neural networks operate remain empirically untested. One common assumption is that convolutional neural networks need to be stable to small translations and deformations to solve image recognition tasks. For many years, this stability was baked into CNN architectures by incorporating interleaved pooling layers. Recently, however, interleaved pooling has largely been abandoned. This raises a number of questions: Are our intuitions about deformation stability right at all? Is it important? Is pooling necessary for deformation invariance? If not, how is deformation invariance achieved in its absence? In this work, we rigorously test these questions, and find that deformation stability in convolutional networks is more nuanced than it first appears: (1) Deformation invariance is not a binary property, but rather that different tasks require different degrees of deformation stability at different layers. (2) Deformation stability is not a fixed property of a network and is heavily adjusted over the course of training, largely through the smoothness of the convolutional filters. (3) Interleaved pooling layers are neither necessary nor sufficient for achieving the optimal form of deformation stability for natural image classification. (4) Pooling confers \emph{too much} deformation stability for image classification at initialization, and during training, networks have to learn to \emph{counteract} this inductive bias. Together, these findings provide new insights into the role of interleaved pooling and deformation invariance in CNNs, and demonstrate the importance of rigorous empirical testing of even our most basic assumptions about the working of neural networks.""","""This paper studies the role of pooling in the success underpinning CNNs. Through several experiments, the authors conclude that pooling is neither necessary nor sufficient to achieve deformation stability, and that its inductive bias can be mostly recovered after training. All reviewers agreed that this is a paper asking an important question, and that it is well-written and reproducible. On the other hand, they also agreed that, in its current form, this paper lacks a 'punchline' that can drive further research. In words of R6, ""the paper does not discuss the links between pooling and aliasing"", or in words of R4, ""it seems to very readily jump to unwarranted conclusions"". In summary, the AC recommends rejection at this time, and encourages the authors to pursue the line of attack by exploring the suggestions of the reviewers and resubmit. """
42,"""Efficiently testing local optimality and escaping saddles for ReLU networks""","['local optimality', 'second-order stationary point', 'escaping saddle points', 'nondifferentiability', 'ReLU', 'empirical risk']","""We provide a theoretical algorithm for checking local optimality and escaping saddles at nondifferentiable points of empirical risks of two-layer ReLU networks. Our algorithm receives any parameter value and returns: local minimum, second-order stationary point, or a strict descent direction. The presence of M data points on the nondifferentiability of the ReLU divides the parameter space into at most 2^M regions, which makes analysis difficult. By exploiting polyhedral geometry, we reduce the total computation down to one convex quadratic program (QP) for each hidden node, O(M) (in)equality tests, and one (or a few) nonconvex QP. For the last QP, we show that our specific problem can be solved efficiently, in spite of nonconvexity. In the benign case, we solve one equality constrained QP, and we prove that projected gradient descent solves it exponentially fast. In the bad case, we have to solve a few more inequality constrained QPs, but we prove that the time complexity is exponential only in the number of inequality constraints. Our experiments show that either benign case or bad case with very few inequality constraints occurs, implying that our algorithm is efficient in most cases.""","""This paper proposes a new method for verifying whether a given point of a two layer ReLU network is a local minima or a second order stationary point and checks for descent directions. All reviewers agree that the algorithm is based on number of new techniques involving both convex and non-convex QPs, and is novel. The method proposed in the paper has significant limitations as the method is not robust to handle approximate stationary points. Given these limitations, there is a disagreement between reviewers about the significance of the result . While I share the same concerns as R4, I agree with R3 and believe that the new ideas in the paper will inspire future work to extend the proposed method towards addressing these limitations. Hence I suggest acceptance. """
43,"""The Limitations of Adversarial Training and the Blind-Spot Attack""","['Adversarial Examples', 'Adversarial Training', 'Blind-Spot Attack']","""The adversarial training procedure proposed by Madry et al. (2018) is one of the most effective methods to defend against adversarial examples in deep neural net- works (DNNs). In our paper, we shed some lights on the practicality and the hardness of adversarial training by showing that the effectiveness (robustness on test set) of adversarial training has a strong correlation with the distance between a test point and the manifold of training data embedded by the network. Test examples that are relatively far away from this manifold are more likely to be vulnerable to adversarial attacks. Consequentially, an adversarial training based defense is susceptible to a new class of attacks, the blind-spot attack, where the input images reside in blind-spots (low density regions) of the empirical distri- bution of training data but is still on the ground-truth data manifold. For MNIST, we found that these blind-spots can be easily found by simply scaling and shifting image pixel values. Most importantly, for large datasets with high dimensional and complex data manifold (CIFAR, ImageNet, etc), the existence of blind-spots in adversarial training makes defending on any valid test examples difficult due to the curse of dimensionality and the scarcity of training data. Additionally, we find that blind-spots also exist on provable defenses including (Kolter & Wong, 2018) and (Sinha et al., 2018) because these trainable robustness certificates can only be practically optimized on a limited set of training data.""","""Reviewers are in a consensus and recommended to accept after engaging with the authors. Please take reviewers' comments into consideration to improve your submission for the camera ready. """
44,"""Using Deep Siamese Neural Networks to Speed up Natural Products Research""","['clustering', 'deep learning', 'application', 'chemistry', 'natural products']","""Natural products (NPs, compounds derived from plants and animals) are an important source of novel disease treatments. A bottleneck in the search for new NPs is structure determination. One method is to use 2D Nuclear Magnetic Resonance (NMR) imaging, which indicates bonds between nuclei in the compound, and hence is the ""fingerprint"" of the compound. Computing a similarity score between 2D NMR spectra for a novel compound and a compound whose structure is known helps determine the structure of the novel compound. Standard approaches to this problem do not appear to scale to larger databases of compounds. Here we use deep convolutional Siamese networks to map NMR spectra to a cluster space, where similarity is given by the distance in the space. This approach results in an AUC score that is more than four times better than an approach using Latent Dirichlet Allocation.""","""The reviewers highlighted that the application in the paper is interesting, but note a lack of new methodology, and also highlight serious flaws in the testing methodology. Specifically, the reviewers are discouraged by the straightforward reuse of Siamese networks without clear modifications. Further, the testing setup might be unfairly easy, since chemical families are represented in both training and test sets, while in true application of the method would be exposed to previously unseen chemical families. The authors did not participate in the discussion, and address concerns. The reviewer consensus is a rejection."""
45,"""SEQUENCE MODELLING WITH AUTO-ADDRESSING AND RECURRENT MEMORY INTEGRATING NETWORKS""","['Memory Network', 'RNN', 'Sequence Modelling']","""Processing sequential data with long term dependencies and learn complex transitions are two major challenges in many deep learning applications. In this paper, we introduce a novel architecture, the Auto-addressing and Recurrent Memory Integrating Network (ARMIN) to address these issues. The ARMIN explicitly stores previous hidden states and recurrently integrate useful past states into current time-step by an efficient memory addressing mechanism. Compared to existing memory networks, the ARMIN is more light-weight and inference-time efficient. Our network can be trained on small slices of long sequential data, and thus, can boost its training speed. Experiments on various tasks demonstrate the efficiency of the ARMIN architecture. Codes and models will be available.""","""there have been many variants of memory augmented neural nets since around 2014 when NTM, attention-based NMT and MemNet were proposed. it is indeed still an interesting and important direction of research, but the bar for introducing yet another variant of memory-augmented neural nets has been significantly raised, which is a sentiment shared by the reviewers. the author's response had not swayed the reviewers' opinion, and i am sticking to the reviewers' decisions. i believe more streamlined and systematic comparison among different memory augmented networks across many different benchmarks (e.g., use the same set of latest variants of memory nets across all the benchmarks) in this submission would make it a better paper and increase the chance of acceptance. """
46,"""Adaptive Sample-space & Adaptive Probability coding: a neural-network based approach for compression""","['Data compression', 'Image compression', 'Deep Learning', 'Convolutional neural networks']","""We propose Adaptive Sample-space & Adaptive Probability (ASAP) coding, an efficient neural-network based method for lossy data compression. Our ASAP coding distinguishes itself from the conventional method based on adaptive arithmetic coding in that it models the probability distribution for the quantization process in such a way that one can conduct back-propagation for the quantization width that determines the support of the distribution. Our ASAP also trains the model with a novel, hyper-parameter free multiplicative loss for the rate-distortion tradeoff. With our ASAP encoder, we are able to compress the image files in the Kodak dataset to as low as one fifth the size of the JPEG-compressed image without compromising their visual quality, and achieved the state-of-the-art result in terms of MS-SSIM based rate-distortion tradeoff. ""","""This paper presents an interesting approach to image compression, as recognized by all reviewers. However, important concerns about evaluating the contribution remains: as noted by reviewers, evaluating the contribution requires disentangling what part of the improvement is due to the proposed approach and what part is due to the loss chosen and evaluation methods. While authors have done a valuable effort adding experiments to incorporate reviewers suggestions with ablation studies, it does not convincingly show that the proposed approach truly improves over existing ones like Balle et al. Authors are encouraged to strengthen their work for future submission by putting particular emphasis on those questions."""
47,"""CNNSAT: Fast, Accurate Boolean Satisfiability using Convolutional Neural Networks""","['Convolutional Neural Networks', 'Boolean satisfiability problem', 'Satisfiability modulo theories']","""Boolean satisfiability (SAT) is one of the most well-known NP-complete problems and has been extensively studied. State-of-the-art solvers exist and have found a wide range of applications. However, they still do not scale well to formulas with hundreds of variables. To tackle this fundamental scalability challenge, we introduce CNNSAT, a fast and accurate statistical decision procedure for SAT based on convolutional neural networks. CNNSAT's effectiveness is due to a precise and compact representation of Boolean formulas. On both real and synthetic formulas, CNNSAT is highly accurate and orders of magnitude faster than the state-of-the-art solver Z3. We also describe how to extend CNNSAT to predict satisfying assignments when it predicts a formula to be satisfiable.""","""The authors provide a convolutional neural network for predicting the satisfiability of SAT instances. The idea is interesting, and the main novelty in the paper is the use of convolutions in the architecture and a procedure to predict a witness when the formula is satisfiable. However, there are concerns about the suitability of convolutions for this problem because of the permutation invariance of SAT. Empirically, the resulting models are accurate (correctly predicting sat/unsat 90-99% of the time) while taking less time than some existing solvers. However, as pointed out by the reviewers, the empirical results are not sufficient to demonstrate the effectiveness of the approach. I want to thank the authors for the great work they did to address the concerns of the reviewers. The paper significantly improved over the reviewing period, and while it is not yet ready for publication, I want to encourage the authors to keep pushing the idea to further and improve the experimental results. """
48,"""Learning Graph Representations by Dendrograms""","['Graph', 'hierarchical clustering', 'dendrogram', 'quality metric', 'reconstruction', 'entropy']","""Hierarchical clustering is a common approach to analysing the multi-scale structure of graphs observed in practice. We propose a novel metric for assessing the quality of a hierarchical clustering. This metric reflects the ability to reconstruct the graph from the dendrogram encoding the hierarchy. The best representation of the graph for this metric in turn yields a novel hierarchical clustering algorithm. Experiments on both real and synthetic data illustrate the efficiency of the approach. ""","""All reviewers agree to reject. While there were many positive points to this work, reviewers believed that it was not yet ready for acceptance."""
49,"""Learn From Neighbour: A Curriculum That Train Low Weighted Samples By Imitating""","['Curriculum Learning', 'Internal Covariate Shift']","""Deep neural networks, which gain great success in a wide spectrum of applications, are often time, compute and storage hungry. Curriculum learning proposed to boost training of network by a syllabus from easy to hard. However, the relationship between data complexity and network training is unclear: why hard example harm the performance at beginning but helps at end. In this paper, we aim to investigate on this problem. Similar to internal covariate shift in network forward pass, the distribution changes in weight of top layers also affects training of preceding layers during the backward pass. We call this phenomenon inverse ""internal covariate shift"". Training hard examples aggravates the distribution shifting and damages the training. To address this problem, we introduce a curriculum loss that consists of two parts: a) an adaptive weight that mitigates large early punishment; b) an additional representation loss for low weighted samples. The intuition of the loss is very simple. We train top layers on ""good"" samples to reduce large shifting, and encourage ""bad"" samples to learn from ""good"" sample. In detail, the adaptive weight assigns small values to hard examples, reducing the influence of noisy gradients. On the other hand, the less-weighted hard sample receives the proposed representation loss. Low-weighted data gets nearly no training signal and can stuck in embedding space for a long time. The proposed representation loss aims to encourage their training. This is done by letting them learn a better representation from its superior neighbours but not participate in learning of top layers. In this way, the fluctuation of top layers is reduced and hard samples also received signals for training. We found in this paper that curriculum learning needs random sampling between tasks for better training. Our curriculum loss is easy to combine with existing stochastic algorithms like SGD. Experimental result shows an consistent improvement over several benchmark datasets.""","""This paper attempts to address a problem they dub ""inverse"" covariate shift where an improperly trained output layer can hamper learning. The idea is to use a form of curriculum learning. The reviewers found that the notion of inverse covariate shift was not formally or empirically well defined. Furthermore the baselines used were too weak: the authors should consider comparing against state-of-the-art curriculum learning methods."""
50,"""Invariant and Equivariant Graph Networks""","['graph learning', 'equivariance', 'deep learning']","""Invariant and equivariant networks have been successfully used for learning images, sets, point clouds, and graphs. A basic challenge in developing such networks is finding the maximal collection of invariant and equivariant \emph{linear} layers. Although this question is answered for the first three examples (for popular transformations, at-least), a full characterization of invariant and equivariant linear layers for graphs is not known. In this paper we provide a characterization of all permutation invariant and equivariant linear layers for (hyper-)graph data, and show that their dimension, in case of edge-value graph data, is pseudo-formula and pseudo-formula , respectively. More generally, for graph data defined on pseudo-formula -tuples of nodes, the dimension is the pseudo-formula -th and pseudo-formula -th Bell numbers. Orthogonal bases for the layers are computed, including generalization to multi-graph data. The constant number of basis elements and their characteristics allow successfully applying the networks to different size graphs. From the theoretical point of view, our results generalize and unify recent advancement in equivariant deep learning. In particular, we show that our model is capable of approximating any message passing neural network. Applying these new linear layers in a simple deep neural network framework is shown to achieve comparable results to state-of-the-art and to have better expressivity than previous invariant and equivariant bases. ""","""The paper provides a comprehensive study and generalisations of previous results on linear permutation invariant and equivariant operators / layers for the case of hypergraph data on multiple node sets. Reviewers indicate that the paper makes a particularly interesting and important contribution, with applications to graphs and hyper-graphs, as demonstrated in experiments. A concern was raised that the paper could be overstating its scope. A point is that the model might not actually give a complete characterization, since the analysis considers permutation action only. The authors have rephrased the claim. Following comments of the reviewer, the authors have also revised the paper to include a discussion of how the model is capable of approximating message passing networks. Two referees give the paper a strong support. One referee considers the paper ok, but not good enough. The authors have made convincing efforts to improve issues and address the concerns. """
51,"""Accelerated Value Iteration via Anderson Mixing""",['Reinforcement Learning'],"""Acceleration for reinforcement learning methods is an important and challenging theme. We introduce the Anderson acceleration technique into the value iteration, developing an accelerated value iteration algorithm that we call Anderson Accelerated Value Iteration (A2VI). We further apply our method to the Deep Q-learning algorithm, resulting in the Deep Anderson Accelerated Q-learning (DA2Q) algorithm. Our approach can be viewed as an approximation of the policy evaluation by interpolating on historical data. A2VI is more efficient than the modified policy iteration, which is a classical approximate method for policy evaluation. We give a theoretical analysis of our algorithm and conduct experiments on both toy problems and Atari games. Both the theoretical and empirical results show the effectiveness of our algorithm.""","""The paper proposes to use Anderson Mixing to accelerate value iteration and DQN. The idea is interesting, with some theoretical and empirical support. However, reviewers feel that the contribution is somewhat limited, and certain parts (e.g., the DP view) can be further developed to strengthen the technical contribution. Furthermore, one reviewer points out that the empirical results are not very strong, where the improvements on 3 Atari games are not very substantial. Overall, while the paper is interesting and does have the potential, it seems too preliminary to be published in its current form. Minor comments: 1. The paper is partially motivated by the claim given at the beginning of section 3: ""Based on the observation that full policy evaluation accelerates convergence, ..."" Can a reference be given? 2. Another way to look at Anderson Mixing is the standard linear value function approximation framework, where the previous K value functions serve as basis functions. See Mahadevan & Maggioni (JMLR'07), Parr et al. (ICML'08) and Konidaris et al. (AAAI'11) for a few examples of constructing basis functions; the approach here seems to provide another way to automatically construct basic functions. A discussion would be helpful."""
52,"""MEAN-FIELD ANALYSIS OF BATCH NORMALIZATION""","['neural networks', 'optimization', 'batch normalization', 'mean field theory', 'Fisher information']","""Batch Normalization (BatchNorm) is an extremely useful component of modern neural network architectures, enabling optimization using higher learning rates and achieving faster convergence. In this paper, we use mean-field theory to analytically quantify the impact of BatchNorm on the geometry of the loss landscape for multi-layer networks consisting of fully-connected and convolutional layers. We show that it has a flattening effect on the loss landscape, as quantified by the maximum eigenvalue of the Fisher Information Matrix. These findings are then used to justify the use of larger learning rates for networks that use BatchNorm, and we provide quantitative characterization of the maximal allowable learning rate to ensure convergence. Experiments support our theoretically predicted maximum learning rate, and furthermore suggest that networks with smaller values of the BatchNorm parameter achieve lower loss after the same number of epochs of training.""","""This paper presents a mean field analysis of the effect of batch norm on optimization. Assuming the weights and biases are independent Gaussians (an assumption that's led to other interesting analysis), they propagate various statistics through the network, which lets them derive the maximum eigenvalue of the Fisher information matrix. This determines the maximum learning rate at which learning is stable. The finding is that batch norm allows larger learning rates. In terms of novelty, the paper builds on the analysis of Karakida et al. (2018). The derivations are mostly mechanical, though there's probably still sufficient novelty. Unfortunately, it's not clear what we learn at the end of the day. The maximum learning rate isn't very meaningful to analyze, since the learning rate is only meaningful relative to the scale of the weights and gradients, and the distance that needs to be moved to reach the optimum. The authors claim that a ""higher learning rate leads to faster convergence"", but this seems false, and at the very least would need more justification. It's well-known that batch norm rescales the norm of the gradients inversely to the norm of the weights; hence, if the weight norm is larger than 1, BN will reduce the gradient norm and hence increase the maximum learning rate. But this isn't a very interesting effect from an optimization perspective. I can't tell from the analysis whether there's a more meaningful sense in which BN speeds up convergence. The condition number might be more relevant from a convergence perspective. Overall, this paper is a promising start, but needs more work before it's ready for publication at ICLR. """
53,"""A preconditioned accelerated stochastic gradient descent algorithm""","['stochastic optimization', 'neural network', 'preconditioned accelerated stochastic gradient descent']","""We propose a preconditioned accelerated stochastic gradient method suitable for large scale optimization. We derive sufficient convergence conditions for the minimization of convex functions using a generic class of diagonal preconditioners and provide a formal convergence proof based on a framework originally used for on-line learning. Inspired by recent popular adaptive per-feature algorithms, we propose a specific preconditioner based on the second moment of the gradient. The sufficient convergence conditions motivate a critical adaptation of the per-feature updates in order to ensure convergence. We show empirical results for the minimization of convex and non-convex cost functions, in the context of neural network training. The method compares favorably with respect to current, first order, stochastic optimization methods.""","""Dear authors, The reviewers pointed out a number of concerns about this work. It is thus not ready for publication. Should you decide to resubmit it to another venue, please address these concerns."""
54,"""Convolutional Neural Networks on Non-uniform Geometrical Signals Using Euclidean Spectral Transformation""","['Non-uniform Fourier Transform', '3D Learning', 'CNN', 'surface reconstruction']","""Convolutional Neural Networks (CNN) have been successful in processing data signals that are uniformly sampled in the spatial domain (e.g., images). However, most data signals do not natively exist on a grid, and in the process of being sampled onto a uniform physical grid suffer significant aliasing error and information loss. Moreover, signals can exist in different topological structures as, for example, points, lines, surfaces and volumes. It has been challenging to analyze signals with mixed topologies (for example, point cloud with surface mesh). To this end, we develop mathematical formulations for Non-Uniform Fourier Transforms (NUFT) to directly, and optimally, sample nonuniform data signals of different topologies defined on a simplex mesh into the spectral domain with no spatial sampling error. The spectral transform is performed in the Euclidean space, which removes the translation ambiguity from works on the graph spectrum. Our representation has four distinct advantages: (1) the process causes no spatial sampling error during initial sampling, (2) the generality of this approach provides a unified framework for using CNNs to analyze signals of mixed topologies, (3) it allows us to leverage state-of-the-art backbone CNN architectures for effective learning without having to design a particular architecture for a particular data structure in an ad-hoc fashion, and (4) the representation allows weighted meshes where each element has a different weight (i.e., texture) indicating local properties. We achieve good results on-par with state-of-the-art for 3D shape retrieval task, and new state-of-the-art for point cloud to surface reconstruction task.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The paper tackles an interesting and challenging problem with a novel approach. - The method gives improves improved performance for the surface reconstruction task. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. The paper - lacks clarity in some areas - doesn't sufficiently explain the trade-offs between performing all computations in the spectral domain vs the spatial domain. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. Reviewers had a divergent set of concerns. After the rebuttal, the remaining concerns were: - the significance of the performance improvements. The AC believes that the quantitative and qualitative results in Table 3 and Figures 5 and 6 show significant improvements with respect to two recent methods. - a feeling that the proposed method could have been more efficient if more computations were done in the spectral domain. This is a fair point but should be considered as suggestions for improvement and future work rather than grounds for rejection in the AC's view. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers did not reach a consensus. The final decision is aligned with the more positive reviewer, AR1, because AR1 was more confident in his/her review and because of the additional reasons stated in the previous section. """
55,"""Learning deep representations by mutual information estimation and maximization""","['representation learning', 'unsupervised learning', 'deep learning']","""This work investigates unsupervised learning of representations by maximizing mutual information between an input and the output of a deep neural network encoder. Importantly, we show that structure matters: incorporating knowledge about locality in the input into the objective can significantly improve a representation's suitability for downstream tasks. We further control characteristics of the representation by matching to a prior distribution adversarially. Our method, which we call Deep InfoMax (DIM), outperforms a number of popular unsupervised learning methods and compares favorably with fully-supervised learning on several classification tasks in with some standard architectures. DIM opens new avenues for unsupervised learning of representations and is an important step towards flexible formulations of representation learning objectives for specific end-goals.""","""This paper proposes a new unsupervised learning approach based on maximizing the mutual information between the input and the representation. The results are strong across several image datasets. Essentially all of the reviewer's concerns were directly addressed in revisions of the paper, including additional experiments. The only weakness is that only image datasets were experimented with; however, the image-based experiments and comparisons are extensive. The reviewers and I all agree that the paper should be accepted, and I think it should be considered for an oral presentation."""
56,"""Sparse Dictionary Learning by Dynamical Neural Networks""","['dynamical neural networks', 'spiking neural networks', 'dynamical system', 'hardware friendly learning', 'feedback', 'contrastive learning', 'dictionary learning', 'sparse coding']","""A dynamical neural network consists of a set of interconnected neurons that interact over time continuously. It can exhibit computational properties in the sense that the dynamical systems evolution and/or limit points in the associated state space can correspond to numerical solutions to certain mathematical optimization or learning problems. Such a computational system is particularly attractive in that it can be mapped to a massively parallel computer architecture for power and throughput efficiency, especially if each neuron can rely solely on local information (i.e., local memory). Deriving gradients from the dynamical networks various states while conforming to this last constraint, however, is challenging. We show that by combining ideas of top-down feedback and contrastive learning, a dynamical network for solving the l1-minimizing dictionary learning problem can be constructed, and the true gradients for learning are provably computable by individual neurons. Using spiking neurons to construct our dynamical network, we present a learning process, its rigorous mathematical analysis, and numerical results on several dictionary learning problems.""","""While there has been lots of previous work on training dictionaries for sparse coding, this work tackles the problem of doing son in a purely local way. While previous work suggests that the exact computation of gradient addressed in the paper is not necessarily critical, as noted by reviewers, all reviewers agree that the work still makes important contributions through both its theoretical analyses and presented experiments. Authors are encouraged to work on improving clarity further and delineating their contribution more precisely with respect to previous results."""
57,"""LEARNING ADVERSARIAL EXAMPLES WITH RIEMANNIAN GEOMETRY""","['Adversarial training', 'Adversarial examples', 'Riemannian Geometry', 'Machine Learning', 'Deep Learning']","""Adversarial examples, referred to as augmented data points generated by imperceptible perturbation of input samples, have recently drawn much attention. Well-crafted adversarial examples may even mislead state-of-the-art deep models to make wrong predictions easily. To alleviate this problem, many studies focus on investigating how adversarial examples can be generated and/or resisted. All the existing work handles this problem in the Euclidean space, which may however be unable to describe data geometry. In this paper, we propose a generalized framework that addresses the learning problem of adversarial examples with Riemannian geometry. Specifically, we define the local coordinate systems on Riemannian manifold, develop a novel model called Adversarial Training with Riemannian Manifold, and design a series of theory that manages to learn the adversarial examples in the Riemannian space feasibly and efficiently. The proposed work is important in that (1) it is a generalized learning methodology since Riemmanian manifold space would be degraded to the Euclidean space in a special case; (2) it is the first work to tackle the adversarial example problem tractably through the perspective of geometry; (3) from the perspective of geometry, our method leads to the steepest direction of the loss function. We also provide a series of theory showing that our proposed method can truly find the decent direction for the loss function with a comparable computational time against traditional adversarial methods. Finally, the proposed framework demonstrates superior performance to the traditional counterpart methods on benchmark data including MNIST, CIFAR-10 and SVHN.""","""On the positive side, this is among the first papers to exploit non-Euclidean geometry, specifically curvature for adversarial learning. However, reviewers are largely in agreement that the technical correctness of this paper is unconvincing despite substantial technical exchanges with the authors. """
58,"""Learning Embeddings into Entropic Wasserstein Spaces""","['Embedding', 'Wasserstein', 'Sinkhorn', 'Optimal Transport']","""Despite their prevalence, Euclidean embeddings of data are fundamentally limited in their ability to capture latent semantic structures, which need not conform to Euclidean spatial assumptions. Here we consider an alternative, which embeds data as discrete probability distributions in a Wasserstein space, endowed with an optimal transport metric. Wasserstein spaces are much larger and more flexible than Euclidean spaces, in that they can successfully embed a wider variety of metric structures. We propose to exploit this flexibility by learning an embedding that captures the semantic information in the Wasserstein distance between embedded distributions. We examine empirically the representational capacity of such learned Wasserstein embeddings, showing that they can embed a wide variety of complex metric structures with smaller distortion than an equivalent Euclidean embedding. We also investigate an application to word embedding, demonstrating a unique advantage of Wasserstein embeddings: we can directly visualize the high-dimensional embedding, as it is a probability distribution on a low-dimensional space. This obviates the need for dimensionality reduction techniques such as t-SNE for visualization.""",""" + An interesting and original idea of embedding words into the (very low dimensional) Wasserstein space, i.e. clouds of points in a low-dimensional space + As the space is low-dimensional (2D), it can be directly visualized. + I could imagine the technique to be useful in social / human science for data visualization, the visualization is more faithful to what the model is doing than t-SNE plots of high-dimensional embeddings + Though not the first method to embed words as densities but seemingly the first one which shows that multi-modality / multiple senses are captured (except for models which capture discrete senses) + The paper is very well written - The results are not very convincing but show that embeddings do capture word similarity (even when training the model on a small dataset) - The approach is not very scalable (hence evaluation on 17M corpus) - The method cannot be used to deal with data sparsity, though (very) interesting for visualization - This is mostly an empirical paper (i.e. an interesting application of an existing method) The reviewers are split. One reviewer is negative as they are unclear what the technical contribution is (but seems a bit biased against empirical papers). Another two find the paper very interesting. """
59,"""Set Transformer""","['attention', 'meta-learning', 'set-input neural networks', 'permutation invariant modeling']","""Many machine learning tasks such as multiple instance learning, 3D shape recognition and few-shot image classification are defined on sets of instances. Since solutions to such problems do not depend on the permutation of elements of the set, models used to address them should be permutation invariant. We present an attention-based neural network module, the Set Transformer, specifically designed to model interactions among elements in the input set. The model consists of an encoder and a decoder, both of which rely on attention mechanisms. In an effort to reduce computational complexity, we introduce an attention scheme inspired by inducing point methods from sparse Gaussian process literature. It reduces computation time of self-attention from quadratic to linear in the number of elements in the set. We show that our model is theoretically attractive and we evaluate it on a range of tasks, demonstrating increased performance compared to recent methods for set-structured data.""","""This paper introduces set transformer for set inputs. The idea is built upon the transformer and introduces the attention mechanism. Major concerns on novelty were raised by the reviewers. """
60,"""Local Binary Pattern Networks for Character Recognition""","['deep learning', 'local binary pattern', 'supervised learning', 'hardware-friendly']","""Memory and computation efficient deep learning architectures are crucial to the continued proliferation of machine learning capabilities to new platforms and systems. Binarization of operations in convolutional neural networks has shown promising results in reducing the model size and computing efficiency. In this paper, we tackle the character recognition problem using a strategy different from the existing literature by proposing local binary pattern networks or LBPNet that can learn and perform bit-wise operations in an end-to-end fashion. LBPNet uses local binary comparisons and random projection in place of conventional convolution (or approximation of convolution) operations, providing important means to improve memory and speed efficiency that is particularly suited for small footprint devices and hardware accelerators. These operations can be implemented efficiently on different platforms including direct hardware implementation. LBPNet demonstrates its particular advantage on the character classification task where the content is composed of strokes. We applied LBPNet to benchmark datasets like MNIST, SVHN, DHCD, ICDAR, and Chars74K and observed encouraging results.""","""This paper proposed a LBPNet for character recognition, which introduces the LBP feature extraction into deep learning. Reviewers are confused on implementation and not convinced on experiments. The only score 6 reviewer is also concerned ""Empirically weak, practical advantage wrt to literature unclear"". Only evaluating on MNIST/SVHN etc is not convincing to demo the effectiveness of the proposed method."""
61,"""Graph2Seq: Scalable Learning Dynamics for Graphs""","['graph neural networks', 'scalable representations', 'combinatorial optimization', 'reinforcement learning']","""Neural networks have been shown to be an effective tool for learning algorithms over graph-structured data. However, graph representation techniques---that convert graphs to real-valued vectors for use with neural networks---are still in their infancy. Recent works have proposed several approaches (e.g., graph convolutional networks), but these methods have difficulty scaling and generalizing to graphs with different sizes and shapes. We present Graph2Seq, a new technique that represents vertices of graphs as infinite time-series. By not limiting the representation to a fixed dimension, Graph2Seq scales naturally to graphs of arbitrary sizes and shapes. Graph2Seq is also reversible, allowing full recovery of the graph structure from the sequence. By analyzing a formal computational model for graph representation, we show that an unbounded sequence is necessary for scalability. Our experimental results with Graph2Seq show strong generalization and new state-of-the-art performance on a variety of graph combinatorial optimization problems. ""","""This was an extremely difficult case. There are many positive aspects of Graph2Seq, as detailed by all of the reviewers, however two of the reviewers have issue with the current theory, specifically the definition of k-local-gather and its relation to existing models. The authors and reviewers have had a detailed and discussion on the issue, however we do not seem to have come to a resolution. I will not wade into the specifics of the argument, however, ultimately, the onus is on the authors to convince the reviewers of the merits/correctness, and in this case two reviewers had the same issue, and their concerns have not been resolved. The best advice I can give is to consider the discussion so far and why this misunderstanding occurred, so that it might lead the best version of this paper possible."""
62,"""Using GANs for Generation of Realistic City-Scale Ride Sharing/Hailing Data Sets""","['ride-sharing', 'generative modeling', 'parallelization', 'application']","""This paper focuses on the synthetic generation of human mobility data in urban areas. We present a novel and scalable application of Generative Adversarial Networks (GANs) for modeling and generating human mobility data. We leverage actual ride requests from ride sharing/hailing services from four major cities in the US to train our GANs model. Our model captures the spatial and temporal variability of the ride-request patterns observed for all four cities on any typical day and over any typical week. Previous works have succinctly characterized the spatial and temporal properties of human mobility data sets using the fractal dimensionality and the densification power law, respectively, which we utilize to validate our GANs-generated synthetic data sets. Such synthetic data sets can avoid privacy concerns and be extremely useful for researchers and policy makers on urban mobility and intelligent transportation.""","""While the reviewers all agree that this paper proposes an interesting application of GANs, they would like to see clearer explanations of the technical details, more convincing evaluations, and better justifications of the assumptions and practical values of the proposed algorithms. """
63,"""Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives""","['variational autoencoder', 'reparameterization trick', 'IWAE', 'VAE', 'RWS', 'JVI']","""Deep latent variable models have become a popular model choice due to the scalable learning algorithms introduced by (Kingma & Welling 2013, Rezende et al. 2014). These approaches maximize a variational lower bound on the intractable log likelihood of the observed data. Burda et al. (2015) introduced a multi-sample variational bound, IWAE, that is at least as tight as the standard variational lower bound and becomes increasingly tight as the number of samples increases. Counterintuitively, the typical inference network gradient estimator for the IWAE bound performs poorly as the number of samples increases (Rainforth et al. 2018, Le et al. 2018). Roeder et a. (2017) propose an improved gradient estimator, however, are unable to show it is unbiased. We show that it is in fact biased and that the bias can be estimated efficiently with a second application of the reparameterization trick. The doubly reparameterized gradient (DReG) estimator does not suffer as the number of samples increases, resolving the previously raised issues. The same idea can be used to improve many recently introduced training techniques for latent variable models. In particular, we show that this estimator reduces the variance of the IWAE gradient, the reweighted wake-sleep update (RWS) (Bornschein & Bengio 2014), and the jackknife variational inference (JVI) gradient (Nowozin 2018). Finally, we show that this computationally efficient, drop-in estimator translates to improved performance for all three objectives on several modeling tasks.""","""The paper is well written and easy to follow. The experiments are adequate to justify the usefulness of an identity for improving existing multi-Monte-Carlo-sample based gradient estimators for deep generative models. The originality and significance are acceptable, as discussed below. The proposed doubly reparameterized gradient estimators are built on an important identity shown in Equation (5). This identity appears straightforward to derive by applying both score-function gradient and reparameterization gradient to the same objective function, which is expressed as an expectation. The AC suspects that this identity might have already appeared in previous publications / implementations, though not being claimed as an important contribution / being explicitly discussed. While that identity may not be claimed as the original contribution of the paper if that suspicion is true, the paper makes another useful contribution in applying that identity to the right problem: improving three distinct training algorithms for deep generative models. The doubly reparameterized versions of IWAE and reweighted wake-sleep (RWS) further show how IWAE and RWS are related to each other and how they can be combined for potentially further improved performance. The AC believes that the paper makes enough contributions by well presenting the identity in (5) and applying it to the right problems. """
64,"""Lorentzian Distance Learning""","['distance learning', 'metric learning', 'hyperbolic geometry', 'hierarchy tree']","""This paper introduces an approach to learn representations based on the Lorentzian distance in hyperbolic geometry. Hyperbolic geometry is especially suited to hierarchically-structured datasets, which are prevalent in the real world. Current hyperbolic representation learning methods compare examples with the Poincar\'e distance metric. They formulate the problem as minimizing the distance of each node in a hierarchy with its descendants while maximizing its distance with other nodes. This formulation produces node representations close to the centroid of their descendants. We exploit the fact that the centroid w.r.t the squared Lorentzian distance can be written in closed-form. We show that the Euclidean norm of such a centroid decreases as the curvature of the hyperbolic space decreases. This property makes it appropriate to represent hierarchies where parent nodes minimize the distances to their descendants and have smaller Euclidean norm than their children. Our approach obtains state-of-the-art results in retrieval and classification tasks on different datasets. ""","""Dear authors, The reviewers all appreciated the treatment of the topic and the quality of the writing. It is rare for all reviewers to agree on this, congratulations. However, all reviewers also felt that the paper could have gone further in its analysis. In particular, they noticed that quite a few points were either mentioned in recent papers or lacked an experimental validation. Given the reviews, I strongly encourage the authors to expand on their findings and submit the improved work to a future conference."""
65,"""A Synaptic Neural Network and Synapse Learning""","['synaptic neural network', 'surprisal', 'synapse', 'probability', 'excitation', 'inhibition', 'synapse learning', 'bose-einstein distribution', 'tensor', 'gradient', 'loss function', 'mnist', 'topologically conjugate']","""A Synaptic Neural Network (SynaNN) consists of synapses and neurons. Inspired by the synapse research of neuroscience, we built a synapse model with a nonlinear synapse function of excitatory and inhibitory channel probabilities. Introduced the concept of surprisal space and constructed a commutative diagram, we proved that the inhibitory probability function -log(1-exp(-x)) in surprisal space is the topologically conjugate function of the inhibitory complementary probability 1-x in probability space. Furthermore, we found that the derivative of the synapse over the parameter in the surprisal space is equal to the negative Bose-Einstein distribution. In addition, we constructed a fully connected synapse graph (tensor) as a synapse block of a synaptic neural network. Moreover, we proved the gradient formula of a cross-entropy loss function over parameters, so synapse learning can work with the gradient descent and backpropagation algorithms. In the proof-of-concept experiment, we performed an MNIST training and testing on the MLP model with synapse network as hidden layers.""","""In this paper, neural networks are taken a step further by increasing their biological likeliness. In particular, a model of the membranes of biological cells are used computationally to train a neural network. The results are validated on MNIST. The paper argumentation is not easy to follow, and all reviewers agree that the text needs to be improved. The neuroscience sources that the models are based on are possibly outdated. Finally, the results are too meagre and, in the end, not well compared with competing approaches. All in all, the merit of this approach is not fully demonstrated, and further work seems to be needed to clarify this."""
66,"""Learning Multi-Level Hierarchies with Hindsight""","['Hierarchical Reinforcement Learning', 'Reinforcement Learning', 'Deep Reinforcement Learning']","""Hierarchical agents have the potential to solve sequential decision making tasks with greater sample efficiency than their non-hierarchical counterparts because hierarchical agents can break down tasks into sets of subtasks that only require short sequences of decisions. In order to realize this potential of faster learning, hierarchical agents need to be able to learn their multiple levels of policies in parallel so these simpler subproblems can be solved simultaneously. Yet, learning multiple levels of policies in parallel is hard because it is inherently unstable: changes in a policy at one level of the hierarchy may cause changes in the transition and reward functions at higher levels in the hierarchy, making it difficult to jointly learn multiple levels of policies. In this paper, we introduce a new Hierarchical Reinforcement Learning (HRL) framework, Hierarchical Actor-Critic (HAC), that can overcome the instability issues that arise when agents try to jointly learn multiple levels of policies. The main idea behind HAC is to train each level of the hierarchy independently of the lower levels by training each level as if the lower level policies are already optimal. We demonstrate experimentally in both grid world and simulated robotics domains that our approach can significantly accelerate learning relative to other non-hierarchical and hierarchical methods. Indeed, our framework is the first to successfully learn 3-level hierarchies in parallel in tasks with continuous state and action spaces.""","""As per R3: This paper presents a novel approach for doing hierarchical deep RL (HRL) on UMDPs by: (a) use of hindsight experience replay at multiple levels; combined with (b) max T timesteps at each level. By effectively learning from missed goals at multiple levels, it allows for fairly data-efficient learning and can (in principle) work for an arbitrary number of levels. HRL is an important open problem. The weaknesses described reviewers include limited comparisons to other HRL methods; its applicaiton to fairly simple domain; its still unclear what the benefit of >=4 levels is, and what the diminishing returns are wrt to the claim of working for an arbitrary number of levels. R1(5) and R3(7) stand by their scores. R1(5) still has some remaining concerns regarding some experiments not being done across all tasks, an older version of the HAL algo baseline being used, and lack of insight regarding >= 4 levels. Based on the balance of the reviewers comments and the AC's own reading of the paper and results, and the importance of the problem, the AC leans towards accept. Using Hindsight Exp Replay across multiple levels is a simple-but-interesting idea, and the terminate-after-T steps is an interesting heuristic to make this effective. While the paper does not give insight for large (>=4) levels, it does make for an interesting framework that will inspire further work. The AC recommends that the claims regarding an ""arbitrary number of levels"" be significantly toned down. """
67,"""Teacher Guided Architecture Search""","['hyperparameter search', 'architecture search', 'convolutional neural networks']","""Strong improvements in neural network performance in vision tasks have resulted from the search of alternative network architectures, and prior work has shown that this search process can be automated and guided by evaluating candidate network performance following limited training (Performance Guided Architecture Search or PGAS). However, because of the large architecture search spaces and the high computational cost associated with evaluating each candidate model, further gains in computational efficiency are needed. Here we present a method termed Teacher Guided Search for Architectures by Generation and Evaluation (TG-SAGE) that produces up to an order of magnitude in search efficiency over PGAS methods. Specifically, TG-SAGE guides each step of the architecture search by evaluating the similarity of internal representations of the candidate networks with those of the (fixed) teacher network. We show that this procedure leads to significant reduction in required per-sample training and that, this advantage holds for two different search spaces of architectures, and two different search algorithms. We further show that in the space of convolutional cells for visual categorization, TG-SAGE finds a cell structure with similar performance as was previously found using other methods but at a total computational cost that is two orders of magnitude lower than Neural Architecture Search (NAS) and more than four times lower than progressive neural architecture search (PNAS). These results suggest that TG-SAGE can be used to accelerate network architecture search in cases where one has access to some or all of the internal representations of a teacher network of interest, such as the brain. ""","""The authors propose to accelerate neural architecture search by using feature similarity with a given teacher network to measure how good a new candidate architecture is. The experiments show that the method accelerates architecture search, and has competitive performance. However, both Reviewers 1 and 3 noted questionable motivation behind the approach, as the method assumes that there already exists a strong teacher network in the domain where we architecture search is performed, which is not always the case. The rebuttal and the revised version of the paper addressed some of the reviewers' concerns, but overall the paper remained below the acceptance bar. I suggest that the authors further expand the evaluation and motivate their approach better before re-submitting to another venue. """
68,"""Dj Vu: An Empirical Evaluation of the Memorization Properties of Convnets""","['membership inference', 'memorization', 'attack', 'privacy']","""Convolutional neural networks memorize part of their training data, which is why strategies such as data augmentation and drop-out are employed to mitigate over- fitting. This paper considers the related question of membership inference, where the goal is to determine if an image was used during training. We con- sider membership tests over either ensembles of samples or individual samples. First, we show how to detect if a dataset was used to train a model, and in particular whether some validation images were used at train time. Then, we introduce a new approach to infer membership when a few of the top layers are not available or have been fine-tuned, and show that lower layers still carry information about the training samples. To support our findings, we conduct large-scale experiments on Imagenet and subsets of YFCC-100M with modern architectures such as VGG and Resnet. ""","""This paper studies memorization properties of convnets by testing their ability to determine if an image/set of images was used during training or not. The experiments are reported on large-scale datasets using high-capacity networks. While acknowledging that the proposed model is potentially useful, the reviewers raised several important concerns that were viewed by AC as critical issues: (1) more formal justifications are required to assess the scope and significance of this work contributions -- see very detailed comments by R2 about measuring networks capacity to memorize and the role of network weights and depth as studied in MacKay,2002. In their response the authors acknowledged they didnt take into account network weights and depth but strived at an empirical evaluation scenario. (2) writing and presentation clarity of the paper could be substantially improved see very detailed comments by R3 and also R2; (3) empirical evaluations and effect of the negative set used for training are not well explained and analysed (R2, R3). AC can confirm that all three reviewers have read the author responses and have contributed to the final discussion. AC suggests, in its current state the manuscript is not ready for a publication. We hope the reviews are useful for improving and revising the paper. """
69,"""Distributionally Robust Optimization Leads to Better Generalization: on SGD and Beyond""","['distributionally robust optimization', 'deep learning', 'SGD', 'learning theory']","""In this paper, we adopt distributionally robust optimization (DRO) (Ben-Tal et al., 2013) in hope to achieve a better generalization in deep learning tasks. We establish the generalization guarantees and analyze the localized Rademacher complexity for DRO, and conduct experiments to show that DRO obtains a better performance. We reveal the profound connection between SGD and DRO, i.e., selecting a batch can be viewed as choosing a distribution over the training set. From this perspective, we prove that SGD is prone to escape from bad stationary points and small batch SGD outperforms large batch SGD. We give an upper bound for the robust loss when SGD converges and keeps stable. We propose a novel Weighted SGD (WSGD) algorithm framework, which assigns high-variance weights to the data of the current batch. We devise a practical implement of WSGD that can directly optimize the robust loss. We test our algorithm on CIFAR-10 and CIFAR-100, and WSGD achieves significant improvements over the conventional SGD.""","""This paper received high quality reviews, which highlighted numerous issues with the paper. A common criticism was that the results in the paper seemed disconnected. Numerous technical concerns were raised. Reading the responses, it seems that some of these issues are nonissues, but it seems also that the writing was not sufficiently up to the standard required of this type of technical work. I suggest the authors produce a rewrite and resubmit to the next ML conference, taking the criticisms they've received here very seriously."""
70,"""Metropolis-Hastings view on variational inference and adversarial training""","['MCMC', 'GANs', 'Variational Inference']","""In this paper we propose to view the acceptance rate of the Metropolis-Hastings algorithm as a universal objective for learning to sample from target distribution -- given either as a set of samples or in the form of unnormalized density. This point of view unifies the goals of such approaches as Markov Chain Monte Carlo (MCMC), Generative Adversarial Networks (GANs), variational inference. To reveal the connection we derive the lower bound on the acceptance rate and treat it as the objective for learning explicit and implicit samplers. The form of the lower bound allows for doubly stochastic gradient optimization in case the target distribution factorizes (i.e. over data points). We empirically validate our approach on Bayesian inference for neural networks and generative models for images.""","""This paper provides a good finding that maximizing a lower bound of the M-H acceptance rate is equivalent to minimizing the symmetric KL divergence between target the proposal. This lower bound is then used to learn sampler for both density and sample-based settings. It also nicely connects GAN with MCMC by providing a novel loss function to train the discriminator. Experiment on MNIST dataset in Sec 4.2 shows training the proposal with the symmetric KL is better than variational inference that optimizes KL(q||p). However, there are a few concerns raised in both the reviews and other comments that should be further clarified. 1. Training an independent proposal may reduce the rate of convergence. 2. In the density-based setting experiments, the learnt independent proposal is only used to provide an initial point and a random-walk kernel is actually used for sampling. This is different from what is proposed algorithm in Section 3. 3. The proposed algorithm is only compared with VI in density-based setting, and there are no comparison with other baselines in the sample-based setting, despite the close connections of the proposed method with other models. Stochastic gradient MCMC methods, A-NICE-MC, GAN will be good baselines for empirical comparisons. Also, the dataset in Sec 4.2 is a subset of the standard MNIST, which makes comparison with other literatures difficult. For the first concern, the authors provided new experiments for low-dimensional synthetic distributions. It is very helpful to show the comparable performance with A-NICE-MC in this case, but the real challenge in high-dimensional distributions remains unexamined. For the second concern, the authors consider the use of random-walk as a heuristic that allows to obtain better samples from the posterior, but that significantly changes the proposed transition kernel in Alg. 1. This paper would be significantly stronger and make a very good contribution to this area by addressing the problems above."""
71,"""Discovering Low-Precision Networks Close to Full-Precision Networks for Efficient Embedded Inference""","['Deep Learning', 'Convolutional Neural Networks', 'Low-precision inference', 'Network quantization']","""To realize the promise of ubiquitous embedded deep network inference, it is essential to seek limits of energy and area efficiency. To this end, low-precision networks offer tremendous promise because both energy and area scale down quadratically with the reduction in precision. Here, for the first time, we demonstrate ResNet-18, ResNet-34, ResNet-50, ResNet-152, Inception-v3, densenet-161, and VGG-16bn networks on the ImageNet classification benchmark that, at 8-bit precision exceed the accuracy of the full-precision baseline networks after one epoch of finetuning, thereby leveraging the availability of pretrained models. We also demonstrate ResNet-18, ResNet-34, and ResNet-50 4-bit models that match the accuracy of the full-precision baseline networks -- the highest scores to date. Surprisingly, the weights of the low-precision networks are very close (in cosine similarity) to the weights of the corresponding baseline networks, making training from scratch unnecessary. We find that gradient noise due to quantization during training increases with reduced precision, and seek ways to overcome this noise. The number of iterations required by stochastic gradient descent to achieve a given training error is related to the square of (a) the distance of the initial solution from the final plus (b) the maximum variance of the gradient estimates. By drawing inspiration from this observation, we (a) reduce solution distance by starting with pretrained fp32 precision baseline networks and fine-tuning, and (b) combat noise introduced by quantizing weights and activations during training, by using larger batches along with matched learning rate annealing. Sensitivity analysis indicates that these techniques, coupled with proper activation function range calibration, offer a promising heuristic to discover low-precision networks, if they exist, close to fp32 precision baseline networks. ""","""This paper proposes methods to improve the performance of the low-precision neural networks. The reviewers raised concern about lack of novelty. Due to insufficient technical contribution, recommend for rejection. """
72,"""Efficient Convolutional Neural Network Training with Direct Feedback Alignment""","['Direct Feedback Alignment', 'Convolutional Neural Network', 'DNN Training']","""There were many algorithms to substitute the back-propagation (BP) in the deep neural network (DNN) training. However, they could not become popular because their training accuracy and the computational efficiency were worse than BP. One of them was direct feedback alignment (DFA), but it showed low training performance especially for the convolutional neural network (CNN). In this paper, we overcome the limitation of the DFA algorithm by combining with the conventional BP during the CNN training. To improve the training stability, we also suggest the feedback weight initialization method by analyzing the patterns of the fixed random matrices in the DFA. Finally, we propose the new training algorithm, binary direct feedback alignment (BDFA) to minimize the computational cost while maintaining the training accuracy compared with the DFA. In our experiments, we use the CIFAR-10 and CIFAR-100 dataset to simulate the CNN learning from the scratch and apply the BDFA to the online learning based object tracking application to examine the training in the small dataset environment. Our proposed algorithms show better performance than conventional BP in both two different training tasks especially when the dataset is small.""","""This paper proposes a training algorithm for ConvNet architectures in which the final few layers are fully connected. The main idea is to use direct feedback alignment with carefully chosen binarized (1) weights to train the fully connected layers and backpropagation to train the convolutional layers. The binarization reduces the memory footprint and computational cost of direct feedback alignment, while the careful selection of feedback weights improves convergence. Experiments on CIFAR-10, CIFAR-100, and an object tracking task are provided to show that the proposed algorithm outperforms backpropagation, especially when the amount of training data is small. The reviewers felt that the paper does a terrific job of introducing the various training algorithms --- backpropagation, feedback alignment, and direct feedback alignment --- and that the paper clearly explained what the novel contributions were. However, the reviewers felt the paper had limited novelty because it combines ideas that were already known, that it has limited applicability because it will not work with fully convolutional architectures, that the baselines in the experiments were somewhat weak, and that the paper provided no insights on why the proposed algorithm might be better than backpropagation in some cases. Regrettably, only one reviewer (R2) participated in the discussion, though this was the reviewer who provided the most constructive review. The AC read the revised paper, and agrees with R2's concerns about the limited applicability of the proposed algorithm and lack of insight or analysis explaining why the proposed training algorithm would improve over backpropagation."""
73,"""Generative Code Modeling with Graphs""","['Generative Model', 'Source Code', 'Graph Learning']","""Generative models forsource code are an interesting structured prediction problem, requiring to reason about both hard syntactic and semantic constraints as well as about natural, likely programs. We present a novel model for this problem that uses a graph to represent the intermediate state of the generated output. Our model generates code by interleaving grammar-driven expansion steps with graph augmentation and neural message passing steps. An experimental evaluation shows that our new model can generate semantically meaningful expressions, outperforming a range of strong baselines.""","""This paper presents an interesting method for code generation using a graph-based generative approach. Empirical evaluation shows that the method outperforms relevant baselines (PHOG). There is consensus among reviewers that the methods are novel and is worth acceptance to ICLR."""
74,"""PPD: Permutation Phase Defense Against Adversarial Examples in Deep Learning""","['permutation phase defense', 'adversarial attacks', 'deep learning']","""Deep neural networks have demonstrated cutting edge performance on various tasks including classification. However, it is well known that adversarially designed imperceptible perturbation of the input can mislead advanced classifiers. In this paper, Permutation Phase Defense (PPD), is proposed as a novel method to resist adversarial attacks. PPD combines random permutation of the image with phase component of its Fourier transform. The basic idea behind this approach is to turn adversarial defense problems analogously into symmetric cryptography, which relies solely on safekeeping of the keys for security. In PPD, safe keeping of the selected permutation ensures effectiveness against adversarial attacks. Testing PPD on MNIST and CIFAR-10 datasets yielded state-of-the-art robustness against the most powerful adversarial attacks currently available.""","""This paper presents a new defense against adversarial examples using random permutations and a Fourier transform. The technique is clearly novel, and the paper is clearly written. However, as the reviewers and commenters pointed out, there is a significant degradation in natural accuracy, which does not seem to be easily recoverable. This degradation is due to the random permutation of the images, which effectively disallows the use of convolutions. Furthermore, Reviewer 1 points out that the baselines are insufficient, as the authors do not explore (a) learning the transformation, or (b) using expectation over transformation to attack the model. This concern is further validated by the fact that Black-box attacks are often the best-performing, which is a sign of gradient masking. The authors try to address this by performing an attack against an ensemble of models, and against a substitute model attack. However, attacking an ensemble is not equivalent to optimizing the expectation, which would require sampling a new permutation at each step. The paper thus requires significantly stronger baselines and attacks."""
75,"""Incremental Few-Shot Learning with Attention Attractor Networks""","['meta-learning', 'few-shot learning', 'incremental learning']","""Machine learning classifiers are often trained to recognize a set of pre-defined classes. However, in many real applications, it is often desirable to have the flexibility of learning additional concepts, without re-training on the full training set. This paper addresses this problem, incremental few-shot learning, where a regular classification network has already been trained to recognize a set of base classes; and several extra novel classes are being considered, each with only a few labeled examples. After learning the novel classes, the model is then evaluated on the overall performance of both base and novel classes. To this end, we propose a meta-learning model, the Attention Attractor Network, which regularizes the learning of novel classes. In each episode, we train a set of new weights to recognize novel classes until they converge, and we show that the technique of recurrent back-propagation can back-propagate through the optimization process and facilitate the learning of the attractor network regularizer. We demonstrate that the learned attractor network can recognize novel classes while remembering old classes without the need to review the original training set, outperforming baselines that do not rely on an iterative optimization process.""","""This paper proposes an approach for incremental learning of new classes using meta-learning. Strengths: The framework is interesting. The reviewers agree that the paper is well-written and clear. The experiments include comparisons to prior work, and the ablation studies are useful for judging the performance of the method. Weaknesses: The paper does not provide significant insights over Gidaris & Komodakis '18. Reviewer 1 was also concerned that the motivation for RBP is not entirely clear. Overall, the reviewers found that the strengths did not outweigh the weaknesses. Hence, I recommend reject. """
76,"""Spreading vectors for similarity search""","['dimensionality reduction', 'similarity search', 'indexing', 'differential entropy']","""Discretizing floating-point vectors is a fundamental step of modern indexing methods. State-of-the-art techniques learn parameters of the quantizers on training data for optimal performance, thus adapting quantizers to the data. In this work, we propose to reverse this paradigm and adapt the data to the quantizer: we train a neural net whose last layers form a fixed parameter-free quantizer, such as pre-defined points of a sphere. As a proxy objective, we design and train a neural network that favors uniformity in the spherical latent space, while preserving the neighborhood structure after the mapping. For this purpose, we propose a new regularizer derived from the Kozachenko-Leonenko differential entropy estimator and combine it with a locality-aware triplet loss. Experiments show that our end-to-end approach outperforms most learned quantization methods, and is competitive with the state of the art on widely adopted benchmarks. Further more, we show that training without the quantization step results in almost no difference in accuracy, but yields a generic catalyser that can be applied with any subsequent quantization technique. """,""". Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The proposed method is novel and effective - The paper is clear and the experiments and literature review are sufficient (especially after revision). 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. The original weaknesses (mainly clarity and missing details) were adequately addressed in the revisions. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. No major points of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be accepted."""
77,"""VHEGAN: Variational Hetero-Encoder Randomized GAN for Zero-Shot Learning""","['Deep generative models', 'deep topic modeling', 'generative adversarial learning', 'variational encoder', 'zero-short learning']","""To extract and relate visual and linguistic concepts from images and textual descriptions for text-based zero-shot learning (ZSL), we develop variational hetero-encoder (VHE) that decodes text via a deep probabilisitic topic model, the variational posterior of whose local latent variables is encoded from an image via a Weibull distribution based inference network. To further improve VHE and add an image generator, we propose VHE randomized generative adversarial net (VHEGAN) that exploits the synergy between VHE and GAN through their shared latent space. After training with a hybrid stochastic-gradient MCMC/variational inference/stochastic gradient descent inference algorithm, VHEGAN can be used in a variety of settings, such as text generation/retrieval conditioning on an image, image generation/retrieval conditioning on a document/image, and generation of text-image pairs. The efficacy of VHEGAN is demonstrated quantitatively with experiments on both conventional and generalized ZSL tasks, and qualitatively on (conditional) image and/or text generation/retrieval.""","""The paper received borderline ratings due to concerns regarding novelty and experimental results/settings (e.g. zero shot learning). On my side, I believe that the proposed method would need more evaluations on other benchmarks (e.g., SUN, AWA1 and AWA2) for both ZSL and GZSL settings to make the results more convincing. Overall, none of the reviewers championed this paper and I would recommend weak rejection."""
78,"""SOLAR: Deep Structured Representations for Model-Based Reinforcement Learning""","['model-based reinforcement learning', 'structured representation learning', 'robotics']","""Model-based reinforcement learning (RL) methods can be broadly categorized as global model methods, which depend on learning models that provide sensible predictions in a wide range of states, or local model methods, which iteratively refit simple models that are used for policy improvement. While predicting future states that will result from the current actions is difficult, local model methods only attempt to understand system dynamics in the neighborhood of the current policy, making it possible to produce local improvements without ever learning to predict accurately far into the future. The main idea in this paper is that we can learn representations that make it easy to retrospectively infer simple dynamics given the data from the current policy, thus enabling local models to be used for policy learning in complex systems. We evaluate our approach against other model-based and model-free RL methods on a suite of robotics tasks, including manipulation tasks on a real Sawyer robotic arm directly from camera images.""","""This paper proposes a method to learn representations to infer simple local models that can be used for policy improvement. All the reviewers agree that the paper has interesting ideas, but they found the main contribution to be a bit weak and the experiments to be insufficient. Post rebuttal, the reviewers discussed extensively with each other and agreed that, given more work is done on a clear presentation and improving the experiments, this paper can be accepted. In its current form however, the paper is not ready to be accepted. I have recommended to reject this paper, but I will encourage the authors to resubmit after improving the work. """
79,"""Improved Learning of One-hidden-layer Convolutional Neural Networks with Overlaps""","['deep learning', 'parameter recovery', 'convolutional neural networks', 'non-convex optimization']","""We propose a new algorithm to learn a one-hidden-layer convolutional neural network where both the convolutional weights and the outputs weights are parameters to be learned. Our algorithm works for a general class of (potentially overlapping) patches, including commonly used structures for computer vision tasks. Our algorithm draws ideas from (1) isotonic regression for learning neural networks and (2) landscape analysis of non-convex matrix factorization problems. We believe these findings may inspire further development in designing provable algorithms for learning neural networks and other complex models. While our focus is theoretical, we also present experiments that illustrate our theoretical findings.""","""The reviewers seem to reach a consensus that the contribution of the paper is somewhat incremental give the prior work of Goel et al and that a main drawback of the paper is that it's not clear the similar technique can be applied to multiple **convolutional filters**. The authors mentioned in the response that some of the techniques can be heuristically applied to multiple layers, but the AC is skeptical about it because, with multiple layers and multiple convolutional filters, one has to deal with the permutation invariance caused by the multiple convolutional filters. (It's unclear to the AC how one could have a meaningful setting with multiple layers but a single convolution filters.) """
80,"""L2-Nonexpansive Neural Networks""","['adversarial defense', 'regularization', 'robustness', 'generalization']","""This paper proposes a class of well-conditioned neural networks in which a unit amount of change in the inputs causes at most a unit amount of change in the outputs or any of the internal layers. We develop the known methodology of controlling Lipschitz constants to realize its full potential in maximizing robustness, with a new regularization scheme for linear layers, new ways to adapt nonlinearities and a new loss function. With MNIST and CIFAR-10 classifiers, we demonstrate a number of advantages. Without needing any adversarial training, the proposed classifiers exceed the state of the art in robustness against white-box L2-bounded adversarial attacks. They generalize better than ordinary networks from noisy data with partially random labels. Their outputs are quantitatively meaningful and indicate levels of confidence and generalization, among other desirable properties.""",""" * Strengths This paper studies adversarial robustness to perturbations that are bounded in the L2 norm. It is motivated by a theoretical sufficient condition (non-expansiveness) but rather than trying to formally verify robustness, it uses this condition as inspiration, modifying standard network architectures in several ways to encourage non-expansiveness while mostly preserving computational efficiency and accuracy. This theory-inspired practically-focused hybrid is a rare perspective in this area and could fruitfully inspire further improvements. Finally, the paper came under substantial scrutiny during the review period (there are 65 comments on the page) and the authors have convincingly answered a number of technical criticisms. * Weaknesses One reviewer and some commenters were concerned that the L2 norm is not a realistic norm to measure adversarial attacks in. There were also concerns that the empirical level of robustness of the network was too weak to be meaningful. In addition, while some parts of the experiments were thorough and some parts of the paper were well-presented, the quality was not uniform throughout. Finally, while the proposed changes improve adversarial robustness, they also decrease the accuracy of the network on clean examples (this is to be expected but may be an issue in practice). * Discussion There was substantial disagreement on whether to accept the paper. On the one hand, there has been limited progress on robustness to adversarial examples (even under simple norms such as the L2 norm) and most methods that do work are based on formal verification and therefore quite computationally expensive. On the other hand, simple norms such as the L2 norm are somewhat contrived and mainly chosen for convenience (although doing well in the L2 norm is a necessary condition for being robust to more general attacks). Moreover, the empirical results are currently too weak to confer meaningful robustness even under the L2 norm. * Decision While I agree with the reviewers and commenters who are skeptical of the L2 norm model (and would very much like to see approaches that consider more realistic threat models), I decided to accept the paper for two reasons: first, doing well in L2 is a necessary condition to doing well in more general models, and the ideas and approach here are simple enough that they might provide inspiration in these more general models as well. Additionally, this was one of the strongest adversarial defense papers at ICLR this year in terms of credibility of the claims (certainly the strongest in my pile) and contains several useful ideas as well as novel empirical findings (such as the increased success of attacks up to 1 million iterations)."""
81,"""Modulated Variational Auto-Encoders for Many-to-Many Musical Timbre Transfer""","['Musical Timbre', 'Instrument Translation', 'Domain Translation', 'Style Transfer', 'Sound Synthesis', 'Musical Information', 'Deep Learning', 'Variational Auto-Encoder', 'Generative Models', 'Network Conditioning']","""Generative models have been successfully applied to image style transfer and domain translation. However, there is still a wide gap in the quality of results when learning such tasks on musical audio. Furthermore, most translation models only enable one-to-one or one-to-many transfer by relying on separate encoders or decoders and complex, computationally-heavy models. In this paper, we introduce the Modulated Variational auto-Encoders (MoVE) to perform musical timbre transfer. First, we define timbre transfer as applying parts of the auditory properties of a musical instrument onto another. We show that we can achieve and improve this task by conditioning existing domain translation techniques with Feature-wise Linear Modulation (FiLM). Then, by replacing the usual adversarial translation criterion by a Maximum Mean Discrepancy (MMD) objective, we alleviate the need for an auxiliary pair of discriminative networks. This allows a faster and more stable training, along with a controllable latent space encoder. By further conditioning our system on several different instruments, we can generalize to many-to-many transfer within a single variational architecture able to perform multi-domain transfers. Our models map inputs to 3-dimensional representations, successfully translating timbre from one instrument to another and supporting sound synthesis on a reduced set of control parameters. We evaluate our method in reconstruction and generation tasks while analyzing the auditory descriptor distributions across transferred domains. We show that this architecture incorporates generative controls in multi-domain transfer, yet remaining rather light, fast to train and effective on small datasets.""","""This paper proposes a VAE-based model which is able to perform musical timbre transfer. The reviewers generally find the approach well-motivated. The idea to perform many-to-many transfer within a single architecture is found to be promising. However, there have been some unaddressed concerns, as detailed below. R3 has some methodological concerns regarding negative transfer and asks for more extended experimental section. R1 and R2 ask for more interpretable results and, ultimately, a more conclusive study. R2 specifically finds the results to be insufficient. The authors have agreed with some of the reviewers' feedback but have left most of it unaddressed in a new revision. That could be because some of the recommendations require significant extra work. Given the above, it seems that this paper needs more work before being accepted in ICLR. """
82,"""Unifying Bilateral Filtering and Adversarial Training for Robust Neural Networks""","['Adversarial examples', 'Image denoising']","""Recent analysis of deep neural networks has revealed their vulnerability to carefully structured adversarial examples. Many effective algorithms exist to craft these adversarial examples, but performant defenses seem to be far away. In this work, we explore the use of edge-aware bilateral filtering as a projection back to the space of natural images. We show that bilateral filtering is an effective defense in multiple attack settings, where the strength of the adversary gradually increases. In the case of adversary who has no knowledge of the defense, bilateral filtering can remove more than 90% of adversarial examples from a variety of different attacks. To evaluate against an adversary with complete knowledge of our defense, we adapt the bilateral filter as a trainable layer in a neural network and show that adding this layer makes ImageNet images significantly more robust to attacks. When trained under a framework of adversarial training, we show that the resulting model is hard to fool with even the best attack methods. ""","""The paper proposes a technique for defending against adversarial examples that relies on averaging pixels that are close to each other both in position and value. This approach seems to be an interesting preprocessing technique in the robust training pipeline. However, the actual claims made are not well-supported and, in fact, seem somewhat implausible. """
83,"""Context Dependent Modulation of Activation Function""","['Artificial Neural Network', 'Convolution Neural Network', 'Long Short-Term Memory', 'Activation Function', 'Neuromodulation']","""We propose a modification to traditional Artificial Neural Networks (ANNs), which provides the ANNs with new aptitudes motivated by biological neurons. Biological neurons work far beyond linearly summing up synaptic inputs and then transforming the integrated information. A biological neuron change firing modes accordingly to peripheral factors (e.g., neuromodulators) as well as intrinsic ones. Our modification connects a new type of ANN nodes, which mimic the function of biological neuromodulators and are termed modulators, to enable other traditional ANN nodes to adjust their activation sensitivities in run-time based on their input patterns. In this manner, we enable the slope of the activation function to be context dependent. This modification produces statistically significant improvements in comparison with traditional ANN nodes in the context of Convolutional Neural Networks and Long Short-Term Memory networks. ""","""The paper adds a new level of complexity to neural networks, by modulating activation functions of a layer as a function of the previous layer activations. The method is evaluated on relatively simple vision and language tasks. The idea is nice, but seems to be a special case of previously published work; and the results are not convincing. Four of five reviewers agree that the work would benefit from: improving comparisons with existing approaches, but also improving its theoretical framework, in light of competing approaches."""
84,"""PAIRWISE AUGMENTED GANS WITH ADVERSARIAL RECONSTRUCTION LOSS""","['Computer vision', 'Deep learning', 'Unsupervised Learning', 'Generative Adversarial Networks']","""We propose a novel autoencoding model called Pairwise Augmented GANs. We train a generator and an encoder jointly and in an adversarial manner. The generator network learns to sample realistic objects. In turn, the encoder network at the same time is trained to map the true data distribution to the prior in latent space. To ensure good reconstructions, we introduce an augmented adversarial reconstruction loss. Here we train a discriminator to distinguish two types of pairs: an object with its augmentation and the one with its reconstruction. We show that such adversarial loss compares objects based on the content rather than on the exact match. We experimentally demonstrate that our model generates samples and reconstructions of quality competitive with state-of-the-art on datasets MNIST, CIFAR10, CelebA and achieves good quantitative results on CIFAR10. """,""" The paper proposes an augmented adversarial reconstruction loss for training a stochastic encoder-decoder architecture. It corresponds to a discriminator loss distinguishing between a pair of a sample from the data distribution and its augmentation and pair containing the sample and its reconstruction. The introduction of the augmentation function is an interesting idea, intensively tested in a set of experiments, but, as two of the reviewers pointed out, the paper could be improved by deeper investigation of the augmentation function and the way of choosing it, which would increase significance of the contribution. """
85,"""Consistent Jumpy Predictions for Videos and Scenes""","['jumpy predictions', 'generative models', 'scene reconstruction', 'video prediction', 'variational auto-encoders', 'DRAW']","""Stochastic video prediction models take in a sequence of image frames, and generate a sequence of consecutive future image frames. These models typically generate future frames in an autoregressive fashion, which is slow and requires the input and output frames to be consecutive. We introduce a model that overcomes these drawbacks by generating a latent representation from an arbitrary set of frames that can then be used to simultaneously and efficiently sample temporally consistent frames at arbitrary time-points. For example, our model can ""jump"" and directly sample frames at the end of the video, without sampling intermediate frames. Synthetic video evaluations confirm substantial gains in speed and functionality without loss in fidelity. We also apply our framework to a 3D scene reconstruction dataset. Here, our model is conditioned on camera location and can sample consistent sets of images for what an occluded region of a 3D scene might look like, even if there are multiple possibilities for what that region might contain. Reconstructions and videos are available at pseudo-url. ""","""This paper proposes a probabilistic model for data indexed by an observed parameter (such as time in video frames, or camera locations in 3d scenes), which enables a global encoding of all available frames and is able to sample consistently at arbitrary indexes. Experiments are reported on several synthetic datasets. Reviewers acknowledged the significance of the proposed model, noted that the paper is well-written, and the design choices are sounds. However, they also expressed concerns about the experimental setup, which only includes synthetic examples. Although the authors acknowledged during the response phase that this is indeed a current limitation, they argued it is not specific to their particular architecture, but to the task itself. Another concern raised by R1 is the lack of clarity in some experimental setups (for instance where only a subset of the best runs are used to compute error bars, and this subset appears to be of different size depending on the experiment, cf fig 5), and the fact that the datasets used in this paper to compare against GQNs are specifically designed. Overall, this is a really borderline submission, with several strengths and weaknesses. After taking the reviewer discussion into account and making his/her own assessment, the AC recommends rejection at this time, but strongly encourages the authors to resubmit their work after improving their experimental setup, which will make the paper much stronger."""
86,"""Learning powerful policies and better dynamics models by encouraging consistency""","['model-based reinforcement learning', 'deep learning', 'generative agents', 'policy gradient', 'imitation learning']","""Model-based reinforcement learning approaches have the promise of being sample efficient. Much of the progress in learning dynamics models in RL has been made by learning models via supervised learning. There is enough evidence that humans build a model of the environment, not only by observing the environment but also by interacting with the environment. Interaction with the environment allows humans to carry out ""experiments"": taking actions that help uncover true causal relationships which can be used for building better dynamics models. Analogously, we would expect such interaction to be helpful for a learning agent while learning to model the environment dynamics. In this paper, we build upon this intuition, by using an auxiliary cost function to ensure consistency between what the agent observes (by acting in the real world) and what it imagines (by acting in the ``learned'' world). Our empirical analysis shows that the proposed approach helps to train powerful policies as well as better dynamics models.""","""The paper proposes and approach for model-based reinforcement learning that adds a constraint to encourage the predictions from the model to be consistent with the observations from the environment. The reviewers had substantial concerns about the clarify of the initial submission, which has been significantly improved in revisions of the paper. The experiments have also been improved. Strengths: The method is simple, the performance is competitive with state-of-the-art approaches, and the experiments are thorough including comparisons on seven different environments. Weaknesses: The main concern of the reviewers is the lack of concrete discussion about how the method compares to prior work. While the paper cites many different prior methods, the paper would be significantly improved by explicitly comparing and contrasting the ideas presented in this paper and those presented in prior work. A secondary weakness is that, while the results appear to be statistically significant, the improvement over prior methods is still relatively small. I do not think that this paper meets the bar for publication without an improved discussion of how this work is placed among the existing literature and without more convincing results. As a side note, the authors should consider comparing to the below NeurIPS '18 paper, which significantly exceeds the performance of Nagabandi et al '17: pseudo-url"""
87,"""Auxiliary Variational MCMC""","['MCMC', 'Variational Inference']","""We introduce Auxiliary Variational MCMC, a novel framework for learning MCMC kernels that combines recent advances in variational inference with insights drawn from traditional auxiliary variable MCMC methods such as Hamiltonian Monte Carlo. Our framework exploits low dimensional structure in the target distribution in order to learn a more efficient MCMC sampler. The resulting sampler is able to suppress random walk behaviour and mix between modes efficiently, without the need to compute gradients of the target distribution. We test our sampler on a number of challenging distributions, where the underlying structure is known, and on the task of posterior sampling in Bayesian logistic regression. Code to reproduce all experiments is available at pseudo-url . ""","""The reviewers all argued for acceptance citing the novelty and potential of the work as strengths. They all found the experiments a little underwhelming and asked for more exciting empirical evaluation. The authors have addressed this somewhat by including multi-modal experiments in the discussion period. The paper would be more impactful if the authors could demonstrate significant improvements on really challenging problems where MCMC is currently prohibitively expensive, such as improving over HMC for highly parameterized deep neural networks. Overall, however, this is a very nice paper and warrants acceptance to the conference."""
88,"""Information-Directed Exploration for Deep Reinforcement Learning""","['reinforcement learning', 'exploration', 'information directed sampling']","""Efficient exploration remains a major challenge for reinforcement learning. One reason is that the variability of the returns often depends on the current state and action, and is therefore heteroscedastic. Classical exploration strategies such as upper confidence bound algorithms and Thompson sampling fail to appropriately account for heteroscedasticity, even in the bandit setting. Motivated by recent findings that address this issue in bandits, we propose to use Information-Directed Sampling (IDS) for exploration in reinforcement learning. As our main contribution, we build on recent advances in distributional reinforcement learning and propose a novel, tractable approximation of IDS for deep Q-learning. The resulting exploration strategy explicitly accounts for both parametric uncertainty and heteroscedastic observation noise. We evaluate our method on Atari games and demonstrate a significant improvement over alternative approaches.""","""The paper introduces a method for using information directed sampling, by taking advantage of recent advances in computing parametric uncertainty and variance estimates for returns. These estimates are used to estimate the information gain, based on a formula from (Kirschner & Krause, 2018) for the bandit setting. This paper takes these ideas and puts them together in a reasonably easy-to-use and understandable way for the reinforcement learning setting, which is both nontrivial and useful. The work then demonstrates some successes in Atari. Though it is of course laudable that the paper runs on 57 Atari games, it would make the paper even stronger if a simpler setting (some toy domain) was investigated to more systematically understand this approach and some choices in the approach."""
89,"""Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet""","['interpretability', 'representation learning', 'bag of features', 'deep learning', 'object recognition']","""Deep Neural Networks (DNNs) excel on many complex perceptual tasks but it has proven notoriously difficult to understand how they reach their decisions. We here introduce a high-performance DNN architecture on ImageNet whose decisions are considerably easier to explain. Our model, a simple variant of the ResNet-50 architecture called BagNet, classifies an image based on the occurrences of small local image features without taking into account their spatial ordering. This strategy is closely related to the bag-of-feature (BoF) models popular before the onset of deep learning and reaches a surprisingly high accuracy on ImageNet (87.6% top-5 for 32 x 32 px features and Alexnet performance for 16 x16 px features). The constraint on local features makes it straight-forward to analyse how exactly each part of the image influences the classification. Furthermore, the BagNets behave similar to state-of-the art deep neural networks such as VGG-16, ResNet-152 or DenseNet-169 in terms of feature sensitivity, error distribution and interactions between image parts. This suggests that the improvements of DNNs over previous bag-of-feature classifiers in the last few years is mostly achieved by better fine-tuning rather than by qualitatively different decision strategies.""","""This paper presents an approach that relies on DNNs and bags of features that are fed into them, towards object recognition. The strength of the papers lie in the strong performance of these simple and interpretable models compared to more complex architectures. The authors stress on the interpretability of the results that is indeed a strength of this paper. There is plenty of discussion between the first reviewer and the authors regarding the novelty of the work as the former point out to several related papers; however, the authors provide relatively convincing rebuttal of the concerns. Overall, after the long discussion, there is enough consensus for this paper to be accepted to the conference."""
90,"""Siamese Capsule Networks""","['capsule networks', 'pairwise learning', 'few-shot learning', 'face verification']","""Capsule Networks have shown encouraging results on defacto benchmark computer vision datasets such as MNIST, CIFAR and smallNORB. Although, they are yet to be tested on tasks where (1) the entities detected inherently have more complex internal representations and (2) there are very few instances per class to learn from and (3) where point-wise classification is not suitable. Hence, this paper carries out experiments on face verification in both controlled and uncontrolled settings that together address these points. In doing so we introduce Siamese Capsule Networks, a new variant that can be used for pairwise learning tasks. The model is trained using contrastive loss with l2-normalized capsule encoded pose features. We find that Siamese Capsule Networks perform well against strong baselines on both pairwise learning datasets, yielding best results in the few-shot learning setting where image pairs in the test set contain unseen subjects.""","""The paper extends capsule networks with a pairwise learning objective and evaluates on small face verification datasets. The authors do a great job describing prior work, but lack clarity when articulating their contribution and proposed method. In addition, some important implementation details, such as hyperparameter selection, are missing causing further confusion as to the final approach. Overall, according to the experiments shown, the approach offers modest improvements over prior work. The approach offers an interesting and promising direction. We encourage the authors to revise the manuscript to clarify their approach and contribution and to improve their evaluation by including the relevant metrics and implementation details. """
91,"""The Expressive Power of Deep Neural Networks with Circulant Matrices""","['deep learning', 'circulant matrices', 'universal approximation']","""Recent results from linear algebra stating that any matrix can be decomposed into products of diagonal and circulant matrices has lead to the design of compact deep neural network architectures that perform well in practice. In this paper, we bridge the gap between these good empirical results and the theoretical approximation capabilities of Deep diagonal-circulant ReLU networks. More precisely, we first demonstrate that a Deep diagonal-circulant ReLU networks of bounded width and small depth can approximate a deep ReLU network in which the dense matrices are of low rank. Based on this result, we provide new bounds on the expressive power and universal approximativeness of this type of networks. We support our experimental results with thorough experiments on a large, real world video classification problem.""","""The paper conveys interesting study but the reviewers expressed concerns regarding the difference of this work compared to existing approaches and pointed a room for more thorough empirical evaluation."""
92,"""Sliced Wasserstein Auto-Encoders""","['optimal transport', 'Wasserstein distances', 'auto-encoders', 'unsupervised learning']","""In this paper we use the geometric properties of the optimal transport (OT) problem and the Wasserstein distances to define a prior distribution for the latent space of an auto-encoder. We introduce Sliced-Wasserstein Auto-Encoders (SWAE), that enable one to shape the distribution of the latent space into any samplable probability distribution without the need for training an adversarial network or having a likelihood function specified. In short, we regularize the auto-encoder loss with the sliced-Wasserstein distance between the distribution of the encoded training samples and a samplable prior distribution. We show that the proposed formulation has an efficient numerical solution that provides similar capabilities to Wasserstein Auto-Encoders (WAE) and Variational Auto-Encoders (VAE), while benefiting from an embarrassingly simple implementation. We provide extensive error analysis for our algorithm, and show its merits on three benchmark datasets.""","""The paper proposed to add the sliced-Wasserstein distance between the distribution of the encoded training samples and a samplable prior distribution to the auto encoder (AE) loss, resulting in a model named sliced-Wasserstein AE. The difference compared to the Wasserstein AE (WAE) lies in using the usage of the sliced-Wasserstein distance instead of GAN or MMD-based penalties. The idea of the paper is interesting, and a theoretical and an empirical analysis supporting the approach are presented. As reviewer 1 noticed, the advantage of using sliced Wasserstein distance is twofold: 1)parametric-free (compared to GANs); 2) almost hyperparameter-free (compared to the MMD with RBF kernels), except setting the number of random projection bases. However, the empirical evaluation in the paper and concurrent ICLR submission on Cramer-World-AEs the authors refer to shows no clear practical advantage over the WAE, which leads to better results at least regarding the FID score. On the other hand, the Cramer-World-AE is based on the ideas presented in this paper (which was previously available on arxive) proving that the paper presents interesting ideas which are of value to the communty. Therefore, the paper is a bit boarderline, but I recommand to accept it. """
93,"""Discrete Structural Planning for Generating Diverse Translations""","['machine translation', 'syntax', 'diversity', 'code learning']","""Planning is important for humans when producing complex languages, which is a missing part in current language generation models. In this work, we add a planning phase in neural machine translation to control the global sentence structure ahead of translation. Our approach learns discrete structural representations to encode syntactic information of target sentences. During translation, we can either let beam search to choose the structural codes automatically or specify the codes manually. The word generation is then conditioned on the selected discrete codes. Experiments show that the translation performance remains intact by learning the codes to capture pure structural variations. Through structural planning, we are able to control the global sentence structure by manipulating the codes. By evaluating with a proposed structural diversity metric, we found that the sentences sampled using different codes have much higher diversity scores. In qualitative analysis, we demonstrate that the sampled paraphrase translations have drastically different structures. ""","""This paper introduces a planning phase for NMT. It first generates a discrete set of tags at decoding time, and then the actual words are generated conditioned on those tags. The idea in the paper is interesting. However, the paper's experimental settings could improve by comparing on larger datasets and also using stronger baselines. The writing could also improve -- why were only the few coarse POS tags used? Have the authors tried a larger set? I think without such controlled comparisons, it would be hard to understand why only those coarse tags are used. The reviewers express concern about some of the above issues and there is consensus that the paper should be improved for acceptance at a venue like ICLR."""
94,"""DOMAIN ADAPTATION VIA DISTRIBUTION AND REPRESENTATION MATCHING: A CASE STUDY ON TRAINING DATA SELECTION VIA REINFORCEMENT LEARNING""","['domain adaptation', 'training data selection', 'reinforcement learning', 'natural language processing']","""Supervised models suffer from domain shifting where distribution mismatch across domains greatly affect model performance. Particularly, noise scattered in each domain has played a crucial role in representing such distribution, especially in various natural language processing (NLP) tasks. In addressing this issue, training data selection (TDS) has been proven to be a prospective way to train supervised models with higher performance and efficiency. Following the TDS methodology, in this paper, we propose a general data selection framework with representation learning and distribution matching simultaneously for domain adaptation on neural models. In doing so, we formulate TDS as a novel selection process based on a learned distribution from the input data, which is produced by a trainable selection distribution generator (SDG) that is optimized by reinforcement learning (RL). Then, the model trained by the selected data not only predicts the target domain data in a specific task, but also provides input for the value function of the RL. Experiments are conducted on three typical NLP tasks, namely, part-of-speech tagging, dependency parsing, and sentiment analysis. Results demonstrate the validity and effectiveness of our approach.""","""This paper investigates a data selection framework for domain adaptation based on reinforcement learning. Pros: The paper presents an approach that can dynamically adjust the data selection strategy via reinforcement learning. More specifically, the RL agent gets reward by selecting a new sample that makes the source training data distribution closer to the target distribution, where the distribution comparison is based on the feature representations that will be used by the prediction classifier. While the use of RL for data selection is not entirely new, the specific method proposed by the paper is reasonably novel and interesting. Cons: The use of RL is not clearly motivated and justified (R1,R3) and the method presented in this paper is rather hard to follow might be overly complex (R1). One fair point R1 raised is more clean-cut empirical evaluation that demonstrates how RL performs clearly better than greedy optimization. The authors came back with additional analysis in Section 4.2 to address this question, but R1 feels the new analysis (e.g., Fig 3) is not clear how to interpret. A more thorough ablation study of the proposed model might have addressed the reviewer's question more clearly. In addition, all reviewers felt that baselines are not convincingly strong enough, though each reviewer pointed out somewhat different aspects of baselines. R3 is most concerned about baselines being not state-of-the-art, and the rebuttal did not address R3's concern well enough. Verdict: Reject. A potentially interesting idea but 2/3 reviewers share strong concerns about the empirical results and overall clarity of the paper."""
95,"""On the Minimal Supervision for Training Any Binary Classifier from Only Unlabeled Data""","['learning from only unlabeled data', 'empirical risk minimization', 'unbiased risk estimator']","""Empirical risk minimization (ERM), with proper loss function and regularization, is the common practice of supervised classification. In this paper, we study training arbitrary (from linear to deep) binary classifier from only unlabeled (U) data by ERM. We prove that it is impossible to estimate the risk of an arbitrary binary classifier in an unbiased manner given a single set of U data, but it becomes possible given two sets of U data with different class priors. These two facts answer a fundamental question---what the minimal supervision is for training any binary classifier from only U data. Following these findings, we propose an ERM-based learning method from two sets of U data, and then prove it is consistent. Experiments demonstrate the proposed method could train deep models and outperform state-of-the-art methods for learning from two sets of U data.""","""This paper studies the task of learning a binary classifier from only unlabeled data. They first provide a negative result, i.e., they show it is impossible to learn an unbiased estimator from a set of unlabeled data. Then they provide an empirical risk minimization method which works when given two sets of unlabeled data, as well as the class priors. The four submitted reviews were unanimous in their vote to accept. The results are impactful, and might make for an interesting oral presentation."""
96,"""Adaptive Neural Trees""","['neural networks', 'decision trees', 'computer vision']","""Deep neural networks and decision trees operate on largely separate paradigms; typically, the former performs representation learning with pre-specified architectures, while the latter is characterised by learning hierarchies over pre-specified features with data-driven architectures. We unite the two via adaptive neural trees (ANTs), a model that incorporates representation learning into edges, routing functions and leaf nodes of a decision tree, along with a backpropagation-based training algorithm that adaptively grows the architecture from primitive modules (e.g., convolutional layers). ANTs allow increased interpretability via hierarchical clustering, e.g., learning meaningful class associations, such as separating natural vs. man-made objects. We demonstrate this on classification and regression tasks, achieving over 99% and 90% accuracy on the MNIST and CIFAR-10 datasets, and outperforming standard neural networks, random forests and gradient boosted trees on the SARCOS dataset. Furthermore, ANT optimisation naturally adapts the architecture to the size and complexity of the training data.""","""This paper proposes adaptive neural trees (ANT), a combination of deep networks and decision tress. Reviewers 1 leans toward reject the paper, pointing out several flaws. Reviewer 3 also raises concerns, despite later increasing rating to marginally above threshold. Of particular note is the weak experimental validation. The paper reports results only on MNIST and CIFAR-10. MNIST performance is too easily saturated to be meaningful. The CIFAR-10 results show ANT models to have far greater error than the state-of-the-art deep neural network models. As Reviewer 1 states, ""performance of the proposed method is also not the best on either of the tested datasets. Please clearly elaborate on why and how to address this issue. It would be more interesting and meaningful to work with a more recent large datasets, such as ImageNet or MS COCO."" The rebuttal fails to offer the type of additional results that would remedy this situation. Without a convincing experimental story, it is not possible to recommend acceptance of this paper."""
97,"""On Difficulties of Probability Distillation""","['Probability distillation', 'Autoregressive models', 'normalizing flows', 'wavenet', 'pixelcnn']","""Probability distillation has recently been of interest to deep learning practitioners as it presents a practical solution for sampling from autoregressive models for deployment in real-time applications. We identify a pathological optimization issue with the commonly adopted stochastic minimization of the (reverse) KL divergence, owing to sparse gradient signal from the teacher model due to curse of dimensionality. We also explore alternative principles for distillation, and show that one can achieve qualitatively better results than with KL minimization. ""","""The paper proposes new methods for optimization of optimization of KL(student_model||teacher_model). The topic is relevant. The paper also contains interesting ideas and the proposed methods are interesting; they are elegant and seems to work reasonably well on the tasks tried. However, the reviewers do not all agree that the paper is well written. The reviewers have pointed out several issues that need to be addresses before the paper can be accepted. """
98,"""Multilingual Neural Machine Translation With Soft Decoupled Encoding""",[],"""Multilingual training of neural machine translation (NMT) systems has led to impressive accuracy improvements on low-resource languages. However, there are still significant challenges in efficiently learning word representations in the face of paucity of data. In this paper, we propose Soft Decoupled Encoding (SDE), a multilingual lexicon encoding framework specifically designed to share lexical-level information intelligently without requiring heuristic preprocessing such as pre-segmenting the data. SDE represents a word by its spelling through a character encoding, and its semantic meaning through a latent embedding space shared by all languages. Experiments on a standard dataset of four low-resource languages show consistent improvements over strong multilingual NMT baselines, with gains of up to 2 BLEU on one of the tested languages, achieving the new state-of-the-art on all four language pairs.""","""although some may find the proposed approach as incremental over e.g. gu et al. (2018) and kiela et al. (2018), i believe the authors' clear motivation, formulation, experimentation and analysis are solid enough to warrant the presentation at the conference. the relative simplicity and successful empirical result show that the proposed approach could be one of the standard toolkits in deep learning for multilingual processing. J Gu, H Hassan, J Devlin, VOK Li. Universal Neural Machine Translation for Extremely Low Resource Languages. NAACL 2018. D Kiela, C Wang, K Cho. Context-Attentive Embeddings for Improved Sentence Representations. EMNLP 2018."""
99,"""Improving machine classification using human uncertainty measurements""","['image classification', 'human experiments', 'risk minimization']","""As deep CNN classifier performance using ground-truth labels has begun to asymptote at near-perfect levels, a key aim for the field is to extend training paradigms to capture further useful structure in natural image data and improve model robustness and generalization. In this paper, we present a novel natural image benchmark for making this extension, which we call CIFAR10H. This new dataset comprises a human-derived, full distribution over labels for each image of the CIFAR10 test set, offering the ability to assess the generalization of state-of-the-art CIFAR10 models, as well as investigate the effects of including this information in model training. We show that classification models trained on CIFAR10 do not generalize as well to our dataset as it does to traditional extensions, and that models fine-tuned using our label information are able to generalize better to related datasets, complement popular data augmentation schemes, and provide robustness to adversarial attacks. We explain these improvements in terms of better empirical approximations to the expected loss function over natural images and their categories in the visual world.""",""" The paper presents a new annotation of the CIFAR-10 dataset (the test set) as a distribution over labels as opposed to one-hot annotations. This datasets forms a testbed analysis for assessing the generalization abilities of the state-of-the-art models and their robustness to adversarial attacks. All the reviewers and AC acknowledge the contribution of dataset annotation and that the idea of using label distribution for training the models is sound and should improve the generalization performance of the models. However the reviewers and AC note the following potential weaknesses: (1) the paper requires major improvement in presentation clarity and in-depth investigation and evidence of the benefits of the proposed framework see detailed comments of R3 on what to address in a subsequent revision; see the suggestions of R2 for improving the scope of the empirical evaluations (e.g. distortions of the images, incorporating time limits for doing the classifications) and the requests of R1 for clarifications; (2) the related work is inadequate and should be substantially extended see the related references suggested by the R2; also R1 rightly pointed out that two out of four future extensions of this framework have been addressed already, which questions the significance of findings in this submission. The R2 raised concerns that the current evaluation is missing comparisons to a) the calibration approaches and b) cheaper/easier ways of getting soft labels -- see R2s suggestion to use the Brier score for model calibration and to use a cost matrix about how critical a misclassification is (cat <-> dog, versus cat <-> car) as soft labels. Among these, (2) did not have a substantial impact on the decision, but would be helpful to address in a subsequent revision. However, (1) and (3) makes it very difficult to assess the benefits of the proposed approach, and was viewed by the AC as a critical issue. There is no author response for this paper. The reviewer with a positive view on the manuscript (R3) was reluctant to champion the paper as the authors have not addressed the concerns of the other reviewers (no rebuttal). """
100,"""Functional Bayesian Neural Networks for Model Uncertainty Quantification""",[],"""In this paper, we extend the Bayesian neural network to functional Bayesian neural network with functional Monte Carlo methods that use the samples of functionals instead of samples of networks' parameters for inference to overcome the curse of dimensionality for uncertainty quantification. Based on the previous work on Riemannian Langevin dynamics, we propose the stochastic gradient functional Riemannian dynamics for training functional Bayesian neural network. We show the effectiveness and efficiency of our proposed approach with various experiments. ""","""This paper addresses a promising and challenging idea in Bayesian deep learning, namely thinking about distributions over functions rather than distributions over parameters. This is formulated by doing MCMC in a functional space rather than directly in the parameter space. The reviewers were unfortunately not convinced by the approach citing a variety of technical flaws, a lack of clarity of exposition and critical experiments. In general, it seems that the motivation of the paper is compelling and the idea promising, but perhaps the paper was hastily written before the ideas were fully developed and comprehensive experiments could be run. Hopefully the reviewer feedback will be helpful to further develop the work and lead to a future submission. Note: Unfortunately one review was too short to be informative. However, fortunately the other two reviews were sufficiently thorough to provide enough signal. """
101,"""Neural Regression Tree""","['regression-via-classification', 'discretization', 'regression tree', 'neural model', 'optimization']","""Regression-via-Classification (RvC) is the process of converting a regression problem to a classification one. Current approaches for RvC use ad-hoc discretization strategies and are suboptimal. We propose a neural regression tree model for RvC. In this model, we employ a joint optimization framework where we learn optimal discretization thresholds while simultaneously optimizing the features for each node in the tree. We empirically show the validity of our model by testing it on two challenging regression tasks where we establish the state of the art.""","""While the idea of revisiting regression-via-classification is interesting, the reviewers all agree that the paper lacks a proper motivating story for why this perspective is important. Furthermore, the baselines are weak, and there is additional relevant work that should be considered and discussed."""
102,"""Bias Also Matters: Bias Attribution for Deep Neural Network Explanation""","['explainable AI', 'interpreting deep neural networks', 'bias', 'attribution method', 'piecewise linear activation function', 'backpropagation']","""The gradient of a deep neural network (DNN) w.r.t. the input provides information that can be used to explain the output prediction in terms of the input features and has been widely studied to assist in interpreting DNNs. In a linear model (i.e., pseudo-formula ), the gradient corresponds solely to the weights pseudo-formula . Such a model can reasonably locally linearly approximate a smooth nonlinear DNN, and hence the weights of this local model are the gradient. The other part, however, of a local linear model, i.e., the bias pseudo-formula , is usually overlooked in attribution methods since it is not part of the gradient. In this paper, we observe that since the bias in a DNN also has a non-negligible contribution to the correctness of predictions, it can also play a significant role in understanding DNN behaviors. In particular, we study how to attribute a DNN's bias to its input features. We propose a backpropagation-type algorithm ``bias back-propagation (BBp)'' that starts at the output layer and iteratively attributes the bias of each layer to its input nodes as well as combining the resulting bias term of the previous layer. This process stops at the input layer, where summing up the attributions over all the input features exactly recovers pseudo-formula . Together with the backpropagation of the gradient generating pseudo-formula , we can fully recover the locally linear model pseudo-formula . Hence, the attribution of the DNN outputs to its inputs is decomposed into two parts, the gradient pseudo-formula and the bias attribution, providing separate and complementary explanations. We study several possible attribution methods applied to the bias of each layer in BBp. In experiments, we show that BBp can generate complementary and highly interpretable explanations of DNNs in addition to gradient-based attributions.""","""The work presents a method to back propagate and visualize bias distribution in network as a form of explainability of network decisions. Reviewers unanimous reject, no rebuttal from authors. """
103,"""Approximation and non-parametric estimation of ResNet-type convolutional neural networks via block-sparse fully-connected neural networks""","['CNN', 'ResNet', 'learning theory', 'approximation theory', 'non-parametric estimation', 'block-sparse']","""We develop new approximation and statistical learning theories of convolutional neural networks (CNNs) via the ResNet-type structure where the channel size, filter size, and width are fixed. It is shown that a ResNet-type CNN is a universal approximator and its expression ability is no worse than fully-connected neural networks (FNNs) with a \textit{block-sparse} structure even if the size of each layer in the CNN is fixed. Our result is general in the sense that we can automatically translate any approximation rate achieved by block-sparse FNNs into that by CNNs. Thanks to the general theory, it is shown that learning on CNNs satisfies optimality in approximation and estimation of several important function classes. As applications, we consider two types of function classes to be estimated: the Barron class and H\""older class. We prove the clipped empirical risk minimization (ERM) estimator can achieve the same rate as FNNs even the channel size, filter size, and width of CNNs are constant with respect to the sample size. This is minimax optimal (up to logarithmic factors) for the H\""older class. Our proof is based on sophisticated evaluations of the covering number of CNNs and the non-trivial parameter rescaling technique to control the Lipschitz constant of CNNs to be constructed.""","""The paper presents an interesting treatment of transforming a block-sparse fully connected neural networks to a ResNet-type Convolutional Network. Equipped with recent development on approximations of function classes (Barron, Holder) via block-sparse fully connected networks in the optimal rates, this enables us to show the equivalent power of ResNet Convolutional Nets. The major weakness in this treatment lies in that the ResNet architecture for realizing the block-sparse fully connected nets is unrealistic. It originates from the recent developments in approximation theory that transforming a fully connected net into a convolutional net via Toeplitz matrix (operator) factorizations. However the convolutional nets or ResNets obtained in this way is different to what have been used successfully in applications. Some special properties associated with convolutions, e.g. translation invariance and local deformation stability, are not natural in original fully connected nets and might be indirect after such a treatment. The presentation of the paper is better polished further. Based on ratings of reviewers, the current version of the paper is on borderline lean reject."""
104,"""Theoretical Analysis of Auto Rate-Tuning by Batch Normalization""","['batch normalization', 'scale invariance', 'learning rate', 'stationary point']","""Batch Normalization (BN) has become a cornerstone of deep learning across diverse architectures, appearing to help optimization as well as generalization. While the idea makes intuitive sense, theoretical analysis of its effectiveness has been lacking. Here theoretical support is provided for one of its conjectured properties, namely, the ability to allow gradient descent to succeed with less tuning of learning rates. It is shown that even if we fix the learning rate of scale-invariant parameters (e.g., weights of each layer with BN) to a constant (say, 0.3), gradient descent still approaches a stationary point (i.e., a solution where gradient is zero) in the rate of T^{1/2} in T iterations, asymptotically matching the best bound for gradient descent with well-tuned learning rates. A similar result with convergence rate T^{1/4} is also shown for stochastic gradient descent.""","""This paper conducted theoretical analysis of the effect of batch normalisation to auto rate-tuning. It provides an explanation for the empirical success of BN. The assumptions for the analysis is also closer to the common practice of batch normalization compared to a related work of Wu et al. 2018. One of the concerns raised by the reviewer is that the analysis does not immediately apply to practical uses of BN, but the authors already discussed how to fill the gap with a slight change of the activation function. Another concern is about the lack of empirical evaluation of the theory, and the authors provide additional experiments in the revision. R1 also points out a few weaknesses in the theoretical analysis, which I think would help improve the paper further if the authors could clarify and provide discussion in their revision. Overall, it is a good paper that will help improve our theoretical understanding about the power tool of batch normalization."""
105,"""EFFICIENT TWO-STEP ADVERSARIAL DEFENSE FOR DEEP NEURAL NETWORKS""","['Adversarial Examples', 'Adversarial Training', 'FGSM', 'IFGSM', 'Robustness']","""In recent years, deep neural networks have demonstrated outstanding performancein many machine learning tasks. However, researchers have discovered that thesestate-of-the-art models are vulnerable to adversarial examples: legitimate examples added by small perturbations which are unnoticeable to human eyes. Adversarial training, which augments the training data with adversarial examples duringthe training process, is a well known defense to improve the robustness of themodel against adversarial attacks. However, this robustness is only effective tothe same attack method used for adversarial training. Madry et al. (2017) suggest that effectiveness of iterative multi-step adversarial attacks and particularlythat projected gradient descent (PGD) may be considered the universal first order adversary and applying the adversarial training with PGD implies resistanceagainst many other first order attacks. However, the computational cost of theadversarial training with PGD and other multi-step adversarial examples is muchhigher than that of the adversarial training with other simpler attack techniques.In this paper, we show how strong adversarial examples can be generated only ata cost similar to that of two runs of the fast gradient sign method (FGSM), allowing defense against adversarial attacks with a robustness level comparable to thatof the adversarial training with multi-step adversarial examples. We empiricallydemonstrate the effectiveness of the proposed two-step defense approach againstdifferent attack methods and its improvements over existing defense strategies.""","""While the proposed method is novel, the evaluation is not convincing. In particular, the datasets and models used are small. Susceptibility to adversarial examples is tightly related to dimensionality. The study could benefit from more massive datasets (e.g., Imagenet)."""
106,"""Graph Classification with Geometric Scattering""","['geometric deep learning', 'graph neural network', 'graph classification', 'scattering']","""One of the most notable contributions of deep learning is the application of convolutional neural networks (ConvNets) to structured signal classification, and in particular image classification. Beyond their impressive performances in supervised learning, the structure of such networks inspired the development of deep filter banks referred to as scattering transforms. These transforms apply a cascade of wavelet transforms and complex modulus operators to extract features that are invariant to group operations and stable to deformations. Furthermore, ConvNets inspired recent advances in geometric deep learning, which aim to generalize these networks to graph data by applying notions from graph signal processing to learn deep graph filter cascades. We further advance these lines of research by proposing a geometric scattering transform using graph wavelets defined in terms of random walks on the graph. We demonstrate the utility of features extracted with this designed deep filter bank in graph classification of biochemistry and social network data (incl. state of the art results in the latter case), and in data exploration, where they enable inference of EC exchange preferences in enzyme evolution.""","""AR1 is concerned about the overlap of this paper with Gama et al., 2018 as well as lack of theoretical analysis and poor results on REDDIT-5k and REDDIT-5B datasets. AR2 reflects the same concerns (lack of clear cut novelty over Zou & Lerman, 2018, Game, 2018. AR3 also points the same issue re. lack of theoretical results. The austhors admit that Zou and Lerman, 2018, and Gama, 2018, focus on stability results while this submission offers empirical evaluations. Unfortunately, reviewers did not find these arguments convincing. Thus, at this point, the paper cannot be accepted for publication in ICLR. AC strongly encourages authors to develop their theoretical 'edge' over this crowded market of GCNs and scattering approaches."""
107,"""Neural Networks for Modeling Source Code Edits""","['Neural Networks', 'Program Synthesis', 'Source Code Modeling']","""Programming languages are emerging as a challenging and interesting domain for machine learning. A core task, which has received significant attention in recent years, is building generative models of source code. However, to our knowledge, previous generative models have always been framed in terms of generating static snapshots of code. In this work, we instead treat source code as a dynamic object and tackle the problem of modeling the edits that software developers make to source code files. This requires extracting intent from previous edits and leveraging it to generate subsequent edits. We develop several neural networks and use synthetic data to test their ability to learn challenging edit patterns that require strong generalization. We then collect and train our models on a large-scale dataset consisting of millions of fine-grained edits from thousands of Python developers.""","""This paper focuses on neural network models for source code edits. Compared to prior literature that focused on generative models of source codes, this paper focuses on the generative models of edit sequences of the source code. The paper explores both explicit and implicit representations of source code edits with experiments on synthetic and real code data. Pros: The task studied has a potential real world impact. The reviewers found the paper is generally clear to read. Cons: While the paper doesn't have a major flaw, the overall impact and novelty of the paper are considered to be relatively marginal. Even after the rebuttal, none of the reviewers felt compelled to increase their score. One point that came up multiple times is that the paper treats the source code as flat text and does not model the semantic and syntactic structure of the source code (via e.g., abstract syntax tree). While this alone would have not been a deal-breaker, the overall substance presented in the paper does not seem strong. Also, the empirical results are reasonable but not impressive given that the experiments are focused more on the synthetic data, and the experiments on the real source code are weaker and less clear as has been also noted by the fourth reviewer. Verdict: Possible weak reject. No significant deal breaker per say but the overall substance and novelty are marginal."""
108,"""REPRESENTATION COMPRESSION AND GENERALIZATION IN DEEP NEURAL NETWORKS""","['Deep neural network', 'information theory', 'training dynamics']","""Understanding the groundbreaking performance of Deep Neural Networks is one of the greatest challenges to the scientific community today. In this work, we introduce an information theoretic viewpoint on the behavior of deep networks optimization processes and their generalization abilities. By studying the Information Plane, the plane of the mutual information between the input variable and the desired label, for each hidden layer. Specifically, we show that the training of the network is characterized by a rapid increase in the mutual information (MI) between the layers and the target label, followed by a longer decrease in the MI between the layers and the input variable. Further, we explicitly show that these two fundamental information-theoretic quantities correspond to the generalization error of the network, as a result of introducing a new generalization bound that is exponential in the representation compression. The analysis focuses on typical patterns of large-scale problems. For this purpose, we introduce a novel analytic bound on the mutual information between consecutive layers in the network. An important consequence of our analysis is a super-linear boost in training time with the number of non-degenerate hidden layers, demonstrating the computational benefit of the hidden layers.""","""The authors admit the paper ""was not written carefully enough and requires major rewriting."" This seems to be a frustratingly common phenomenon with work on the information bottleneck. """
109,"""HR-TD: A Regularized TD Method to Avoid Over-Generalization""","['Reinforcement Learning', 'TD Learning', 'Deep Learning']","""Temporal Difference learning with function approximation has been widely used recently and has led to several successful results. However, compared with the original tabular-based methods, one major drawback of temporal difference learning with neural networks and other function approximators is that they tend to over-generalize across temporally successive states, resulting in slow convergence and even instability. In this work, we propose a novel TD learning method, Hadamard product Regularized TD (HR-TD), that reduces over-generalization and thus leads to faster convergence. This approach can be easily applied to both linear and nonlinear function approximators. HR-TD is evaluated on several linear and nonlinear benchmark domains, where we show improvement in learning behavior and performance.""","""All three reviewers raised the issues that (a) the problem tackled in the paper was insufficiently motivated, (b) the solution strategy was also not sufficiently motivated and (c) the experiments had serious methodological issues."""
110,"""Partially Mutual Exclusive Softmax for Positive and Unlabeled data""","['Negative Sampling', 'Sampled Softmax', 'Word embeddings', 'Adversarial Networks']","""In recent years, softmax together with its fast approximations has become the de-facto loss function for deep neural networks with multiclass predictions. However, softmax is used in many problems that do not fully fit the multiclass framework and where the softmax assumption of mutually exclusive outcomes can lead to biased results. This is often the case for applications such as language modeling, next event prediction and matrix factorization, where many of the potential outcomes are not mutually exclusive, but are more likely to be independent conditionally on the state. To this end, for the set of problems with positive and unlabeled data, we propose a relaxation of the original softmax formulation, where, given the observed state, each of the outcomes are conditionally independent but share a common set of negatives. Since we operate in a regime where explicit negatives are missing, we create an adversarially-trained model of negatives and derive a new negative sampling and weighting scheme which we denote as Cooperative Importance Sampling (CIS). We show empirically the advantages of our newly introduced negative sampling scheme by pluging it in the Word2Vec algorithm and benching it extensively against other negative sampling schemes on both language modeling and matrix factorization tasks and show large lifts in performance.""","""All reviewers agree that the paper is not quite ready for publication. """
111,"""Ergodic Measure Preserving Flows""","['Markov chain Monte Carlo', 'variational inference', 'deep generative models']","""Training probabilistic models with neural network components is intractable in most cases and requires to use approximations such as Markov chain Monte Carlo (MCMC), which is not scalable and requires significant hyper-parameter tuning, or mean-field variational inference (VI), which is biased. While there has been attempts at combining both approaches, the resulting methods have some important limitations in theory and in practice. As an alternative, we propose a novel method which is scalable, like mean-field VI, and, due to its theoretical foundation in ergodic theory, is also asymptotically accurate, like MCMC. We test our method on popular benchmark problems with deep generative models and Bayesian neural networks. Our results show that we can outperform existing approximate inference methods.""","""This paper proposes to a simple method for tuning parameters of HMC by maximizing the log density under the final sample of the MCMC, and apply it for training VAE. The reviews and discussion raises some critical concerns and questions, which unfortunately, which unfortunately, is not adequately addressed. """
112,"""SnapQuant: A Probabilistic and Nested Parameterization for Binary Networks""","['Binary weight networks', 'neural network quantization', 'reinforcement learning']","""In this paper, we study the problem of training real binary weight networks (without layer-wise or filter-wise scaling factors) from scratch under the Bayesian deep learning perspective, meaning that the final objective is to approximate the posterior distribution of binary weights rather than reach a point estimation. The proposed method, named as SnapQuant, has two intriguing features: (1) The posterior distribution is parameterized as a policy network trained with a reinforcement learning scheme. During the training phase, we generate binary weights on-the-fly since what we actually maintain is the policy network, and all the binary weights are used in a burn-after-reading style. At the testing phase, we can sample binary weight instances for a given recognition architecture from the learnt policy network. (2) The policy network, which has a nested parameter structure consisting of layer-wise, filter-wise and kernel-wise parameter sharing designs, is applicable to any neural network architecture. Such a nested parameterization explicitly and hierarchically models the joint posterior distribution of binary weights. The performance of SnapQuant is evaluated with several visual recognition tasks including ImageNet. The code will be made publicly available.""","""Reviewers mostly recommended to reject. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit. """
113,"""Learning Neural PDE Solvers with Convergence Guarantees""","['Partial differential equation', 'deep learning']","""Partial differential equations (PDEs) are widely used across the physical and computational sciences. Decades of research and engineering went into designing fast iterative solution methods. Existing solvers are general purpose, but may be sub-optimal for specific classes of problems. In contrast to existing hand-crafted solutions, we propose an approach to learn a fast iterative solver tailored to a specific domain. We achieve this goal by learning to modify the updates of an existing solver using a deep neural network. Crucially, our approach is proven to preserve strong correctness and convergence guarantees. After training on a single geometry, our model generalizes to a wide variety of geometries and boundary conditions, and achieves 2-3 times speedup compared to state-of-the-art solvers.""","""Quality: The overall quality of the work is high. The main idea and technical choices are well-motivated, and the method is about as simple as it could be while achieving its stated objectives. Clarity: The writing is clear, with the exception of using alternative scripts for some letters in definitions. Originality: The biggest weakness of this work is originality, in that there is a lot of closely related work, and similar ideas without convergence guarantees have begun to be explored. For example, the (very natural) U-net architecture was explored in previous work. Significance: This seems like an example of work that will be of interest both to the machine learning community, and also the numerics community, because it also achieves the properties that the numerics community has historically cared about. It is significant on its own as an improved method, but also as a demonstration that using deep learning doesn't require scrapping existing frameworks but can instead augment them."""
114,"""Perfect Match: A Simple Method for Learning Representations For Counterfactual Inference With Neural Networks""",[],"""Learning representations for counterfactual inference from observational data is of high practical relevance for many domains, such as healthcare, public policy and economics. Counterfactual inference enables one to answer ""What if...?"" questions, such as ""What would be the outcome if we gave this patient treatment pseudo-formula ?"". However, current methods for training neural networks for counterfactual inference on observational data are either overly complex, limited to settings with only two available treatment options, or both. Here, we present Perfect Match (PM), a method for training neural networks for counterfactual inference that is easy to implement, compatible with any architecture, does not add computational complexity or hyperparameters, and extends to any number of treatments. PM is based on the idea of augmenting samples within a minibatch with their propensity-matched nearest neighbours. Our experiments demonstrate that PM outperforms a number of more complex state-of-the-art methods in inferring counterfactual outcomes across several real-world and semi-synthetic datasets.""","""The reviewers found the paper to be well written, the work novel and they appreciated the breadth of the empirical evaluation. However, they did not seem entirely convinced that the improvements over the baseline are statistically significant. Reviewer 1 has lingering concerns about the experimental conditions and whether propensity-score matching within a minibatch would provide a substantial improvement over propensity-score matching across the dataset. Overall the reviewers found this to be a good paper and noted that the discussion was illuminating and demonstrated the merits of this work and interest to the community. However, no reviewers were prepared to champion the paper and thus it falls just below borderline for acceptance."""
115,"""Graph U-Net""","['graph', 'pooling', 'unpooling', 'U-Net']","""We consider the problem of representation learning for graph data. Convolutional neural networks can naturally operate on images, but have significant challenges in dealing with graph data. Given images are special cases of graphs with nodes lie on 2D lattices, graph embedding tasks have a natural correspondence with image pixel-wise prediction tasks such as segmentation. While encoder-decoder architectures like U-Net have been successfully applied on many image pixel-wise prediction tasks, similar methods are lacking for graph data. This is due to the fact that pooling and up-sampling operations are not natural on graph data. To address these challenges, we propose novel graph pooling (gPool) and unpooling (gUnpool) operations in this work. The gPool layer adaptively selects some nodes to form a smaller graph based on their scalar projection values on a trainable projection vector. We further propose the gUnpool layer as the inverse operation of the gPool layer. The gUnpool layer restores the graph into its original structure using the position information of nodes selected in the corresponding gPool layer. Based on our proposed gPool and gUnpool layers, we develop an encoder-decoder model on graph, known as the graph U-Net. Our experimental results on node classification tasks demonstrate that our methods achieve consistently better performance than previous models.""","""The authors supplied an updated paper resolving the most important reviewer concerns after the deadline for revisions. In part, this was due to reviewers requesting new experiments that take substantial time to complete. After discussion with the reviewers, I believe that if the revised manuscript had arrived earlier, then it should be accepted. Without the new results I would recommend rejecting since I believe the original submission lacked important experiments to justify the approach (inductive setting experiments are very useful). The community has an interest in uniform application of the rules surrounding the revision process. It is not fair to other authors to consider revisions past the deadline and we do not want to encourage late revisions. Better to submit a finished piece of work initially and not assume it will be possible to use up a lot of reviewer time and fix during the review process. We also don't want to encourage shoddy, rushed experimental work. However, the way we typically handle requests from reviewers that require a lot of work to complete is by rejecting papers and encouraging them to be resubmitted sometime in the future, typically to another similar conference. Thus I am recommending rejecting this paper on policy grounds, not on the merits of the latest draft. I believe that we should base the decision on the state of the paper at the same deadline that applies to all other authors. However, I am asking the program chairs to review this case since ultimately they will be the final arbiters of policy questions like this."""
116,"""Optimizing for Generalization in Machine Learning with Cross-Validation Gradients""",[],"""Cross-validation is the workhorse of modern applied statistics and machine learning, as it provides a principled framework for selecting the model that maximizes generalization performance. In this paper, we show that the cross-validation risk is differentiable with respect to the hyperparameters and training data for many common machine learning algorithms, including logistic regression, elastic-net regression, and support vector machines. Leveraging this property of differentiability, we propose a cross-validation gradient method (CVGM) for hyperparameter optimization. Our method enables efficient optimization in high-dimensional hyperparameter spaces of the cross-validation risk, the best surrogate of the true generalization ability of our learning algorithm.""","""This paper gives explicit hyperparameter gradients for several models with convex losses. The idea is well-motivated and clearly presented, but because it's relatively incremental, it needs a more systematic experimental section, or at least a stronger characterization of its scope and limitations. I would also recommend an investigation of more expressive hyperparameterizations (like in Maclaurin et al 2015) and/or an investigation of overfitting on the validation set."""
117,"""Gradient-based learning for F-measure and other performance metrics""",[],"""Many important classification performance metrics, e.g. pseudo-formula -measure, are non-differentiable and non-decomposable, and are thus unfriendly to gradient descent algorithm. Consequently, despite their popularity as evaluation metrics, these metrics are rarely optimized as training objectives in neural network community. In this paper, we propose an empirical utility maximization scheme with provable learning guarantees to address the non-differentiability of these metrics. We then derive a strongly consistent gradient estimator to handle non-decomposability. These innovations enable end-to-end optimization of these metrics with the same computational complexity as optimizing a decomposable and differentiable metric, e.g. cross-entropy loss.""","""This manuscript proposes a gradient-based learning scheme for non-differentiable and non-decomposable metrics. The key idea is to optimize a soft predictor directly (instead of aiming for a deterministic predictor), which results in a differentiable loss for many of these metrics. Theoretical results are provided which describe the performance of this approach. The reviewers and ACs noted weakness in the original submission related to the clarity of the presentation and novelty as related to already published work. There was also a concern about the usefulness the main theoretical results due to asymptotic assumptions. The manuscript would be significantly strengthened if the reliance on infinite sample sizes is resolved, or sufficient empirical evidence is provided which suggests that the asymptotic issues are not practically significant."""
118,"""Policy Optimization via Stochastic Recursive Gradient Algorithm""","['reinforcement learning', 'policy gradient', 'variance reduction', 'stochastic recursive gradient algorithm']","""In this paper, we propose the StochAstic Recursive grAdient Policy Optimization (SARAPO) algorithm which is a novel variance reduction method on Trust Region Policy Optimization (TRPO). The algorithm incorporates the StochAstic Recursive grAdient algoritHm(SARAH) into the TRPO framework. Compared with the existing Stochastic Variance Reduced Policy Optimization (SVRPO), our algorithm is more stable in the variance. Furthermore, by theoretical analysis the ordinary differential equation and the stochastic differential equation (ODE/SDE) of SARAH, we analyze its convergence property and stability. Our experiments demonstrate its performance on a variety of benchmark tasks. We show that our algorithm gets better improvement in each iteration and matches or even outperforms SVRPO and TRPO. ""","""The use of SARAH for Policy optimization in RL is novel, with some theoretical analysis to demonstrate convergence of this approach. However, concerns were raised in terms of clarity of the paper, empirical results and in placement of this theory relative to a previous variance reduction algorithm called SVRPG. The author response similarly did not explain the novelty of the theory beyond the convergence results of what was given by the paper on SVRPG. By incorporating some of the reviewer comments, this paper could be a meaningful and useful contribution."""
119,"""SHE2: Stochastic Hamiltonian Exploration and Exploitation for Derivative-Free Optimization""",['derivative-free optimization'],"""Derivative-free optimization (DFO) using trust region methods is frequently used for machine learning applications, such as (hyper-)parameter optimization without the derivatives of objective functions known. Inspired by the recent work in continuous-time minimizers, our work models the common trust region methods with the exploration-exploitation using a dynamical system coupling a pair of dynamical processes. While the first exploration process searches the minimum of the blackbox function through minimizing a time-evolving surrogation function, another exploitation process updates the surrogation function time-to-time using the points traversed by the exploration process. The efficiency of derivative-free optimization thus depends on ways the two processes couple. In this paper, we propose a novel dynamical system, namely \ThePrev---\underline{S}tochastic \underline{H}amiltonian \underline{E}xploration and \underline{E}xploitation, that surrogates the subregions of blackbox function using a time-evolving quadratic function, then explores and tracks the minimum of the quadratic functions using a fast-converging Hamiltonian system. The \ThePrev\ algorithm is later provided as a discrete-time numerical approximation to the system. To further accelerate optimization, we present \TheName\ that parallelizes multiple \ThePrev\ threads for concurrent exploration and exploitation. Experiment results based on a wide range of machine learning applications show that \TheName\ outperform a boarder range of derivative-free optimization algorithms with faster convergence speed under the same settings.""","""In general the reviewers found the work to be interesting and the results to be promising. However, all the reviewers shared significant concerns about the clarity of the paper and the correctness of technical claims made. This paper would significantly benefit from rewriting and restructuring the paper to improve clarity, better motivate the approach and provide more careful exposition of related work and technical claims."""
120,"""How Powerful are Graph Neural Networks?""","['graph neural networks', 'theory', 'deep learning', 'representational power', 'graph isomorphism', 'deep multisets']","""Graph Neural Networks (GNNs) are an effective framework for representation learning of graphs. GNNs follow a neighborhood aggregation scheme, where the representation vector of a node is computed by recursively aggregating and transforming representation vectors of its neighboring nodes. Many GNN variants have been proposed and have achieved state-of-the-art results on both node and graph classification tasks. However, despite GNNs revolutionizing graph representation learning, there is limited understanding of their representational properties and limitations. Here, we present a theoretical framework for analyzing the expressive power of GNNs to capture different graph structures. Our results characterize the discriminative power of popular GNN variants, such as Graph Convolutional Networks and GraphSAGE, and show that they cannot learn to distinguish certain simple graph structures. We then develop a simple architecture that is provably the most expressive among the class of GNNs and is as powerful as the Weisfeiler-Lehman graph isomorphism test. We empirically validate our theoretical findings on a number of graph classification benchmarks, and demonstrate that our model achieves state-of-the-art performance.""","""Graph neural networks are an increasingly popular topic of research in machine learning, and this paper does a good job of studying the representational power of some newly proposed variants. The framing of the problem in terms of the WL test, and the proposal of the GIN architecture is a valuable contribution. Through the reviews and subsequent discussion, it looks like the issues surrounding Theorem 3 have been resolved, and therefore all of the reviewers now agree that this paper should be accepted. There may be some interesting followup work based on studying depth, as pointed out by reviewer 1, but this may not be an issue in GIN and is regardless a topic for future research."""
121,"""Optimized Gated Deep Learning Architectures for Sensor Fusion""","['deep learning', 'convolutional neural network', 'sensor fusion', 'activity recognition']","""Sensor fusion is a key technology that integrates various sensory inputs to allow for robust decision making in many applications such as autonomous driving and robot control. Deep neural networks have been adopted for sensor fusion in a body of recent studies. Among these, the so-called netgated architecture was proposed, which has demonstrated improved performances over the conventional convolu- tional neural networks (CNN). In this paper, we address several limitations of the baseline negated architecture by proposing two further optimized architectures: a coarser-grained gated architecture employing (feature) group-level fusion weights and a two-stage gated architectures leveraging both the group-level and feature- level fusion weights. Using driving mode prediction and human activity recogni- tion datasets, we demonstrate the significant performance improvements brought by the proposed gated architectures and also their robustness in the presence of sensor noise and failures. ""","""The paper builds up on the gated fusion network architectures, and adapt those approaches to reach improved results. In that it is incrementally worthwhile. All the same, all reviewers agree that the work is not yet up to par. In particular, the paper is only incremental, and the novelty of it is not clear. It does not relate well to existing work in this field, and the results are not rigorously evaluated; thus its merit is unclear experimentally. """
122,"""Directed-Info GAIL: Learning Hierarchical Policies from Unsegmented Demonstrations using Directed Information""","['Imitation Learning', 'Reinforcement Learning', 'Deep Learning']","""The use of imitation learning to learn a single policy for a complex task that has multiple modes or hierarchical structure can be challenging. In fact, previous work has shown that when the modes are known, learning separate policies for each mode or sub-task can greatly improve the performance of imitation learning. In this work, we discover the interaction between sub-tasks from their resulting state-action trajectory sequences using a directed graphical model. We propose a new algorithm based on the generative adversarial imitation learning framework which automatically learns sub-task policies from unsegmented demonstrations. Our approach maximizes the directed information flow in the graphical model between sub-task latent variables and their generated trajectories. We also show how our approach connects with the existing Options framework, which is commonly used to learn hierarchical policies.""","""This paper proposes an approach for imitation learning from unsegmented demonstrations. The paper addresses an important problem and is well-motivated. Many of the concerns about the experiments have been addressed with follow-up comments. We strongly encourage the authors to integrate the new results and additional literature to the final version. With these changes, the reviewers agree that the paper exceeds the bar for acceptance. Thus, I recommend acceptance."""
123,"""Classification from Positive, Unlabeled and Biased Negative Data""","['positive-unlabeled learning', 'dataset shift', 'empirical risk minimization']","""Positive-unlabeled (PU) learning addresses the problem of learning a binary classifier from positive (P) and unlabeled (U) data. It is often applied to situations where negative (N) data are difficult to be fully labeled. However, collecting a non-representative N set that contains only a small portion of all possible N data can be much easier in many practical situations. This paper studies a novel classification framework which incorporates such biased N (bN) data in PU learning. The fact that the training N data are biased also makes our work very different from those of standard semi-supervised learning. We provide an empirical risk minimization-based method to address this PUbN classification problem. Our approach can be regarded as a variant of traditional example-reweighting algorithms, with the weight of each example computed through a preliminary step that draws inspiration from PU learning. We also derive an estimation error bound for the proposed method. Experimental results demonstrate the effectiveness of our algorithm in not only PUbN learning scenarios but also ordinary PU leaning scenarios on several benchmark datasets.""","""The paper proposes an algorithm for semi-supervised learning, which incorporate biased negative data into the existing PU learning framework. The reviewers and AC commonly note the critical limitation of practical value of the paper and results are rather straightforward. AC decided the paper might not be ready to publish as other contributions are not enough to compensate the issue."""
124,"""Where and when to look? Spatial-temporal attention for action recognition in videos""","['visual attention', 'video action recognition', 'network interpretability']","""Inspired by the observation that humans are able to process videos efficiently by only paying attention when and where it is needed, we propose a novel spatial-temporal attention mechanism for video-based action recognition. For spatial attention, we learn a saliency mask to allow the model to focus on the most salient parts of the feature maps. For temporal attention, we employ a soft temporal attention mechanism to identify the most relevant frames from an input video. Further, we propose a set of regularizers that ensure that our attention mechanism attends to coherent regions in space and time. Our model is efficient, as it proposes a separable spatio-temporal mechanism for video attention, while being able to identify important parts of the video both spatially and temporally. We demonstrate the efficacy of our approach on three public video action recognition datasets. The proposed approach leads to state-of-the-art performance on all of them, including the new large-scale Moments in Time dataset. Furthermore, we quantitatively and qualitatively evaluate our model's ability to accurately localize discriminative regions spatially and critical frames temporally. This is despite our model only being trained with per video classification labels. ""","""Strengths: The paper presentation was assessed as being of high quality. Experiments were diverse in terms of datasets and tasks. Weaknesses: Multiple reviewers commented that the paper does not present substantial novelty compared to previous work. Contention: One reviewer holding out on giving a stronger rating to the paper due to the issue of novelty. Consensus: Final scores were two 6s one 3. This work has merit, but the degree of concern over the level of novelty leads to an aggregate rating that is too low to justify acceptance. Authors are encourage to re-submit to another venue. """
125,"""Learning to Describe Scenes with Programs""","['Structured scene representations', 'program synthesis']","""Human scene perception goes beyond recognizing a collection of objects and their pairwise relations. We understand higher-level, abstract regularities within the scene such as symmetry and repetition. Current vision recognition modules and scene representations fall short in this dimension. In this paper, we present scene programs, representing a scene via a symbolic program for its objects, attributes, and their relations. We also propose a model that infers such scene programs by exploiting a hierarchical, object-based scene representation. Experiments demonstrate that our model works well on synthetic data and transfers to real images with such compositional structure. The use of scene programs has enabled a number of applications, such as complex visual analogy-making and scene extrapolation.""","""This paper presents a dataset and method for training a model to infer, from a visual scene, the program that would generate/describe it. In doing so, it produces abstract disentangled representations of the scene which could be used by agents, models, and other ML methods to reason about the scene. This is yet another paper where the reviewers disappointingly did not interact. The first round of reviews were mediocre-to-acceptable. The authors, I think, did a good job of responding to the concerns raised by the reviewers and edited their paper accordingly. Unfortunately, not one of the reviewers took the time to consider author responses. In light of my reading of the responses and the revisions in the paper, I am leaning towards treating this as a paper where the review process has failed the authors, and recommending acceptance. The paper presents a novel method and dataset, and the experiments are reasonably convincing. The paper has flaws and the authors are advised to carefully take into account the concerns flagged by reviewersmany of which they have responded toin producing their final manuscript."""
126,"""LEARNING GENERATIVE MODELS FOR DEMIXING OF STRUCTURED SIGNALS FROM THEIR SUPERPOSITION USING GANS""","['Generative Models', 'GANs', 'Denosing', 'Demixing', 'Structured Recovery']","""Recently, Generative Adversarial Networks (GANs) have emerged as a popular alternative for modeling complex high dimensional distributions. Most of the existing works implicitly assume that the clean samples from the target distribution are easily available. However, in many applications, this assumption is violated. In this paper, we consider the problem of learning GANs under the observation setting when the samples from target distribution are given by the superposition of two structured components. We propose two novel frameworks: denoising-GAN and demixing-GAN. The denoising-GAN assumes access to clean samples from the second component and try to learn the other distribution, whereas demixing-GAN learns the distribution of the components at the same time. Through comprehensive numerical experiments, we demonstrate that proposed frameworks can generate clean samples from unknown distributions, and provide competitive performance in tasks such as denoising, demixing, and compressive sensing.""","""The paper proposes two simple generator architecture variants enabling the use of GAN training for the tasks of denoising (from known noise types) and demixing (of two added sources). While the denoising approach is very similar to AmbientGAN and could thus be considered somewhat incremental, all reviewers and the AC agree that the developed use of GANs for demixing is an interesting novel direction. The paper is well written, and the approach is supported by encouraging experimental results on MNIST and Fashion-MNIST. Reviewers and AC noted the following weaknesses of the paper: a) no theoretical support or analysis is provided for the approach, this makes it primarily an empirical study of a nice idea. b) For an empirical study, the experimental evaluation is very limited, both in terms of dataset/problems it is tested on; and in terms of algorithms for demixing/source-separation that it is compared against. Following these reviews, the authors added the experiments on Fashion-MNIST and comparison with ICA which are steps in the right direction. This improvement moved one reviewer to positively update his score, but not the others. Taking everything into account, the AC judges that it is a very promising direction, but that more extensive experiments on additional benchmark tasks for demixing and comparison with other demixing algorithms are needed to make this work a more complete contribution. """
127,"""What Would pi* Do?: Imitation Learning via Off-Policy Reinforcement Learning""","['imitation learning', 'reinforcement learning']","""Learning to imitate expert actions given demonstrations containing image observations is a difficult problem in robotic control. The key challenge is generalizing behavior to out-of-distribution states that differ from those in the demonstrations. State-of-the-art imitation learning algorithms perform well in environments with low-dimensional observations, but typically involve adversarial optimization procedures, which can be difficult to use with high-dimensional image observations. We propose a remarkably simple alternative based on off-policy soft Q-learning, which we call soft Q imitation learning (SQIL, pronounced ""skill""), that rewards the agent for matching demonstrated actions in demonstrated states. The key idea is initially filling the agent's experience replay buffer with demonstrations, where rewards are set to a positive constant, and setting rewards to zero in all additional experiences. We derive SQIL from first principles as a method for performing approximate inference under the MaxCausalEnt model of expert behavior. The approximate inference objective trades off between a pure behavioral cloning loss and a regularization term that incorporates information about state transitions via the soft Bellman error. Our experiments show that SQIL matches the state of the art in low-dimensional environments, and significantly outperforms prior work in playing video games from high-dimensional images.""",""" This is an interesting direction and all reviewers thought the idea has merit, but pointed out some significant limitations. The authors did an admirable job at addressing some of these but some remain, including R1s point 2 which is a significant issue. The authors are encouraged to submit a revised version of their work which addresses all the discussed limitations and will likely be a competitive submission to another top ML conference. """
128,"""Efficient Codebook and Factorization for Second Order Representation Learning""",['Second order pooling'],"""Learning rich and compact representations is an open topic in many fields such as word embedding, visual question-answering, object recognition or image retrieval. Although deep neural networks (convolutional or not) have made a major breakthrough during the last few years by providing hierarchical, semantic and abstract representations for all of these tasks, these representations are not necessary as rich as needed nor as compact as expected. Models using higher order statistics, such as bilinear pooling, provide richer representations at the cost of higher dimensional features. Factorization schemes have been proposed but without being able to reach the original compactness of first order models, or at a heavy loss in performances. This paper addresses these two points by extending factorization schemes to codebook strategies, allowing compact representations with the same dimensionality as first order representations, but with second order performances. Moreover, we extend this framework with a joint codebook and factorization scheme, granting a reduction both in terms of parameters and computation cost. This formulation leads to state-of-the-art results and compact second-order models with few additional parameters and intermediate representations with a dimension similar to that of first-order statistics.""","""AR1 is is concerned that the only contribution of this work is combining second-order pooling with with a codebook style assignments. After discussions, AR1 still maintains that that the proposed factorization is a marginal contribution. AR2 feels that the proposed paper is highly related to numerous current works (e.g. mostly a mixture of existing contributions) and that evaluations have not been improved. AR3 also points that this paper lacks important comparisons for fairly evaluating the effectiveness of the proposed formulation and it lacks detailed description and discussion for the methods. AC has also pointed several works to the authors which are highly related (but by no means this is not an exhaustive list and authors need to explore google scholar to retrieve more relevant papers than the listed ones): [1] MoNet: Moments Embedding Network by Gou et al. (e.g. Stanford Cars via Tensor Sketching: 90.8 vs. 90.4 in this submission, Airplane: 88.1 vs. 87.3% in this submission, 85.7 vs. 84.3% in this submission) [2] Second-order Democratic Aggregation by Lin et al. (e.g. Stanford Cars: 90.8 vs. 90.4 in this submission) [3] Statistically-motivated Second-order Pooling by Yu et al (CUB: 85%) [4] DeepKSPD: Learning Kernel-matrix-based SPD Representation for Fine-grained Image Recognition by Engin et al. [5] Global Gated Mixture of Second-order Pooling for Improving Deep Convolutional Neural Networks by Q. Wang et al. (512D representations) [6] Low-rank Bilinear Pooling for Fine-Grained Classification' by S. Kong et al. (CVPR I believe). They get some reduction of size of 10x less than tensor sketch, higher results than here by some >2% (CUB), and all this obtained in somewhat more sophisticated way. The authors brushed under the carpet some comparisons. Some methods above are simply better performing even if cited, e.g. MoNet [1] uses sketching and seems a better performer on several datasets, see [2] that uses sketching (Section 4.4), see [5] which also generates compact representation (8K). [4] may be not compact but the whole point is to compare compact methods with non-compact second-order ones too (e.g. small performance loss for compact methods is OK but big loss warrants a question whether they are still useful). Approach [6] seems to also obtain better results on some sets (common testbed comparisons are essentially encouraged). At this point, AC will also point authors to sparse coding methods on matrices (bilinear) and tensors (higher-order) from years 2013-2018 (TPAMI, CVPR, ECCV, ICCV, etc.). These all methods can produce compact representations (512 to 10K or so) of bilinear or higher-order descriptors for classification. This manuscript fails to mention this family of methods. For a paper to be improved for the future, the authors should consider the following: - make a thorough comparison with existing second-order/bilinear methods in the common testbed (most of the codes are out there on-line) - the authors should vary the size of representation (from 512 to 8K or more) and plot this against accuracy - the authors should provide theoretical discussion and guarantees on the quality of their low-rank approximations (e.g. sketching has clear bounds on its approximation quality, rates, computational cost). The authors should provide some bounds on the loss of information in the proposed method. - authors should discuss the theoretical complexity of proposed method (and other methods in the literature) Additionally, the authors should improve their references and the story line. Citing (Lin et al. (2015)) in Eq. 1 and 2 as if they are the father of bilinear pooling is misleading. Citing (Gao et al. (2016)) in the context of polynomial kernel approximation in Eq. 3 to obtain bilinear pooling should be preceded with earlier works that expand polynomial kernel to obtain bilieanr pooling. AC can think of at least two papers from 2012/2013 which do derive bilinear pooling and could be cited here instead. AC encourages the authors to revise their references and story behind bilinear pooling to give unsuspected readers a full/honest story of bilinear representations and compact methods (whether they are branded as compact or just use sketching etc., whether they use dictionaries or low-rank representations). In conclusion, it feels this manuscript is not ready for publication with ICLR and requires a major revision. However, there is some merit in the proposed direction and authors are encouraged to explore further."""
129,"""Penetrating the Fog: the Path to Efficient CNN Models""","['Efficient CNN models', 'Computer Vision']","""With the increasing demand to deploy convolutional neural networks (CNNs) on mobile platforms, the sparse kernel approach was proposed, which could save more parameters than the standard convolution while maintaining accuracy. However, despite the great potential, no prior research has pointed out how to craft an sparse kernel design with such potential (i.e., effective design), and all prior works just adopt simple combinations of existing sparse kernels such as group convolution. Meanwhile due to the large design space it is also impossible to try all combinations of existing sparse kernels. In this paper, we are the first in the field to consider how to craft an effective sparse kernel design by eliminating the large design space. Specifically, we present a sparse kernel scheme to illustrate how to reduce the space from three aspects. First, in terms of composition we remove designs composed of repeated layers. Second, to remove designs with large accuracy degradation, we find an unified property named~\emph{information field} behind various sparse kernel designs, which could directly indicate the final accuracy. Last, we remove designs in two cases where a better parameter efficiency could be achieved. Additionally, we provide detailed efficiency analysis on the final 4 designs in our scheme. Experimental results validate the idea of our scheme by showing that our scheme is able to find designs which are more efficient in using parameters and computation with similar or higher accuracy.""","""This paper points out methods to obtain sparse convolutional operators. The reviewers have a consensus on rejection due to clarity and lack of support to the claims."""
130,"""COLLABORATIVE MULTIAGENT REINFORCEMENT LEARNING IN HOMOGENEOUS SWARMS""","['Reinforcement Learning', 'Multi Agent', 'policy gradient']","""A deep reinforcement learning solution is developed for a collaborative multiagent system. Individual agents choose actions in response to the state of the environment, their own state, and possibly partial information about the state of other agents. Actions are chosen to maximize a collaborative long term discounted reward that encompasses the individual rewards collected by each agent. The paper focuses on developing a scalable approach that applies to large swarms of homogeneous agents. This is accomplished by forcing the policies of all agents to be the same resulting in a constrained formulation in which the experiences of each agent inform the learning process of the whole team, thereby enhancing the sample efficiency of the learning process. A projected coordinate policy gradient descent algorithm is derived to solve the constrained reinforcement learning problem. Experimental evaluations in collaborative navigation, a multi-predator-multi-prey game, and a multiagent survival game show marked improvements relative to methods that do not exploit the policy equivalence that naturally arises in homogeneous swarms.""","""Pros: - interesting novel formulation of policy learning in homogeneous swarms - multi-stage learning process that trades off diversity and consistency (fig 1) Cons: - implausible mechanisms like averaging weights of multiple networks - minor novelty - missing ablations of which aspect is crucial - dubious baseline results - no rebuttal One reviewer out of three would have accepted the paper, the other two have major concerns. Unfortunately the authors did not revise the paper or engage with the reviewers to clear up these points, so as it stand the paper should be rejected."""
131,"""Analysis of Quantized Models""","['weight quantization', 'gradient quantization', 'distributed learning']","""Deep neural networks are usually huge, which significantly limits the deployment on low-end devices. In recent years, many weight-quantized models have been proposed. They have small storage and fast inference, but training can still be time-consuming. This can be improved with distributed learning. To reduce the high communication cost due to worker-server synchronization, recently gradient quantization has also been proposed to train deep networks with full-precision weights. In this paper, we theoretically study how the combination of both weight and gradient quantization affects convergence. We show that (i) weight-quantized models converge to an error related to the weight quantization resolution and weight dimension; (ii) quantizing gradients slows convergence by a factor related to the gradient quantization resolution and dimension; and (iii) clipping the gradient before quantization renders this factor dimension-free, thus allowing the use of fewer bits for gradient quantization. Empirical experiments confirm the theoretical convergence results, and demonstrate that quantized networks can speed up training and have comparable performance as full-precision networks.""","""This paper provides the first convergence analysis for convex model distributed training with quantized weights and gradients. It is well written and organized. Extensive experiments are carried out beyond the assumption of convex models in the theoretical study. Analysis with weight and gradient quantization has been separately studied, and this paper provides a combined analysis, which renders the contribution incremental. As pointed out by R2 and R3, it is somewhat unclear under which problem setting, the proposed quantized training would help improve the convergence. The authors provide clarification in the feedback. It is important to include those, together with other explanations in the feedback, in the future revision. Another limitation pointed out by R3 is that the theoretical analysis applies to convex models only. Nevertheless, it is nice to show in experiments that deep networks training is benefitted from the gradient quantization empirically."""
132,"""K for the Price of 1: Parameter-efficient Multi-task and Transfer Learning""","['deep learning', 'mobile', 'transfer learning', 'multi-task learning', 'computer vision', 'small models', 'imagenet', 'inception', 'batch normalization']","""We introduce a novel method that enables parameter-efficient transfer and multi-task learning with deep neural networks. The basic approach is to learn a model patch - a small set of parameters - that will specialize to each task, instead of fine-tuning the last layer or the entire network. For instance, we show that learning a set of scales and biases is sufficient to convert a pretrained network to perform well on qualitatively different problems (e.g. converting a Single Shot MultiBox Detection (SSD) model into a 1000-class image classification model while reusing 98% of parameters of the SSD feature extractor). Similarly, we show that re-learning existing low-parameter layers (such as depth-wise convolutions) while keeping the rest of the network frozen also improves transfer-learning accuracy significantly. Our approach allows both simultaneous (multi-task) as well as sequential transfer learning. In several multi-task learning problems, despite using much fewer parameters than traditional logits-only fine-tuning, we match single-task performance. ""","""Reviewers largely agree that the proposed method for finetuning the deep neural networks is interesting and empirical results clearly show the benefits over finetuning only the last layer. I recommend acceptance. """
133,"""Successor Options : An Option Discovery Algorithm for Reinforcement Learning""",['Hierarchical Reinforcement Learning'],"""Hierarchical Reinforcement Learning is a popular method to exploit temporal abstractions in order to tackle the curse of dimensionality. The options framework is one such hierarchical framework that models the notion of skills or options. However, learning a collection of task-agnostic transferable skills is a challenging task. Option discovery typically entails using heuristics, the majority of which revolve around discovering bottleneck states. In this work, we adopt a method complementary to the idea of discovering bottlenecks. Instead, we attempt to discover ``landmark"" sub-goals which are prototypical states of well connected regions. These sub-goals are points from which densely connected set of states are easily accessible. We propose a new model called Successor options that leverages Successor Representations to achieve the same. We also design a novel pseudo-reward for learning the intra-option policies. Additionally, we describe an Incremental Successor options model that iteratively builds options and explores in environments where exploration through primitive actions is inadequate to form the Successor Representations. Finally, we demonstrate the efficacy of our approach on a collection of grid worlds and on complex high dimensional environments like Deepmind-Lab. ""","""Pros: - simple, sensible subgoal discovery method - strong inuitions, visualizations - detailed rebuttal, 15 appendix sections Cons: - moderate novelty - lack of ablations - assessments don't back up all claims - ill-justified/mismatching design decisions - inefficiency due to relying on a random policy in the first phase There is consensus among the reviewers that the paper is not quite good enough, and should be (borderline) rejected."""
134,"""Learning Discriminators as Energy Networks in Adversarial Learning""","['adversarial learning', 'structured prediction', 'energy networks']","""We propose a novel adversarial learning framework in this work. Existing adversarial learning methods involve two separate networks, i.e., the structured prediction models and the discriminative models, in the training. The information captured by discriminative models complements that in the structured prediction models, but few existing researches have studied on utilizing such information to improve structured prediction models at the inference stage. In this work, we propose to refine the predictions of structured prediction models by effectively integrating discriminative models into the prediction. Discriminative models are treated as energy-based models. Similar to the adversarial learning, discriminative models are trained to estimate scores which measure the quality of predicted outputs, while structured prediction models are trained to predict contrastive outputs with maximal energy scores. In this way, the gradient vanishing problem is ameliorated, and thus we are able to perform inference by following the ascent gradient directions of discriminative models to refine structured prediction models. The proposed method is able to handle a range of tasks, \emph{e.g.}, multi-label classification and image segmentation. Empirical results on these two tasks validate the effectiveness of our learning method.""","""All three reviewers expressed concerns about the writing of the paper. The AC thus recommends ""revise and resubmit""."""
135,"""Complexity of Training ReLU Neural Networks""","['NP-hardness', 'ReLU activation', 'Two hidden layer networks']","""In this paper, we explore some basic questions on complexity of training Neural networks with ReLU activation function. We show that it is NP-hard to train a two-hidden layer feedforward ReLU neural network. If dimension d of the data is fixed then we show that there exists a polynomial time algorithm for the same training problem. We also show that if sufficient over-parameterization is provided in the first hidden layer of ReLU neural network then there is a polynomial time algorithm which finds weights such that output of the over-parameterized ReLU neural network matches with the output of the given data.""","""Dear authors, All reviewers agreed that, while the problem considered was of interest, the theoretical result presented in this work was of too limited scope to be of interest for the ICLR audience. Based on their comments, you might want to consider a more theoretically-oriented venue for such a submission."""
136,"""What do you learn from context? Probing for sentence structure in contextualized word representations""","['natural language processing', 'word embeddings', 'transfer learning', 'interpretability']","""Contextualized representation models such as ELMo (Peters et al., 2018a) and BERT (Devlin et al., 2018) have recently achieved state-of-the-art results on a diverse array of downstream NLP tasks. Building on recent token-level probing work, we introduce a novel edge probing task design and construct a broad suite of sub-sentence tasks derived from the traditional structured NLP pipeline. We probe word-level contextual representations from four recent models and investigate how they encode sentence structure across a range of syntactic, semantic, local, and long-range phenomena. We find that existing models trained on language modeling and translation produce strong representations for syntactic phenomena, but only offer comparably small improvements on semantic tasks over a non-contextual baseline.""","""Pros - Thorough analysis on a large number of diverse tasks - Extending the probing technique typically applied to individual encoder states to testing for presence of certain (linguistic) information based on pairs of encoders states (corresponding to pairs of words) - The comparison can be useful when deciding which representations to use for a given task Cons - Nothing serious, it is solid and important empirical study The reviewers are in consensus."""
137,"""Probabilistic Recursive Reasoning for Multi-Agent Reinforcement Learning""","['Multi-agent Reinforcement Learning', 'Recursive Reasoning']","""Humans are capable of attributing latent mental contents such as beliefs, or intentions to others. The social skill is critical in everyday life to reason about the potential consequences of their behaviors so as to plan ahead. It is known that humans use this reasoning ability recursively, i.e. considering what others believe about their own beliefs. In this paper, we start from level- pseudo-formula recursion and introduce a probabilistic recursive reasoning (PR2) framework for multi-agent reinforcement learning. Our hypothesis is that it is beneficial for each agent to account for how the opponents would react to its future behaviors. Under the PR2 framework, we adopt variational Bayes methods to approximate the opponents' conditional policy, to which each agent finds the best response and then improve their own policy. We develop decentralized-training-decentralized-execution algorithms, PR2-Q and PR2-Actor-Critic, that are proved to converge in the self-play scenario when there is one Nash equilibrium. Our methods are tested on both the matrix game and the differential game, which have a non-trivial equilibrium where common gradient-based methods fail to converge. Our experiments show that it is critical to reason about how the opponents believe about what the agent believes. We expect our work to contribute a new idea of modeling the opponents to the multi-agent reinforcement learning community. ""","""Pros: - novel idea of endowing RL agents with recursive reasoning - clear, well presented paper - thorough rebuttal and revision with new results Cons: - small-scale experiments The reviewers agree that the paper should be accepted."""
138,"""Few-shot Classification on Graphs with Structural Regularized GCNs""","['Graph Convolutional Networks', 'Few-shot', 'Classification']","""We consider the fundamental problem of semi-supervised node classification in attributed graphs with a focus on \emph{few-shot} learning. Here, we propose Structural Regularized Graph Convolutional Networks (SRGCN), novel neural network architectures extending the well-known GCN structures by stacking transposed convolutional layers for reconstruction of input features. We add a reconstruction error term in the loss function as a regularizer. Unlike standard regularization such as pseudo-formula or pseudo-formula , which controls the model complexity by including a penalty term depends solely on parameters, our regularization function is parameterized by a trainable neural network whose structure depends on the topology of the underlying graph. The new approach effectively addresses the shortcomings of previous graph convolution-based techniques for learning classifiers in the few-shot regime and significantly improves generalization performance over original GCNs when the number of labeled samples is insufficient. Experimental studies on three challenging benchmarks demonstrate that the proposed approach has matched state-of-the-art results and can improve classification accuracies by a notable margin when there are very few examples from each class.""","""A new regularized graph CNN approach is proposed for semi-supervised learning on graphs. The conventional Graph CNN is concatenated with a Transposed Network, which is used to supplement the supervised loss w.r.t. the labeled part of the graph with an unsupervised loss that serves as a regularizer measuring reconstruction errors of features. While this extension performs well and was found to be interesting in general by the reviewers, the novelty of the approach (adding a reconstruction loss), the completeness of the experimental evaluation, and the presentation quality have also been questioned consistently. The paper has improved during the course of the review, but overall the AC evaluates that paper is not upto ICLR-2019 standards in its current form. """
139,"""Universal Successor Features Approximators""","['reinforcement learning', 'zero-shot transfer', 'successor features', 'universal value functions', 'general value functions']","""The ability of a reinforcement learning (RL) agent to learn about many reward functions at the same time has many potential benefits, such as the decomposition of complex tasks into simpler ones, the exchange of information between tasks, and the reuse of skills. We focus on one aspect in particular, namely the ability to generalise to unseen tasks. Parametric generalisation relies on the interpolation power of a function approximator that is given the task description as input; one of its most common form are universal value function approximators (UVFAs). Another way to generalise to new tasks is to exploit structure in the RL problem itself. Generalised policy improvement (GPI) combines solutions of previous tasks into a policy for the unseen task; this relies on instantaneous policy evaluation of old policies under the new reward function, which is made possible through successor features (SFs). Our proposed \emph{universal successor features approximators} (USFAs) combine the advantages of all of these, namely the scalability of UVFAs, the instant inference of SFs, and the strong generalisation of GPI. We discuss the challenges involved in training a USFA, its generalisation properties and demonstrate its practical benefits and transfer abilities on a large-scale domain in which the agent has to navigate in a first-person perspective three-dimensional environment. ""","""This paper addresses an importnant and more realistic setting of multi-task RL where the reward function changes; the approach is elegant, and empirical results are convincing. The paper presents an importnant contribution to the challenging multi-task RL problem."""
140,"""Adaptive Gradient Methods with Dynamic Bound of Learning Rate""","['Optimization', 'SGD', 'Adam', 'Generalization']","""Adaptive optimization methods such as AdaGrad, RMSprop and Adam have been proposed to achieve a rapid training process with an element-wise scaling term on learning rates. Though prevailing, they are observed to generalize poorly compared with SGD or even fail to converge due to unstable and extreme learning rates. Recent work has put forward some algorithms such as AMSGrad to tackle this issue but they failed to achieve considerable improvement over existing methods. In our paper, we demonstrate that extreme learning rates can lead to poor performance. We provide new variants of Adam and AMSGrad, called AdaBound and AMSBound respectively, which employ dynamic bounds on learning rates to achieve a gradual and smooth transition from adaptive methods to SGD and give a theoretical proof of convergence. We further conduct experiments on various popular tasks and models, which is often insufficient in previous work. Experimental results show that new variants can eliminate the generalization gap between adaptive methods and SGD and maintain higher learning speed early in training at the same time. Moreover, they can bring significant improvement over their prototypes, especially on complex deep networks. The implementation of the algorithm can be found at pseudo-url .""","""The paper was found to be well-written and conveys interesting idea. However the AC notices a large body of clarifications that were provided to the reviewers (regarding the theory, experiments, and setting in general) that need to be well addressed in the paper. """
141,"""Visual Imitation with a Minimal Adversary""","['imitation', 'from pixels', 'adversarial']","""High-dimensional sparse reward tasks present major challenges for reinforcement learning agents. In this work we use imitation learning to address two of these challenges: how to learn a useful representation of the world e.g. from pixels, and how to explore efficiently given the rarity of a reward signal? We show that adversarial imitation can work well even in this high dimensional observation space. Surprisingly the adversary itself, acting as the learned reward function, can be tiny, comprising as few as 128 parameters, and can be easily trained using the most basic GAN formulation. Our approach removes limitations present in most contemporary imitation approaches: requiring no demonstrator actions (only video), no special initial conditions or warm starts, and no explicit tracking of any single demo. The proposed agent can solve a challenging robot manipulation task of block stacking from only video demonstrations and sparse reward, in which the non-imitating agents fail to learn completely. Furthermore, our agent learns much faster than competing approaches that depend on hand-crafted, staged dense reward functions, and also better compared to standard GAIL baselines. Finally, we develop a new adversarial goal recognizer that in some cases allows the agent to learn stacking without any task reward, purely from imitation.""","""The paper extends an existing approach to imitation learning, GAIL (Generative Adversarial Imitation Learning, based on an adversarial approach where a policy learner competes with a discriminator) in several ways and demonstrates that the resulting approach can learn in settings with high dimensional observation spaces, even with a very low dimensional discriminator. Empirical results show promising performance on a (simulated) robotics block stacking task, as well as a standard benchmark - Walker2D (DeepMind control suite). The reviewers and the AC note several potential weaknesses. Most importantly, the contributions of the paper are ""muddled"" (R2). The authors introduce several modifications to their baseline, GAIL, and show empirical improvements over the baseline. However, the presented experiments do systematically identify which modifications have what impact on the empirical results. For example, R2 mentions this for figure 4, where it appears on first look that the proposed approach is compared to the vanilla GAIL baseline - however, there appear to be differences from vanilla GAIL, e.g., in terms of reward structure (and possibly other modeling choices - how close is the GAIL implementation used to the original method, e.g., in terms of the policy learner and discriminator)? There is also confusion on which setting is addressed in which part of the paper, given that there is both a ""RL+IL"" and an ""imitation only"" component. In their rebuttal, the authors respond to, and clarify some of the questions raised by the reviewers, but the AC and corresponding reviewers consider many issues to remain unclear. Overall, the presentation could be much improved by indicating, for each set of experiments, what research question or hypothesis it is designed to address, and to clearly indicate conclusions on each question once the results have been discussed. In its current state, the paper reads as a list of interesting and potentially highly valuable ideas, together with a list of empirical results. The real value of the paper should come in when these are synthesized into lessons learned, e.g., why specific results are observed and what novel insights they afford the reader. Overall, the paper will benefit from a thorough revision and is not considered ready for publication at ICLR at this stage. The AC notes that they placed less weight on R3's assessment, due to their relatively low confidence, because they appear not to be familiar with key related work (GAIL), and did not respond to further requests for comments in the discussion phase. The AC also notes a potential weakness that was not brought up by the reviewers, and which they therefore did not weigh into their assessment of the paper, but nevertheless want to share to hopefully help improve a future version of the paper. Figure 6(b) should be interpreted with caution given that performance with a greater number of demonstrations (120 vs 60) showed lower performance. The authors note in the caption that one of the ""120 demos"" runs ""failed to take of"". This suggests that variance for all these runs may be underestimated with the currently used number of seeds. It is not clear what the shaded region indicates (another drawback) but if I interpret these as standard errors then this plot would suggest lower performance for higher numbers of demonstrations with some confidence - clearly that conclusion is unlikely to be correct."""
142,"""An Energy-Based Framework for Arbitrary Label Noise Correction""","['label noise', 'feature dependent noise', 'label correction', 'unsupervised machine learning', 'semi-supervised machine learning']","""We propose an energy-based framework for correcting mislabelled training examples in the context of binary classification. While existing work addresses random and class-dependent label noise, we focus on feature dependent label noise, which is ubiquitous in real-world data and difficult to model. Two elements distinguish our approach from others: 1) instead of relying on the original feature space, we employ an autoencoder to learn a discriminative representation and 2) we introduce an energy-based formalism for the label correction problem. We prove that a discriminative representation can be learned by training a generative model using a loss function comprised of the difference of energies corresponding to each class. The learned energy value for each training instance is compared to the original training labels and contradictions between energy assignment and training label are used to correct labels. We validate our method across eight datasets, spanning synthetic and realistic settings, and demonstrate the technique's state-of-the-art label correction performance. Furthermore, we derive analytical expressions to show the effect of label noise on the gradients of empirical risk.""","""The authors present an algorithm for label noise correction when the label error is a function of the input features. Strengths - Well motivated problem and a well written paper. Weaknesses - The reviewers raised concerns about theoretical guarantees on generalization; it is not clear why energy based auto-encoder / contrastive divergence would be a good measure of label accuracy especially when the feature distribution has high variance, and when there are not enough clean examples to model this distribution correctly. - Evaluations are all on toy-like tasks with small training sets, which makes it harder to gauge how well the techniques work for real-world tasks. - Its not clear how well the algorithm can be extended to multi-class problems. The authors suggested 1-vs-all, but have no experiments or results to support the claim. The authors tried to address some of the concerns raised by the reviewers in the rebuttal, e.g., how to address unavailability of correctly labeled data to train an auto-encoder. But other concerns remain. Therefore, the recommendation is to reject the paper. """
143,"""Assessing Generalization in Deep Reinforcement Learning""","['reinforcement learning', 'generalization', 'benchmark']","""Deep reinforcement learning (RL) has achieved breakthrough results on many tasks, but has been shown to be sensitive to system changes at test time. As a result, building deep RL agents that generalize has become an active research area. Our aim is to catalyze and streamline community-wide progress on this problem by providing the first benchmark and a common experimental protocol for investigating generalization in RL. Our benchmark contains a diverse set of environments and our evaluation methodology covers both in-distribution and out-of-distribution generalization. To provide a set of baselines for future research, we conduct a systematic evaluation of state-of-the-art algorithms, including those that specifically tackle the problem of generalization. The experimental results indicate that in-distribution generalization may be within the capacity of current algorithms, while out-of-distribution generalization is an exciting challenge for future work.""","""The manuscript proposes benchmarks for studying generalization in reinforcement learning, primarily through the alteration of the environment parameters of standard tasks such as Mountain Car and Half Cheetah. In contrast with methodological innovations where a numerical argument can often be made for the new method's performance on well-understood tasks, a paper introducing a new benchmark must be held to a high standard in terms of the usefulness of the benchmark in studying the phenomenon under consideration. Reviewers commended the quality of writing and considered the experiments given the set of tasks to be thorough, but there were serious concerns from several reviewers regarding how well-motivated this benchmark is and restrictions viewed as artificial (no training at test-time), concerns which the updated manuscript has failed to address. I therefore recommend rejection at this stage, and urge the authors to carefully consider the desiderata for a generalization benchmark and why their current proposed set of tasks satisfies (or doesn't satisfy) those desiderata."""
144,"""Deli-Fisher GAN: Stable and Efficient Image Generation With Structured Latent Generative Space""","['Generative Adversarial Networks', 'Structured Latent Space', 'Stable Training']","""Generative Adversarial Networks (GANs) are powerful tools for realistic image generation. However, a major drawback of GANs is that they are especially hard to train, often requiring large amounts of data and long training time. In this paper we propose the Deli-Fisher GAN, a GAN that generates photo-realistic images by enforcing structure on the latent generative space using similar approaches in \cite{deligan}. The structure of the latent space we consider in this paper is modeled as a mixture of Gaussians, whose parameters are learned in the training process. Furthermore, to improve stability and efficiency, we use the Fisher Integral Probability Metric as the divergence measure in our GAN model, instead of the Jensen-Shannon divergence. We show by experiments that the Deli-Fisher GAN performs better than DCGAN, WGAN, and the Fisher GAN as measured by inception score.""","""This paper combines two recently proposed ideas for GAN training: Fisher integral probability metrics, and the Deli-GAN. As the reviewers have pointed out, the writing is somewhat haphazard, and it's hard to identify the key contributions, why the proposed method is expected to help, and so on. The experiments are rather minimal: a single experiment comparing Inception scores to previous models on CIFAR; Inception scores are not a great measure, and the experiments don't yield much insight into where the improvement comes from. No author response was given. I don't think this paper is ready for publication in ICLR. """
145,"""UaiNets: From Unsupervised to Active Deep Anomaly Detection""","['Anomaly Detection', 'Active  Learning', 'Unsupervised Learning']","""This work presents a method for active anomaly detection which can be built upon existing deep learning solutions for unsupervised anomaly detection. We show that a prior needs to be assumed on what the anomalies are, in order to have performance guarantees in unsupervised anomaly detection. We argue that active anomaly detection has, in practice, the same cost of unsupervised anomaly detection but with the possibility of much better results. To solve this problem, we present a new layer that can be attached to any deep learning model designed for unsupervised anomaly detection to transform it into an active method, presenting results on both synthetic and real anomaly detection datasets.""","""Following the unanimous vote of the reviewers, this paper is not ready for publication at ICLR. The most significant concern raised is that there does not seem to be an adequate research contribution. Moreover, unsubstantiated claims of novelty do not adequately discuss or compare to past work."""
146,"""Zero-Resource Multilingual Model Transfer: Learning What to Share""","['cross-lingual transfer learning', 'multilingual transfer learning', 'zero-resource model transfer', 'adversarial training', 'mixture of experts', 'multilingual natural language understanding']","""Modern natural language processing and understanding applications have enjoyed a great boost utilizing neural networks models. However, this is not the case for most languages especially low-resource ones with insufficient annotated training data. Cross-lingual transfer learning methods improve the performance on a low-resource target language by leveraging labeled data from other (source) languages, typically with the help of cross-lingual resources such as parallel corpora. In this work, we propose a zero-resource multilingual transfer learning model that can utilize training data in multiple source languages, while not requiring target language training data nor cross-lingual supervision. Unlike most existing methods that only rely on language-invariant features for cross-lingual transfer, our approach utilizes both language-invariant and language-specific features in a coherent way. Our model leverages adversarial networks to learn language-invariant features and mixture-of-experts models to dynamically exploit the relation between the target language and each individual source language. This enables our model to learn effectively what to share between various languages in the multilingual setup. It results in significant performance gains over prior art, as shown in an extensive set of experiments over multiple text classification and sequence tagging tasks including a large-scale real-world industry dataset.""","""The proposed method proposes a new architecture that uses mixture of experts to determine what to share between multiple languages for transfer learning. The results are quite good. There is still a bit of a lack of framing compared to the large amount of previous work in the field, even after initial revisions to cover reviewer comments. I think that probably this requires a significant rewrite of the intro and maybe even title of the paper to scope the contributions, and also make sure in the empirical analysis that the novel contributions are evaluated independently on their own (within the same experimental setting and hyperparameters). As such, and given the high quality bar of ICLR, I can't recommend this paper be accepted at this time, but I encourage the authors to revise this explanation and re-submit a new version elsewhere."""
147,"""Architecture Compression""","['compression', 'architecture search']","""In this paper we propose a novel approach to model compression termed Architecture Compression. Instead of operating on the weight or filter space of the network like classical model compression methods, our approach operates on the architecture space. A 1-D CNN encoder/decoder is trained to learn a mapping from discrete architecture space to a continuous embedding and back. Additionally, this embedding is jointly trained to regress accuracy and parameter count in order to incorporate information about the architecture's effectiveness on the dataset. During the compression phase, we first encode the network and then perform gradient descent in continuous space to optimize a compression objective function that maximizes accuracy and minimizes parameter count. The final continuous feature is then mapped to a discrete architecture using the decoder. We demonstrate the merits of this approach on visual recognition tasks such as CIFAR-10/100, FMNIST and SVHN and achieve a greater than 20x compression on CIFAR-10.""","""The authors propose a scheme to learn a mapping between the discrete space of network architectures into a continuous embedding, and from the continuous embedding back into the space of network architectures. During the training phase, the models regress the number of parameters, and expected accuracy given the continuous embedding. Once trained, the model can be used for compression by first embedding the network structure and then performing gradient descent to maximize accuracy by minimizing the number of parameters. The optimized representation can then be mapped back into the discrete architecture space. Overall, the main idea of this work is very interesting, and the experiments show that the method has some promise. However, as was noted by the reviewers, the paper could be significantly strengthened by performing additional experiments and analyses. As such, the AC agrees with the reviewers that the paper in its present form is not suitable for acceptance, but the authors are encouraged to revise and resubmit this work to a future venue. """
148,"""ProxQuant: Quantized Neural Networks via Proximal Operators""","['Model quantization', 'Optimization', 'Regularization']","""To make deep neural networks feasible in resource-constrained environments (such as mobile devices), it is beneficial to quantize models by using low-precision weights. One common technique for quantizing neural networks is the straight-through gradient method, which enables back-propagation through the quantization mapping. Despite its empirical success, little is understood about why the straight-through gradient method works. Building upon a novel observation that the straight-through gradient method is in fact identical to the well-known Nesterovs dual-averaging algorithm on a quantization constrained optimization problem, we propose a more principled alternative approach, called ProxQuant , that formulates quantized network training as a regularized learning problem instead and optimizes it via the prox-gradient method. ProxQuant does back-propagation on the underlying full-precision vector and applies an efficient prox-operator in between stochastic gradient steps to encourage quantizedness. For quantizing ResNets and LSTMs, ProxQuant outperforms state-of-the-art results on binary quantization and is on par with state-of-the-art on multi-bit quantization. We further perform theoretical analyses showing that ProxQuant converges to stationary points under mild smoothness assumptions, whereas variants such as lazy prox-gradient method can fail to converge in the same setting.""","""A novel approach for quantized deep neural nets is proposed, which is more principled than commonly used straight-through gradient method. A theoretical analysis of the algorithm's converegence is presented, and empirical results show advantages of the proposed approach. """
149,"""Integer Networks for Data Compression with Latent-Variable Models""","['data compression', 'variational models', 'network quantization']","""We consider the problem of using variational latent-variable models for data compression. For such models to produce a compressed binary sequence, which is the universal data representation in a digital world, the latent representation needs to be subjected to entropy coding. Range coding as an entropy coding technique is optimal, but it can fail catastrophically if the computation of the prior differs even slightly between the sending and the receiving side. Unfortunately, this is a common scenario when floating point math is used and the sender and receiver operate on different hardware or software platforms, as numerical round-off is often platform dependent. We propose using integer networks as a universal solution to this problem, and demonstrate that they enable reliable cross-platform encoding and decoding of images using variational models.""","""This paper addresses the issue of numerical rounding-off errors that can arise when using latent variable models for data compression, e.g., because of differences in floating point arithmetic across different platforms (sender and receiver). The authors propose using neural networks that perform integer arithmetic (integer networks) to mitigate this issue. The problem statement is well described, and the presentation is generally OK, although it could be improved in certain aspects as pointed out by the reviewers. The experiments are properly carried out, and the experimental results are good. Thank you for addressing the questions raised by the reviewers. After taking into account the author's responds, there is consensus that the paper is worthy of publication. I therefore recommend acceptance. """
150,"""Stochastic Learning of Additive Second-Order Penalties with Applications to Fairness""",['fairness'],"""Many notions of fairness may be expressed as linear constraints, and the resulting constrained objective is often optimized by transforming the problem into its Lagrangian dual with additive linear penalties. In non-convex settings, the resulting problem may be difficult to solve as the Lagrangian is not guaranteed to have a deterministic saddle-point equilibrium. In this paper, we propose to modify the linear penalties to second-order ones, and we argue that this results in a more practical training procedure in non-convex, large-data settings. For one, the use of second-order penalties allows training the penalized objective with a fixed value of the penalty coefficient, thus avoiding the instability and potential lack of convergence associated with two-player min-max games. Secondly, we derive a method for efficiently computing the gradients associated with the second-order penalties in stochastic mini-batch settings. Our resulting algorithm performs well empirically, learning an appropriately fair classifier on a number of standard benchmarks.""","""The paper presents a method to stochastically optimize second-order penalties and show how this could be applied to training fairness-aware classifiers, where the linear penalties associated with common fairness criteria are expressed as the second order penalties. While the reviewers acknowledged the potential usefulness of the proposed approach, all of them agreed that the paper requires: (1) major improvement in clarifying important points related to the approach (see R3s detailed comments; R2s concern on using the double sampling method to train non-convex models; see R1s and R3s concerns regarding the double summation/integral terms and how this effects runtime), and (2) major improvement in justifying its application to fairness; as noted by R2, there is no sufficient evidence why non-convex models are actually useful in the experiments. Given that fairness problems are currently studied on the small scale datasets (which is not this papers fault), a comparison to simpler methods for fairness or other applications could substantially strengthen the contribution and evaluation of this work. We hope the reviews are useful for improving and revising the paper. """
151,"""Generating Multi-Agent Trajectories using Programmatic Weak Supervision""","['deep learning', 'generative models', 'imitation learning', 'hierarchical methods', 'data programming', 'weak supervision', 'spatiotemporal']","""We study the problem of training sequential generative models for capturing coordinated multi-agent trajectory behavior, such as offensive basketball gameplay. When modeling such settings, it is often beneficial to design hierarchical models that can capture long-term coordination using intermediate variables. Furthermore, these intermediate variables should capture interesting high-level behavioral semantics in an interpretable and manipulable way. We present a hierarchical framework that can effectively learn such sequential generative models. Our approach is inspired by recent work on leveraging programmatically produced weak labels, which we extend to the spatiotemporal regime. In addition to synthetic settings, we show how to instantiate our framework to effectively model complex interactions between basketball players and generate realistic multi-agent trajectories of basketball gameplay over long time periods. We validate our approach using both quantitative and qualitative evaluations, including a user study comparison conducted with professional sports analysts.""","""The paper presents generative models to produce multi-agent trajectories. The approach of using a simple heuristic labeling function that labels variables that would otherwise be latent in training data is novel and and results in higher quality than the previously proposed baselines. In response to reviewer suggestions, authors included further results with models that share parameters across agents as well as agent-specific parameters and further clarifications were made for other main comments (i.e., baselines that train the hierarchical model by maximizing an ELBO on the marginal likelihood?)."""
152,"""MARGINALIZED AVERAGE ATTENTIONAL NETWORK FOR WEAKLY-SUPERVISED LEARNING""","['feature aggregation', 'weakly supervised learning', 'temporal action localization']","""In weakly-supervised temporal action localization, previous works have failed to locate dense and integral regions for each entire action due to the overestimation of the most salient regions. To alleviate this issue, we propose a marginalized average attentional network (MAAN) to suppress the dominant response of the most salient regions in a principled manner. The MAAN employs a novel marginalized average aggregation (MAA) module and learns a set of latent discriminative probabilities in an end-to-end fashion. MAA samples multiple subsets from the video snippet features according to a set of latent discriminative probabilities and takes the expectation over all the averaged subset features. Theoretically, we prove that the MAA module with learned latent discriminative probabilities successfully reduces the difference in responses between the most salient regions and the others. Therefore, MAAN is able to generate better class activation sequences and identify dense and integral action regions in the videos. Moreover, we propose a fast algorithm to reduce the complexity of constructing MAA from pseudo-formula to pseudo-formula . Extensive experiments on two large-scale video datasets show that our MAAN achieves a superior performance on weakly-supervised temporal action localization. ""","""The paper proposes a new attentional pooling mechanism that potentially addresses the issues of simple attention-based weighted averaging (where discriminative parts/frames might get disportionately high attentions). A nice contribution of the paper is to propose an alternative mechanism with theoretical proofs, and it also presents a method for fast recurrent computation. The experimental results show that the proposed attention mechanism improves over prior methods (e.g., STPN) on THUMOS14 and ActivityNet1.3 datasets. In terms of weaknesses: (1) the computational cost may be quite significant. (2) the proposed method should be evaluated over several tasks beyond activity recognition, but its unclear how it would work. The authors provided positive proof-of-concept results on weakly supervised object localization task, improving over CAM-based methods. However, CAM baseline is a reasonable but not the strongest method and the weakly-supervised object recognition/segmentation domains are much more competitive domains, so it's unclear if the proposed method would achieve the state-of-the-art by simply replacing the weighted-averaging-attentional-pooling with the proposed attention mechanism. In addition, the description on how to perform attentional pooling over images is not clearly described (its not clear how the 1D sequence-based recurrent attention method can be extended to 2-D cases). However, this would not be a reason to reject the paper. Finally, the papers presentation would need improvement. I would suggest that the authors give more intuitive explanations and rationale before going into technical details. The paper starts with Figure 1 which is not really well motivated/explained, so it could be moved to a later part. Overall, there are interesting technical contributions with positive results, but there are issues to be addressed. """
153,"""Amortized Context Vector Inference for Sequence-to-Sequence Networks""","['neural attention', 'sequence-to-sequence', 'variational inference']","""Neural attention (NA) has become a key component of sequence-to-sequence models that yield state-of-the-art performance in as hard tasks as abstractive document summarization (ADS), machine translation (MT), and video captioning (VC). NA mechanisms perform inference of context vectors; these constitute weighted sums of deterministic input sequence encodings, adaptively sourced over long temporal horizons. Inspired from recent work in the field of amortized variational inference (AVI), in this work we consider treating the context vectors generated by soft-attention (SA) models as latent variables, with approximate finite mixture model posteriors inferred via AVI. We posit that this formulation may yield stronger generalization capacity, in line with the outcomes of existing applications of AVI to deep networks. To illustrate our method, we implement it and experimentally evaluate it considering challenging ADS, VC, and MT benchmarks. This way, we exhibit its improved effectiveness over state-of-the-art alternatives.""","""The overall view of the reviewers is that the paper is not quite good enough as it stands. The reviewers also appreciate the contributions so taking the comments into account and resubmit elsewhere is encouraged. """
154,"""Double Viterbi: Weight Encoding for High Compression Ratio and Fast On-Chip Reconstruction for Deep Neural Network""","['quantization', 'pruning', 'memory footprint', 'model compression', 'sparse matrix']","""Weight pruning has been introduced as an efficient model compression technique. Even though pruning removes significant amount of weights in a network, memory requirement reduction was limited since conventional sparse matrix formats require significant amount of memory to store index-related information. Moreover, computations associated with such sparse matrix formats are slow because sequential sparse matrix decoding process does not utilize highly parallel computing systems efficiently. As an attempt to compress index information while keeping the decoding process parallelizable, Viterbi-based pruning was suggested. Decoding non-zero weights, however, is still sequential in Viterbi-based pruning. In this paper, we propose a new sparse matrix format in order to enable a highly parallel decoding process of the entire sparse matrix. The proposed sparse matrix is constructed by combining pruning and weight quantization. For the latest RNN models on PTB and WikiText-2 corpus, LSTM parameter storage requirement is compressed 19x using the proposed sparse matrix format compared to the baseline model. Compressed weight and indices can be reconstructed into a dense matrix fast using Viterbi encoders. Simulation results show that the proposed scheme can feed parameters to processing elements 20 % to 106 % faster than the case where the dense matrix values directly come from DRAM.""","""The authors propose an efficient scheme for encoding sparse matrices which allow weights to be compressed efficiently. At the same time, the proposed scheme allows for fast parallelizable decompression into a dense matrix using Viterbi-based pruning. The reviewers noted that the techniques address an important problem relevant to deploying neural networks on resource-constrained platforms, and although the work builds on previous work, it is important from a practical standpoint. The reviewers noted a number of concerns on the initial draft of this work related to the experimental methodology and the absence of runtime comparison against the baseline, which the reviewers have since fixed in the revised draft. The reviewers were unanimous in recommending that the revision be accepted, and the authors are requested to incorporate the final changes that they said they would make in the camera-ready version. """
155,"""Dynamic Pricing on E-commerce Platform with Deep Reinforcement Learning""","['reinforcement learning', 'dynamic pricing', 'e-commerce', 'revenue management', 'field experiment']","""In this paper we develop an approach based on deep reinforcement learning (DRL) to address dynamic pricing problem on E-commerce platform. We models real-world E-commerce dynamic pricing problem as Markov Decision Process. Environment state are defined with four groups of different business data. We make several main improvements on the state-of-the-art DRL-based dynamic pricing approaches: 1. We first extend the application of dynamic pricing to a continuous pricing action space. 2. We solve the unknown demand function problem by designing different reward functions. 3. The cold-start problem is addressed by introducing pre-training and evaluation using the historical sales data. Field experiments are designed and conducted on real-world E-commerce platform, pricing thousands of SKUs of products lasting for months. The experiment results shows that, on E-commerce platform, the difference of the revenue conversion rates (DRCR) is a more suitable reward function than the revenue only, which is different from the conclusion from previous researches. Meanwhile, the proposed continuous action model performs better than the discrete one.""",""" This is an interesting topic but the reviewers had substantial concerns on the clarity and significance of the contribution. """
156,"""Learning to Infer and Execute 3D Shape Programs""","['Program Synthesis', '3D Shape Modeling', 'Self-supervised Learning']","""Human perception of 3D shapes goes beyond reconstructing them as a set of points or a composition of geometric primitives: we also effortlessly understand higher-level shape structure such as the repetition and reflective symmetry of object parts. In contrast, recent advances in 3D shape sensing focus more on low-level geometry but less on these higher-level relationships. In this paper, we propose 3D shape programs, integrating bottom-up recognition systems with top-down, symbolic program structure to capture both low-level geometry and high-level structural priors for 3D shapes. Because there are no annotations of shape programs for real shapes, we develop neural modules that not only learn to infer 3D shape programs from raw, unannotated shapes, but also to execute these programs for shape reconstruction. After initial bootstrapping, our end-to-end differentiable model learns 3D shape programs by reconstructing shapes in a self-supervised manner. Experiments demonstrate that our model accurately infers and executes 3D shape programs for highly complex shapes from various categories. It can also be integrated with an image-to-shape module to infer 3D shape programs directly from an RGB image, leading to 3D shape reconstructions that are both more accurate and more physically plausible.""","""This paper presents a method whereby a model learns to describe 3D shapes as programs which generate said shapes. Beyond introducing some new techniques in neural program synthesis through the use of loops, this method also produces disentangled representations of the shapes by deconstructing them into the program that produced them, thereby introducing an interesting and useful level of abstraction that could be exploited by models, agents, and other learning algorithms. Despite some slightly aggressive anonymous comments by a third party, the reviewers agree that this paper is solid and publishable, and I have no qualms in recommending it from inclusion in the proceedings."""
157,"""Transformer-XL: Language Modeling with Longer-Term Dependency""","['Language Modeling', 'Self-Attention']","""We propose a novel neural architecture, Transformer-XL, for modeling longer-term dependency. To address the limitation of fixed-length contexts, we introduce a notion of recurrence by reusing the representations from the history. Empirically, we show state-of-the-art (SoTA) results on both word-level and character-level language modeling datasets, including WikiText-103, One Billion Word, Penn Treebank, and enwiki8. Notably, we improve the SoTA results from 1.06 to 0.99 in bpc on enwiki8, from 33.0 to 18.9 in perplexity on WikiText-103, and from 28.0 to 23.5 in perplexity on One Billion Word. Performance improves when the attention length increases during evaluation, and our best model attends to up to 1,600 words and 3,800 characters. To quantify the effective length of dependency, we devise a new metric and show that on WikiText-103 Transformer-XL manages to model dependency that is about 80% longer than recurrent networks and 450% longer than Transformer. Moreover, Transformer-XL is up to 1,800+ times faster than vanilla Transformer during evaluation.""","""despite the (significant) improvement in language modelling, it has always been a thorny issue whether better language models (at this level) lead to better performance in the downstream task or whether such a technique could be used to build a better conditional language model which often focuses on the aspect of generation. in this context, the reviewers found it difficult to see the merit of the proposed approach, as the technique itself may be considered a rather trivial application of earlier approaches such as truncated backprop. it would be good to apply this technique to e.g. document-level generation and see if the proposed approach can strike an amazing balance between computational efficiency and generation performance."""
158,"""The wisdom of the crowd: reliable deep reinforcement learning through ensembles of Q-functions""","['reinforcement learning', 'ensembles', 'deep learning', 'neural network']","""Reinforcement learning agents learn by exploring the environment and then exploiting what they have learned. This frees the human trainers from having to know the preferred action or intrinsic value of each encountered state. The cost of this freedom is reinforcement learning is slower and more unstable than supervised learning. We explore the possibility that ensemble methods can remedy these shortcomings and do so by investigating a novel technique which harnesses the wisdom of the crowds by bagging Q-function approximator estimates. Our results show that this proposed approach improves all three tasks and reinforcement learning approaches attempted. We are able to demonstrate that this is a direct result of the increased stability of the action portion of the state-action-value function used by Q-learning to select actions and by policy gradient methods to train the policy. ""","""The paper suggests using an ensemble of Q functions for Q-learning. This idea is related to bootstrapped DQN and more recent work on distributional RL and quantile regression in RL. Given the similarity, a comparison against these approaches (or a subset of those) is necessary. The experiments are limited to very simple environment (e.g. swing-up and cart-pole). The paper in its current form does not pass the bar for acceptance at ICLR."""
159,"""Iteratively Learning from the Best""","['noisy samples', 'deep learning', 'generative adversarial network']","""We study a simple generic framework to address the issue of bad training data; both bad labels in supervised problems, and bad samples in unsupervised ones. Our approach starts by fitting a model to the whole training dataset, but then iteratively improves it by alternating between (a) revisiting the training data to select samples with lowest current loss, and (b) re-training the model on only these selected samples. It can be applied to any existing model training setting which provides a loss measure for samples, and a way to refit on new ones. We show the merit of this approach in both theory and practice We first prove statistical consistency, and linear convergence to the ground truth and global optimum, for two simpler model settings: mixed linear regression, and gaussian mixture models. We then demonstrate its success empirically in (a) saving the accuracy of existing deep image classifiers when there are errors in the labels of training images, and (b) improving the quality of samples generated by existing DC-GAN models, when it is given training data that contains a fraction of the images from a different and unintended dataset. The experimental results show significant improvement over the baseline methods that ignore the existence of bad labels/samples. ""","""This paper addresses the problem of learning with outliers, which many reviewers agree is an important direction. However, reviewers point to issues with the experiments (missing baselines, ablations, etc.) and are concerned that the assumptions in the theoretical analysis are too strong. These were potentially addressed in a revised version of the paper, but the revisions are so major that I do not think it is appropriate to consider them in the review process (and it is hard to assess to what extent they address the issues without asking reviewers to do a thorough re-appraisal, which goes beyond the scope of their duties). I encourage the authors to take reviewer comments into account and prepare a more polished version of the manuscript for future submission."""
160,"""Open Vocabulary Learning on Source Code with a Graph-Structured Cache""","['deep learning', 'graph neural network', 'open vocabulary', 'natural language processing', 'source code', 'abstract syntax tree', 'code completion', 'variable naming']","""Machine learning models that take computer program source code as input typically use Natural Language Processing (NLP) techniques. However, a major challenge is that code is written using an open, rapidly changing vocabulary due to, e.g., the coinage of new variable and method names. Reasoning over such a vocabulary is not something for which most NLP methods are designed. We introduce a Graph-Structured Cache to address this problem; this cache contains a node for each new word the model encounters with edges connecting each word to its occurrences in the code. We find that combining this graph-structured cache strategy with recent Graph-Neural-Network-based models for supervised learning on code improves the models' performance on a code completion task and a variable naming task --- with over 100\% relative improvement on the latter --- at the cost of a moderate increase in computation time.""","""This paper introduces fairly complex methods for dealing with OOV words in graphs representing source code, and aims to show that these improve over existing methods. The chief and valid concern raised by the reviewers was that the experiments had been changed so as to not allow proper comparison to prior work, or where comparison can be made. It is essential that a new method such as this be properly evaluated against existing benchmarks, under the same experimental conditions as presented in related literature. It seems that while the method is interesting, the empirical section of this paper needs reworking in order to be suitable for publication."""
161,"""Quantization for Rapid Deployment of Deep Neural Networks""",[],"""This paper aims at rapid deployment of the state-of-the-art deep neural networks (DNNs) to energy efficient accelerators without time-consuming fine tuning or the availability of the full datasets. Converting DNNs in full precision to limited precision is essential in taking advantage of the accelerators with reduced memory footprint and computation power. However, such a task is not trivial since it often requires the full training and validation datasets for profiling the network statistics and fine tuning the networks to recover the accuracy lost after quantization. To address these issues, we propose a simple method recognizing channel-level distribution to reduce the quantization-induced accuracy loss and minimize the required image samples for profiling. We evaluated our method on eleven networks trained on the ImageNet classification benchmark and a network trained on the Pascal VOC object detection benchmark. The results prove that the networks can be quantized into 8-bit integer precision without fine tuning.""","""This paper proposes an 8-bit quantization strategy for rapid DNN deployment. 3 reviewers all rated this paper as marginally below acceptance threshold due to lack of novelty. 8 bit quantization (including channel-wise) is a well studied task. The paper lacks comparison with peer work. """
162,"""Graph2Seq: Graph to Sequence Learning with Attention-Based Neural Networks""","['Graph Encoder', 'Graph Decoder', 'Graph2Seq', 'Graph Attention']","""The celebrated Sequence to Sequence learning (Seq2Seq) technique and its numerous variants achieve excellent performance on many tasks. However, many machine learning tasks have inputs naturally represented as graphs; existing Seq2Seq models face a significant challenge in achieving accurate conversion from graph form to the appropriate sequence. To address this challenge, we introduce a general end-to-end graph-to-sequence neural encoder-decoder architecture that maps an input graph to a sequence of vectors and uses an attention-based LSTM method to decode the target sequence from these vectors. Our method first generates the node and graph embeddings using an improved graph-based neural network with a novel aggregation strategy to incorporate edge direction information in the node embeddings. We further introduce an attention mechanism that aligns node embeddings and the decoding sequence to better cope with large graphs. Experimental results on bAbI, Shortest Path, and Natural Language Generation tasks demonstrate that our model achieves state-of-the-art performance and significantly outperforms existing graph neural networks, Seq2Seq, and Tree2Seq models; using the proposed bi-directional node embedding aggregation strategy, the model can converge rapidly to the optimal performance.""","""Strengths: The work proposes a novel architecture for graph to sequence learning. The paper shows improved performance on synthetic transduction tasks and for graph to text generation. Weaknesses: Multiple reviewers felt that the experiments were insufficient to evaluate the novel aspects of the submission relative to prior work. Newer experiments with the proposed aggregation strategy and a different graph representation were not as promising with respect to simple baselines. Points of contention: The discussion with the authors and one of the reviewers was particular contentious. The title of the paper & sentences within the paper such as ""We propose a new attention-based neural networks paradigm to elegantly address graph- to-sequence learning problems"" caused significant contention, as this was perceived to discount the importance of prior work on graph-to-sequence problems which led to a perception of the paper ""overclaiming"" novelty. Consensus: Consensus was not reached, but both the reviewer with the lowest score and one of the reviewers giving a 6 came to the consensus that the experimental evaluation does not yet evaluate the novel aspects of the submission thoroughly enough. Due to the aggregate score, factors discussed above (and others) the AC recommends rejection; however, this work shows promise and additional experimental work should allow a new set of reviewers to better understand the behaviour and utility of the proposed method. """
163,"""Parameter efficient training of deep convolutional neural networks by dynamic sparse reparameterization""","['sparse', 'reparameterization', 'overparameterization', 'convolutional neural network', 'training', 'compression', 'pruning']","""Modern deep neural networks are highly overparameterized, and often of huge sizes. A number of post-training model compression techniques, such as distillation, pruning and quantization, can reduce the size of network parameters by a substantial fraction with little loss in performance. However, training a small network of the post-compression size de novo typically fails to reach the same level of accuracy achieved by compression of a large network, leading to a widely-held belief that gross overparameterization is essential to effective learning. In this work, we argue that this is not necessarily true. We describe a dynamic sparse reparameterization technique that closed the performance gap between a model compressed through iterative pruning and a model of the post-compression size trained de novo. We applied our method to training deep residual networks and showed that it outperformed existing reparameterization techniques, yielding the best accuracy for a given parameter budget for training. Compared to existing dynamic reparameterization methods that reallocate non-zero parameters during training, our approach achieved better performance at lower computational cost. Our method is not only of practical value for training under stringent memory constraints, but also potentially informative to theoretical understanding of generalization properties of overparameterized deep neural networks. """,""" The authors presents a technique for training neural networks, through dynamic sparse reparameterization. The work builds on previous work notably SET (Mocanu et al., 18), but the authors propose to use an adaptive threshold for and a heuristic for determining how to reparameterize weights across layers. The reviewers raised a number of concerns on the original manuscript, most notably 1) that the work lacked comparisons against existing dynamic reparameterization schemes, 2) an analysis of the computational complexity of the proposed method relative to other works, and that 3) the work is an incremental improvement over SET. In the revised version, the authors revised the paper to address the various concerns raised by the reviewers. To address weakness 1) the authors ran experiments comparing the proposed approach to SET and DeepR, and demonstrated that the proposed method performs at least as well, or is better than either approach. While the new draft is in the ACs view a significant improvement over the initial version, the reviewers still had concerns about the fact that the work appears to be incremental relative to SET, and that the differences in performance between the two models were not very large (although the authors note that the differences are statistically significant). The reviewers were not entirely unanimous in their decision, which meant that the scores that this work received placed it at the borderline for acceptance. As such, the AC ultimately decide to recommend rejection, though the authors are encouraged to resubmit the revised version of the paper to a future venue. """
164,"""How to train your MAML""","['meta-learning', 'deep-learning', 'few-shot learning', 'supervised learning', 'neural-networks', 'stochastic optimization']","""The field of few-shot learning has recently seen substantial advancements. Most of these advancements came from casting few-shot learning as a meta-learning problem.Model Agnostic Meta Learning or MAML is currently one of the best approaches for few-shot learning via meta-learning. MAML is simple, elegant and very powerful, however, it has a variety of issues, such as being very sensitive to neural network architectures, often leading to instability during training, requiring arduous hyperparameter searches to stabilize training and achieve high generalization and being very computationally expensive at both training and inference times. In this paper, we propose various modifications to MAML that not only stabilize the system, but also substantially improve the generalization performance, convergence speed and computational overhead of MAML, which we call MAML++.""","""This paper proposes several improvements for the MAML algorithm that improve its stability and performance. Strengths: The improvements are useful for future researchers building upon the MAML algorithm. The results demonstrate a significant improvement over MAML. The authors revised the paper to address concerns about overstatements Weaknesses: The paper does not present a major conceptual advance. It would also be very helpful to present a more careful ablation study of the six individual techniques. Overall, the significance of the results outweights the weaknesses. However, the authors are strongly encouraged to perform and include a more detailed ablation study in the final paper. I recommend accept."""
165,"""A Rotation and a Translation Suffice: Fooling CNNs with Simple Transformations""","['robustness', 'spatial transformations', 'invariance', 'rotations', 'data augmentation', 'robust optimization']","""We show that simple spatial transformations, namely translations and rotations alone, suffice to fool neural networks on a significant fraction of their inputs in multiple image classification tasks. Our results are in sharp contrast to previous work in adversarial robustness that relied on more complicated optimization ap- proaches unlikely to appear outside a truly adversarial context. Moreover, the misclassifying rotations and translations are easy to find and require only a few black-box queries to the target model. Overall, our findings emphasize the need to design robust classifiers even for natural input transformations in benign settings. ""","""This paper demonstrated interesting observations that simple transformations such as a rotation and a translation is enough to fool CNNs. Major concern of the paper is the novelty. Similar ideas have been proposed before by many previous researchers. Other networks trying to address this issue have been proposed. Such as those rotation-invariant neural networks. The grid search attack used in the experiments may be not convincing. Overall, this paper is not ready for publication."""
166,"""Prior Networks for Detection of Adversarial Attacks""","['Uncertainty', 'Prior Networks', 'Adversarial Attacks', 'Detection']","""Adversarial examples are considered a serious issue for safety critical applications of AI, such as finance, autonomous vehicle control and medicinal applications. Though significant work has resulted in increased robustness of systems to these attacks, systems are still vulnerable to well-crafted attacks. To address this problem several adversarial attack detection methods have been proposed. However, system can still be vulnerable to adversarial samples that are designed to specifically evade these detection methods. One recent detection scheme that has shown good performance is based on uncertainty estimates derived from Monte-Carlo dropout ensembles. Prior Networks, a new method of estimating predictive uncertainty, have been shown to outperform Monte-Carlo dropout on a range of tasks. One of the advantages of this approach is that the behaviour of a Prior Network can be explicitly tuned to, for example, predict high uncertainty in regions where there are no training data samples. In this work Prior Networks are applied to adversarial attack detection using measures of uncertainty in a similar fashion to Monte-Carlo Dropout. Detection based on measures of uncertainty derived from DNNs and Monte-Carlo dropout ensembles are used as a baseline. Prior Networks are shown to significantly out-perform these baseline approaches over a range of adversarial attacks in both detection of whitebox and blackbox configurations. Even when the adversarial attacks are constructed with full knowledge of the detection mechanism, it is shown to be highly challenging to successfully generate an adversarial sample.""","""This paper addresses an important topic and was generally well-written. However, reviewers pointed out serious issues with the evaluation (using weak or poorly chosen attacks), and some conceptual confusions (e.g. conflating adversarial examples with out-of-distribution examples, unsubstantiated claim that adversarial examples lie off the data manifold)."""
167,"""Mental Fatigue Monitoring using Brain Dynamics Preferences""","['mental fatigue', 'brain dynamics preference', 'brain dynamics ranking', 'channel reliability', 'channel Selection']","""Driver's cognitive state of mental fatigue significantly affects driving performance and more importantly public safety. Previous studies leverage the response time (RT) as the metric for mental fatigue and aim at estimating the exact value of RT using electroencephalogram (EEG) signals within a regression model. However, due to the easily corrupted EEG signals and also non-smooth RTs during data collection, regular regression methods generally suffer from poor generalization performance. Considering that human response time is the reflection of brain dynamics preference rather than a single value, a novel model called Brain Dynamic ranking (BDrank) has been proposed. BDrank could learn from brain dynamics preferences using EEG data robustly and preserve the ordering corresponding to RTs. BDrank model is based on the regularized alternative ordinal classification comparing to regular regression based practices. Furthermore, a transition matrix is introduced to characterize the reliability of each channel used in EEG data, which helps in learning brain dynamics preferences only from informative EEG channels. In order to handle large-scale EEG signals~and obtain higher generalization, an online-generalized Expectation Maximum (OnlineGEM) algorithm also has been proposed to update BDrank in an online fashion. Comprehensive empirical analysis on EEG signals from 44 participants shows that BDrank together with OnlineGEM achieves substantial improvements in reliability while simultaneously detecting possible less informative and noisy EEG channels.""","""The manuscript describes a novel technique predicting metal fatigue based on EEG measurements. The work is motivated by an application to driving safety. Reviewers and the AC agreed that the main motivation for the proposed work, and perhaps the results, are likely to be of interest to the applied BCI community. The reviewers and ACs noted weakness in the original submission related to the clarity of the presentation and breadth of empirical evaluation. In particular, only a few baselines were considered. As a result, for the non-expert, it is also unclear if the proposed methods are compared against the state of the art. There was also a particular concern that this work may not be a good fit for an ICLR audience. """
168,"""VARIATIONAL SGD: DROPOUT , GENERALIZATION AND CRITICAL POINT AT THE END OF CONVEXITY""","['Bayesian inference', 'neural networks', 'generalization', 'critical point solution']","""The goal of the paper is to propose an algorithm for learning the most generalizable solution from given training data. It is shown that Bayesian approach leads to a solution that dependent on statistics of training data and not on particular samples. The solution is stable under perturbations of training data because it is defined by an integral contribution of multiple maxima of the likelihood and not by a single global maximum. Specifically, the Bayesian probability distribution of parameters (weights) of a probabilistic model given by a neural network is estimated via recurrent variational approximations. Derived recurrent update rules correspond to SGD-type rules for finding a minimum of an effective loss that is an average of an original negative log-likelihood over the Gaussian distributions of weights, which makes it a function of means and variances. The effective loss is convex for large variances and non-convex in the limit of small variances. Among stationary solutions of the update rules there are trivial solutions with zero variances at local minima of the original loss and a single non-trivial solution with finite variances that is a critical point at the end of convexity of the effective loss in the mean-variance space. At the critical point both first- and second-order gradients of the effective loss w.r.t. means are zero. The empirical study confirms that the critical point represents the most generalizable solution. While the location of the critical point in the weight space depends on specifics of the used probabilistic model some properties at the critical point are universal and model independent.""","""This paper studies a variational formulation of the loss minimization to study the solution that generalizes the most. An expectation of the loss wrt a Gaussian distribution is minimized to find the mean and variance of the Gaussian distribution. As the variance goes to zero, we recover the original loss, but for a higher value of variance, the loss may be convex. This is used to study the generalizability of the landscape. Both objective and solutions of the paper are unclear and not communicated well. There is not enough citation to previous work (e.g., Gaussian homotopy exactly considers this problem, and there are papers that study the convexity of the expectation of the loss function). There are no experimental results either to confirm the theoretical finding. All the reviewers struggle to understand both the problem and solutions discussed in this paper. I believe that the paper could become useful if reviewers' feedback is taken seriously to improve the paper."""
169,"""Count-Based Exploration with the Successor Representation""","['reinforcement learning', 'successor representation', 'exploration', 'atari']","""The problem of exploration in reinforcement learning is well-understood in the tabular case and many sample-efficient algorithms are known. Nevertheless, it is often unclear how the algorithms in the tabular setting can be extended to tasks with large state-spaces where generalization is required. Recent promising developments generally depend on problem-specific density models or handcrafted features. In this paper we introduce a simple approach for exploration that allows us to develop theoretically justified algorithms in the tabular case but that also give us intuitions for new algorithms applicable to settings where function approximation is required. Our approach and its underlying theory is based on the substochastic successor representation, a concept we develop here. While the traditional successor representation is a representation that defines state generalization by the similarity of successor states, the substochastic successor representation is also able to implicitly count the number of times each state (or feature) has been observed. This extension connects two until now disjoint areas of research. We show in traditional tabular domains (RiverSwim and SixArms) that our algorithm empirically performs as well as other sample-efficient algorithms. We then describe a deep reinforcement learning algorithm inspired by these ideas and show that it matches the performance of recent pseudo-count-based methods in hard exploration Atari 2600 games.""","""This paper was on the borderline. I am sympathetic to the authors' point about computational resources. It is helpful to demonstrate performance gains that offer ""jump start"" performance benefits, as the authors argue. However, the empirical results even on this part are still somewhat mixed-- for example, the proposed approach struggles on Private Eye (doing far worse than DQN) in Table 2. In addition, while it is beneficial to remove the need for training a density model, it would be good to show a place where a density model fails (perhaps because it is so hard to find a good one) compared to their proposed approach. """
170,"""Transferring SLU Models in Novel Domains""","['transfer learning', 'semantic representation', 'spoken language understanding']","""Spoken language understanding (SLU) is a critical component in building dialogue systems. When building models for novel natural language domains, a major challenge is the lack of data in the new domains, no matter whether the data is annotated or not. Recognizing and annotating ``intent'' and ``slot'' of natural languages is a time-consuming process. Therefore, spoken language understanding in low resource domains remains a crucial problem to address. In this paper, we address this problem by proposing a transfer-learning method, whereby a SLU model is transferred to a novel but data-poor domain via a deep neural network framework. We also introduce meta-learning in our work to bridge the semantic relations between seen and unseen data, allowing new intents to be recognized and new slots to be filled with much lower new training effort. We show the performance improvement with extensive experimental results for spoken language understanding in low resource domains. We show that our method can also handle novel intent recognition and slot-filling tasks. Our methodology provides a feasible solution for alleviating data shortages in spoken language understanding.""","""This paper proposes a transfer learning approach based on previous works on this area, to build language understanding models for new domains. Experimental results show improved performance in comparison to previous studies in terms of slot and intent accuracies in multiple setups. The work is interesting and useful, but is not novel given the previous work. The paper organization is also not great, for example, the intro should introduce the approach beyond just mentioning transfer learning and meta-learning. The improvements over the baselines look good, but the baselines themselves are quite simple. It'd be better to include comparisons with other state of the art methods. Also, the improvements over DNN are not consistent, it would be good to analyze and come up with suggestions on when to use which approach. """
171,"""Hierarchically Clustered Representation Learning""","['Representation learning', 'Hierarchical clustering', 'Nonparametric Bayesian modeling']","""The joint optimization of representation learning and clustering in the embedding space has experienced a breakthrough in recent years. In spite of the advance, clustering with representation learning has been limited to flat-level categories, which oftentimes involves cohesive clustering with a focus on instance relations. To overcome the limitations of flat clustering, we introduce hierarchically clustered representation learning (HCRL), which simultaneously optimizes representation learning and hierarchical clustering in the embedding space. Specifically, we place a nonparametric Bayesian prior on embeddings to handle dynamic mixture hierarchies under the variational autoencoder framework, and to adopt the generative process of a hierarchical-versioned Gaussian mixture model. Compared with a few prior works focusing on unifying representation learning and hierarchical clustering, HCRL is the first model to consider a generation of deep embeddings from every component of the hierarchy, not just leaf components. This generation process enables more meaningful separations and mergers of clusters via branches in a hierarchy. In addition to obtaining hierarchically clustered embeddings, we can reconstruct data by the various abstraction levels, infer the intrinsic hierarchical structure, and learn the level-proportion features. We conducted evaluations with image and text domains, and our quantitative analyses showed competent likelihoods and the best accuracies compared with the baselines.""","""While this was a borderline paper, concerns about the novelty and significance of the presented work exist on the part of all reviewers, and no reviewer was willing to argue for acceptance. Many good points to the work exist, and a stronger case on these issues would greatly strengthen the paper overall. I look forward to a future submission."""
172,"""Active Learning with Partial Feedback""","['Active Learning', 'Learning from Partial Feedback']","""While many active learning papers assume that the learner can simply ask for a label and receive it, real annotation often presents a mismatch between the form of a label (say, one among many classes), and the form of an annotation (typically yes/no binary feedback). To annotate examples corpora for multiclass classification, we might need to ask multiple yes/no questions, exploiting a label hierarchy if one is available. To address this more realistic setting, we propose active learning with partial feedback (ALPF), where the learner must actively choose both which example to label and which binary question to ask. At each step, the learner selects an example, asking if it belongs to a chosen (possibly composite) class. Each answer eliminates some classes, leaving the learner with a partial label. The learner may then either ask more questions about the same example (until an exact label is uncovered) or move on immediately, leaving the first example partially labeled. Active learning with partial labels requires (i) a sampling strategy to choose (example, class) pairs, and (ii) learning from partial labels between rounds. Experiments on Tiny ImageNet demonstrate that our most effective method improves 26% (relative) in top-1 classification accuracy compared to i.i.d. baselines and standard active learners given 30% of the annotation budget that would be required (naively) to annotate the dataset. Moreover, ALPF-learners fully annotate TinyImageNet at 42% lower cost. Surprisingly, we observe that accounting for per-example annotation costs can alter the conventional wisdom that active learners should solicit labels for hard examples.""","""This paper is on active deep learning in the setting where a label hierarchy is available for multiclass classification problems: a fairly natural and pervasive setting. The extension where the learner can ask for example labels as well as a series of questions to adequately descend the label hierarchy is an interesting twist on active learning. The paper is well written and develops several natural formulations which are then benchmarked on CIFAR10, CIFAR100, and Tiny ImageNet using a ResNet-18 architecture. The empirical results are carefully analyzed and appear to set interesting new baselines for active learning. """
173,"""Learning Backpropagation-Free Deep Architectures with Kernels""","['supervised learning', 'backpropagation-free deep architecture', 'kernel method']","""One can substitute each neuron in any neural network with a kernel machine and obtain a counterpart powered by kernel machines. The new network inherits the expressive power and architecture of the original but works in a more intuitive way since each node enjoys the simple interpretation as a hyperplane (in a reproducing kernel Hilbert space). Further, using the kernel multilayer perceptron as an example, we prove that in classification, an optimal representation that minimizes the risk of the network can be characterized for each hidden layer. This result removes the need of backpropagation in learning the model and can be generalized to any feedforward kernel network. Moreover, unlike backpropagation, which turns models into black boxes, the optimal hidden representation enjoys an intuitive geometric interpretation, making the dynamics of learning in a deep kernel network simple to understand. Empirical results are provided to validate our theory.""","""The reviewers mostly raised two concerns regarding the paper: a) why this algorithm is more interpretability than BP (which is just gradient descent); b) the exposition of the paper is somewhat confusing at various places; c) the lack of large-scale experiment results to show this is practically relevant. In the AC's opinion, a principled kernel-based approach can be counted as interpretable, and there the AC would support the paper if a) is the only concern. However, c) seems to be a serious concern since the paper doesn't seem to have experiments beyond fashion MNIST (e.g., CIFAR is pretty easy to train these days) and doesn't have experiments with convolutional models. Based on c), the AC decided that the paper is not quite ready for acceptance. """
174,"""Empirical Study of Easy and Hard Examples in CNN Training""","['easy examples', 'hard example', 'CNN']","""Deep Neural Networks (DNNs) generalize well despite their massive size and capability of memorizing all examples. There is a hypothesis that DNNs start learning from simple patterns based on the observations that are consistently well-classified at early epochs (i.e., easy examples) and examples misclassified (i.e., hard examples). However, despite the importance of understanding the learning dynamics of DNNs, properties of easy and hard examples are not fully investigated. In this paper, we study the similarities of easy and hard examples respectively among different CNNs, assessing those examples contributions to generalization. Our results show that most easy examples are identical among different CNNs, as they share similar dataset-dependent patterns (e.g., colors, structures, and superficial cues in high-frequency). Moreover, while hard examples tend to contribute more to generalization than easy examples, removing a large number of easy examples leads to poor generalization, and we find that most misclassified examples in validation dataset are hard examples. By analyzing intriguing properties of easy and hard examples, we discover that the reason why easy and hard examples have such properties can be explained by biases in a dataset and Stochastic Gradient Descent (SGD).""","""There is no author response for this paper. The paper formulates a definition of easy and hard examples for training a neural network (NN) in terms of their frequency of being classified correctly over several repeats. One repeat corresponds to training the NN from scratch. Top 10% and bottom 10% of the samples with the highest and the lowest frequency define easy and hard instances for training. The authors also compare easy and hard examples across different architectures of NNs. On the positive side, all the reviewers acknowledge the potential usefulness of quantifying easy and hard examples in training NNs, and R1 was ready to improve his/her initial rating if the authors revisited the paper. On the other hand, all the reviewers and AC agreed that the paper requires (1) major improvement in presentation clarity -- see detailed comments of R1 on how to improve as well as comments/questions from R3 and R2; try to avoid confusing terminology such as contradicted patterns. R1 raised important concerns that the proposed notion of easiness is drawn from the experiment in Fig. 1 of Arpit et al (2017) which is not properly attributed. R3 and R2 agreed that in its current state the experimental results are not conclusive and often non informative. To strengthen the paper the reviewers suggested to include more experiments in terms of different datasets, to propose a better metric for defining easy and hard samples (see R3s suggestions). We hope the reviews are useful for improving the paper. """
175,"""Over-parameterization Improves Generalization in the XOR Detection Problem""","['deep learning', 'theory', 'non convex optimization', 'over-parameterization']","""Empirical evidence suggests that neural networks with ReLU activations generalize better with over-parameterization. However, there is currently no theoretical analysis that explains this observation. In this work, we study a simplified learning task with over-parameterized convolutional networks that empirically exhibits the same qualitative phenomenon. For this setting, we provide a theoretical analysis of the optimization and generalization performance of gradient descent. Specifically, we prove data-dependent sample complexity bounds which show that over-parameterization improves the generalization performance of gradient descent.""","""`This paper tackles the problem of learning with one hidden layer non-overlapping conv net for XOR detection problem. For this problem the paper shows that over parametrized models perform better, giving insights into why larger neural networks generalize better - an interesting question to study. However reviews opined that the setting considered in this paper is too specific to this XOR problem and the simplified network architecture, and the techniques are not generalizable to other models. Generalizing these results to more complex architectures or other learning problems will make the paper more interesting. """
176,"""INFORMATION MAXIMIZATION AUTO-ENCODING""","['Information maximization', 'unsupervised learning of hybrid of discrete and continuous representations']","""We propose the Information Maximization Autoencoder (IMAE), an information theoretic approach to simultaneously learn continuous and discrete representations in an unsupervised setting. Unlike the Variational Autoencoder framework, IMAE starts from a stochastic encoder that seeks to map each input data to a hybrid discrete and continuous representation with the objective of maximizing the mutual information between the data and their representations. A decoder is included to approximate the posterior distribution of the data given their representations, where a high fidelity approximation can be achieved by leveraging the informative representations. We show that the proposed objective is theoretically valid and provides a principled framework for understanding the tradeoffs regarding informativeness of each representation factor, disentanglement of representations, and decoding quality. ""","""The paper proposes a principled modeling framework to train a stochastic auto-encoder that is regularized with mutual information maximization. For unsupervised learning, this auto-encoder produces a hybrid continuous-discrete latent representation. While the authors' response and revision have partially addressed some of the raised concerns on the technical analyses, the experimental evaluations presented in the paper do not appear adequate to justify the advantages of the proposed method over previously proposed ones, and the clarity (in particular, notation) needs further improvement. The proposed framework and techniques are potentially of interest to the machine learning community, but the paper of its current form fells below the acceptance bar. The authors are encouraged to improve the clarify of the paper and provide more convincing experiments (e.g., on high-dimensional datasets beyond MNIST)."""
177,"""Complementary-label learning for arbitrary losses and models""","['complementary labels', 'weak supervision']","""In contrast to the standard classification paradigm where the true (or possibly noisy) class is given to each training pattern, complementary-label learning only uses training patterns each equipped with a complementary label. This only specifies one of the classes that the pattern does not belong to. The seminal paper on complementary-label learning proposed an unbiased estimator of the classification risk that can be computed only from complementarily labeled data. How- ever, it required a restrictive condition on the loss functions, making it impossible to use popular losses such as the softmax cross-entropy loss. Recently, another formulation with the softmax cross-entropy loss was proposed with consistency guarantee. However, this formulation does not explicitly involve a risk estimator. Thus model/hyper-parameter selection is not possible by cross-validation we may need additional ordinarily labeled data for validation purposes, which is not available in the current setup. In this paper, we give a novel general framework of complementary-label learning, and derive an unbiased risk estimator for arbitrary losses and models. We further improve the risk estimator by non-negative correction and demonstrate its superiority through experiments.""","""The paper studies learning from complementary labels the setting when example comes with the label information about one of the classes that the example does not belong to. The paper core contribution is an unbiased risk estimator for arbitrary losses and models under this learning scenario, which is an improvement over the previous work, as rightly acknowledged by R1 and R2. The reviewers and AC note the following potential weaknesses: (1) R3 raised an important concern that the core technical contribution is a special case of previously published more general framework which is not cited in the paper. The authors agree with R3 on this matter; (2) the proposed unbiased estimator is not practical, e.g. it leads to overfitting when the cross-entropy loss is used, it is unbounded from below as pointed out by R1; (3) the two proposed modifications of the unbiased estimator are biased estimators, which defeats the motivation of the work and limits its main technical contributions; (4) R2 rightly pointed out that the assumption that complementary label is selected uniformly at random is unrealistic see R2s suggestions on how to address this issue. While all the reviewers acknowledged that the proposed biased estimators show advantageous performance on practice, the AC decides that in its current state the paper does not present significant contributions to the prior work, given (1)-(3), and needs major revision before submitting for another round of reviews. """
178,"""RNNs implicitly implement tensor-product representations""","['tensor-product representations', 'compositionality', 'neural network interpretability', 'recurrent neural networks']","""Recurrent neural networks (RNNs) can learn continuous vector representations of symbolic structures such as sequences and sentences; these representations often exhibit linear regularities (analogies). Such regularities motivate our hypothesis that RNNs that show such regularities implicitly compile symbolic structures into tensor product representations (TPRs; Smolensky, 1990), which additively combine tensor products of vectors representing roles (e.g., sequence positions) and vectors representing fillers (e.g., particular words). To test this hypothesis, we introduce Tensor Product Decomposition Networks (TPDNs), which use TPRs to approximate existing vector representations. We demonstrate using synthetic data that TPDNs can successfully approximate linear and tree-based RNN autoencoder representations, suggesting that these representations exhibit interpretable compositional structure; we explore the settings that lead RNNs to induce such structure-sensitive representations. By contrast, further TPDN experiments show that the representations of four models trained to encode naturally-occurring sentences can be largely approximated with a bag of words, with only marginal improvements from more sophisticated structures. We conclude that TPDNs provide a powerful method for interpreting vector representations, and that standard RNNs can induce compositional sequence representations that are remarkably well approximated byTPRs; at the same time, existing training tasks for sentence representation learning may not be sufficient for inducing robust structural representations""","""AR1 seeks the paper to be more standalone and easier to read. As this comment comes from the reviewer who is very experienced in tensor models, it is highly recommended that the authors make further efforts to make the paper easier to follow. AR2 is concerned about the manually crafted role schemes and alignment discrepancy of results between these schemes and RNNs. To this end, the authors hypothesized further reasons as to why this discrepancy occurs. AC encourages authors to make further efforts to clarify this point without overstating the ability of tensors to model RNNs (it would be interesting to see where these schemes and RNN differ). Lastly, AR3 seeks more clarifications on contributions. While the paper is not ground breaking, it offers some starting point on relating tensors and RNNs. Thus, AC recommends an accept. Kindly note that tensor outer products have been used heavily in computer vision, i.e.: - Higher-Order Occurrence Pooling for Bags-of-Words: Visual Concept Detection by Koniusz et al. (e.g. section 3 considers bi-modal outer tensor product for combining multiple sources: one source can be considered a filter, another as role (similar to Smolensky at al. 1990), e.g. a spatial grid number refining local role of a visual word. This further is extended to multi-modal cases (multiple filter or role modes etc.) ) - Multilinear image analysis for facial recognition (e.g. so called tensor-faces) by Vasilescu et al. - Multilinear independent components analysis by Vasilescu et al. - Tensor decompositions for learning latent variable models by Anandkumar et al. Kindly make connections to these works in your final draft (and to more prior works). """
179,"""Fast Exploration with Simplified Models and Approximately Optimistic Planning in Model Based Reinforcement Learning""","['Reinforcement Learning', 'Strategic Exploration', 'Model Based Reinforcement Learning']","""Humans learn to play video games significantly faster than the state-of-the-art reinforcement learning (RL) algorithms. People seem to build simple models that are easy to learn to support planning and strategic exploration. Inspired by this, we investigate two issues in leveraging model-based RL for sample efficiency. First we investigate how to perform strategic exploration when exact planning is not feasible and empirically show that optimistic Monte Carlo Tree Search outperforms posterior sampling methods. Second we show how to learn simple deterministic models to support fast learning using object representation. We illustrate the benefit of these ideas by introducing a novel algorithm, Strategic Object Oriented Reinforcement Learning (SOORL), that outperforms state-of-the-art algorithms in the game of Pitfall! in less than 50 episodes.""","""Pros: - rather novel approach to using optimistic MCTS for exploration with deterministic models - positive rewards on Pitfall Cons: - lost of domain-specific knowledge - deteministic models - lacking clarity - lacking ablations - no rebuttal I agree with both reviewers that the paper is not good enough to be accepted."""
180,"""PRUNING IN TRAINING: LEARNING AND RANKING SPARSE CONNECTIONS IN DEEP CONVOLUTIONAL NETWORKS""","['Split LBI', 'sparse penalty', 'network pruning', 'feature selection']","""This paper proposes a Pruning in Training (PiT) framework of learning to reduce the parameter size of networks. Different from existing works, our PiT framework employs the sparse penalties to train networks and thus help rank the importance of weights and filters. Our PiT algorithms can directly prune the network without any fine-tuning. The pruned networks can still achieve comparable performance to the original networks. In particular, we introduce the (Group) Lasso-type Penalty (L-P /GL-P), and (Group) Split LBI Penalty (S-P / GS-P) to regularize the networks, and a pruning strategy proposed is used in help prune the network. We conduct the extensive experiments on MNIST, Cifar-10, and miniImageNet. The results validate the efficacy of our proposed methods. Remarkably, on MNIST dataset, our PiT framework can save 17.5% parameter size of LeNet-5, which achieves the 98.47% recognition accuracy.""","""This paper propose to obtain high pruning ratio by adding constraints to obtain small weights. Reviewers have a consensus on rejection due to not convincing experiments and lack of novelty."""
181,"""INVASE: Instance-wise Variable Selection using Neural Networks""","['Instance-wise feature selection', 'interpretability', 'actor-critic methodology']","""The advent of big data brings with it data with more and more dimensions and thus a growing need to be able to efficiently select which features to use for a variety of problems. While global feature selection has been a well-studied problem for quite some time, only recently has the paradigm of instance-wise feature selection been developed. In this paper, we propose a new instance-wise feature selection method, which we term INVASE. INVASE consists of 3 neural networks, a selector network, a predictor network and a baseline network which are used to train the selector network using the actor-critic methodology. Using this methodology, INVASE is capable of flexibly discovering feature subsets of a different size for each instance, which is a key limitation of existing state-of-the-art methods. We demonstrate through a mixture of synthetic and real data experiments that INVASE significantly outperforms state-of-the-art benchmarks.""","""This manuscript proposes a new algorithm for instance-wise feature selection. To this end, the selection is achieved by combining three neural networks trained via an actor-critic methodology. The manuscript highlight that beyond prior work, this strategy enables the selection of a different number of features for each example. Encouraging results are provided on simulated data in comparison to related work, and on real data. The reviewers and AC note issues with the evaluation of the proposed method. In particular, the evaluation of computer vision and natural language processing datasets may have further highlighted the performance of the proposed method. Further, while technically innovative, the approach is closely related to prior work (L2X) -- limiting the novelty. The paper presents a promising new algorithm for training generative adversarial networks. The mathematical foundation for the method is novel and thoroughly motivated, the theoretical results are non-trivial and correct, and the experimental evaluation shows a substantial improvement over the state of the art."""
182,"""Model-Predictive Policy Learning with Uncertainty Regularization for Driving in Dense Traffic""","['model-based reinforcement learning', 'stochastic video prediction', 'autonomous driving']",""" Learning a policy using only observational data is challenging because the distribution of states it induces at execution time may differ from the distribution observed during training. In this work, we propose to train a policy while explicitly penalizing the mismatch between these two distributions over a fixed time horizon. We do this by using a learned model of the environment dynamics which is unrolled for multiple time steps, and training a policy network to minimize a differentiable cost over this rolled-out trajectory. This cost contains two terms: a policy cost which represents the objective the policy seeks to optimize, and an uncertainty cost which represents its divergence from the states it is trained on. We propose to measure this second cost by using the uncertainty of the dynamics model about its own predictions, using recent ideas from uncertainty estimation for deep networks. We evaluate our approach using a large-scale observational dataset of driving behavior recorded from traffic cameras, and show that we are able to learn effective driving policies from purely observational data, with no environment interaction. ""","""Reviewers are in a consensus and recommended to accept after engaging with the authors. Please take reviewers' comments into consideration to improve your submission for the camera ready. """
183,"""Learning what and where to attend""","['Attention models', 'human feature importance', 'object recognition', 'cognitive science']","""Most recent gains in visual recognition have originated from the inclusion of attention mechanisms in deep convolutional networks (DCNs). Because these networks are optimized for object recognition, they learn where to attend using only a weak form of supervision derived from image class labels. Here, we demonstrate the benefit of using stronger supervisory signals by teaching DCNs to attend to image regions that humans deem important for object recognition. We first describe a large-scale online experiment (ClickMe) used to supplement ImageNet with nearly half a million human-derived ""top-down"" attention maps. Using human psychophysics, we confirm that the identified top-down features from ClickMe are more diagnostic than ""bottom-up"" saliency features for rapid image categorization. As a proof of concept, we extend a state-of-the-art attention network and demonstrate that adding ClickMe supervision significantly improves its accuracy and yields visual features that are more interpretable and more similar to those used by human observers.""","""This paper presents a large-scale annotation of human-derived attention maps for ImageNet dataset. This annotation can be used for training more accurate and more interpretable attention models (deep neural networks) for object recognition. All reviewers and AC agree that this work is clearly of interest to ICLR and that extensive empirical evaluations show clear advantages of the proposed approach in terms of improved classification accuracy. In the initial review, R3 put this paper below the acceptance bar requesting major revision of the manuscript and addressing three important weaknesses: (1) no analysis on interpretability; (2) no details about statistical analysis; (3) design choices of the experiments are not motivated. Pleased to report that based on the author respond, the reviewer was convinced that the most crucial concerns have been addressed in the revision. R3 subsequently increased assigned score to 6. As a result, the paper is not in the borderline bucket anymore. The specific recommendation for the authors is therefore to further revise the paper taking into account a better split of the material in the main paper and its appendix. The additional experiments conducted during rebuttal (on interpretability) would be better to include in the main text, as well as explanation regarding statistical analysis. """
184,"""Improving Generative Adversarial Imitation Learning with Non-expert Demonstrations""","['Imitation learning', 'Generative adversarial imitation learning']","""Imitation learning aims to learn an optimal policy from expert demonstrations and its recent combination with deep learning has shown impressive performance. However, collecting a large number of expert demonstrations for deep learning is time-consuming and requires much expert effort. In this paper, we propose a method to improve generative adversarial imitation learning by using additional information from non-expert demonstrations which are easier to obtain. The key idea of our method is to perform multiclass classification to learn discriminator functions where non-expert demonstrations are regarded as being drawn from an extra class. Experiments in continuous control tasks demonstrate that our method learns better policies than the generative adversarial imitation learning baseline when the number of expert demonstrations is small.""","""This paper proposes a variant of GAIL that can learn from both expert and non-expert demonstrations. The paper is generally well-written, and the general topic is of interest to the ICLR community. Further, the empirical comparisons provide some interesting insights. However, the reviewers are concerned that the conceptual contribution is quite small, and that the relatively small conceptual contribution also does not lead to large empirical gains. As such, the paper does not meet the bar for publication at ICLR."""
185,"""LEARNING TO PROPAGATE LABELS: TRANSDUCTIVE PROPAGATION NETWORK FOR FEW-SHOT LEARNING""","['few-shot learning', 'meta-learning', 'label propagation', 'manifold learning']","""The goal of few-shot learning is to learn a classifier that generalizes well even when trained with a limited number of training instances per class. The recently introduced meta-learning approaches tackle this problem by learning a generic classifier across a large number of multiclass classification tasks and generalizing the model to a new task. Yet, even with such meta-learning, the low-data problem in the novel classification task still remains. In this paper, we propose Transductive Propagation Network (TPN), a novel meta-learning framework for transductive inference that classifies the entire test set at once to alleviate the low-data problem. Specifically, we propose to learn to propagate labels from labeled instances to unlabeled test instances, by learning a graph construction module that exploits the manifold structure in the data. TPN jointly learns both the parameters of feature embedding and the graph construction in an end-to-end manner. We validate TPN on multiple benchmark datasets, on which it largely outperforms existing few-shot learning approaches and achieves the state-of-the-art results. ""","""As far as I know, this is the first paper to combine transductive learning with few-shot classification. The proposed algorithm, TPN, combines label propagation with episodic training, as well as learning an adaptive kernel bandwidth in order to determine the label propagation graph. The reviewers liked the idea, however there were concerns of novelty and clarity. I think the contributions of the paper and the strong empirical results are sufficient to merit acceptance, however the paper has not undergone a revision since September. It is therefore recommended that the authors improve the clarity based on the reviewer feedback. In particular, clarifying the details around learning \sigma_i and graph construction. It would also be useful to include the discussion of timing complexity in the final draft."""
186,"""Towards a better understanding of Vector Quantized Autoencoders""","['machine translation', 'vector quantized autoencoders', 'non-autoregressive', 'NMT']","""Deep neural networks with discrete latent variables offer the promise of better symbolic reasoning, and learning abstractions that are more useful to new tasks. There has been a surge in interest in discrete latent variable models, however, despite several recent improvements, the training of discrete latent variable models has remained challenging and their performance has mostly failed to match their continuous counterparts. Recent work on vector quantized autoencoders (VQ-VAE) has made substantial progress in this direction, with its perplexity almost matching that of a VAE on datasets such as CIFAR-10. In this work, we investigate an alternate training technique for VQ-VAE, inspired by its connection to the Expectation Maximization (EM) algorithm. Training the discrete autoencoder with EM and combining it with sequence level knowledge distillation alows us to develop a non-autoregressive machine translation model whose accuracy almost matches a strong greedy autoregressive baseline Transformer, while being 3.3 times faster at inference. ""","""Strengths: - well-written - strong results for non-autoregressive NMT - a novel soft EM version of VQ-VAE Weaknesses: - as pointed out by reviewers, the improvements are mostly not due to the VQ-VAE modification rather due to orthogonal (and not interesting) changes e.g., knowledge distillation. If there is a genuine contribution of VQ-VAE, it is small and required extensive parameter selection - the explanations provided in the paper do not match the empirical results Two reviewers criticize the experiments / experimental section: rigour / their discussion. Overall, there is nothing wrong with the method but the experiments are not showing that the modification is particularly beneficial. Given these results and also given that the method is not particularly novel (switching from EM to Soft EM in VQ-VAE), it is hard for me to argue for accepting the paper."""
187,"""Understanding GANs via Generalization Analysis for Disconnected Support""","['Generalization analysis', 'Statistical estimation', 'Understanding GANs', 'Disconnected support']","""This paper provides theoretical analysis of generative adversarial networks (GANs) to explain its advantages over other standard methods of learning probability measures. GANs learn a probability through observations, using the objective function with a generator and a discriminator. While many empirical results indicate that GANs can generate realistic samples, the reason for such successful performance remains unelucidated. This paper focuses the situation where the target probability measure satisfies the disconnected support property, which means a separate support of a probability, and relates it with the advantage of GANs. It is theoretically shown that, unlike other popular models, GANs do not suffer from the decrease of generalization performance caused by the disconnected support property. We rigorously quantify the generalization performance of GANs of a given architecture, and compare it with the performance of the other models. Based on the theory, we also provide a guideline for selecting deep network architecture for GANs. We demonstrate some numerical examples which support our results.""","""This paper provides a theoretical analysis of GANs, showing its advantages when the measure satisfies the disconnected support property. Its main theoretical results are interesting, but the reviews and discussion shows some misleading places. It was also found some of the claims and proof are not mathematically rigorous. """
188,"""Rethinking the Value of Network Pruning""","['network pruning', 'network compression', 'architecture search', 'train from scratch']","""Network pruning is widely used for reducing the heavy inference cost of deep models in low-resource settings. A typical pruning algorithm is a three-stage pipeline, i.e., training (a large model), pruning and fine-tuning. During pruning, according to a certain criterion, redundant weights are pruned and important weights are kept to best preserve the accuracy. In this work, we make several surprising observations which contradict common beliefs. For all state-of-the-art structured pruning algorithms we examined, fine-tuning a pruned model only gives comparable or worse performance than training that model with randomly initialized weights. For pruning algorithms which assume a predefined target network architecture, one can get rid of the full pipeline and directly train the target network from scratch. Our observations are consistent for multiple network architectures, datasets, and tasks, which imply that: 1) training a large, over-parameterized model is often not necessary to obtain an efficient final model, 2) learned ``important'' weights of the large model are typically not useful for the small pruned model, 3) the pruned architecture itself, rather than a set of inherited ``important'' weights, is more crucial to the efficiency in the final model, which suggests that in some cases pruning can be useful as an architecture search paradigm. Our results suggest the need for more careful baseline evaluations in future research on structured pruning methods. We also compare with the ""Lottery Ticket Hypothesis"" (Frankle & Carbin 2019), and find that with optimal learning rate, the ""winning ticket"" initialization as used in Frankle & Carbin (2019) does not bring improvement over random initialization.""","""The paper presents a lot of empirical evidence that fine tuning pruned networks is inferior to training them from scratch. These results seem unsurprising in retrospect, but hindsight is 20-20. The reviewers raised a wide range of issues, some of which were addressed and some which were not. I recommend to the authors that they make sure that any claims they draw from their experiments are sufficiently prescribed. E.g., the lottery ticket experiments done by Anonymous in response to this paper show that the random initialization does poorer than restarting with the initial weights (other than in resnet, though this seems possibly due to the learning rate). There is something different in their setting, and so your claims should be properly circumscribed. I don't think the ""standard"" versus ""nonstandard"" terminology is appropriate until the actual boundary between these two behaviors is identified. I would recommend the authors make guarded claims here."""
189,"""Graph Generation via Scattering""","['graph generative neural network', 'link prediction', 'graph and signal generation', 'scattering network']","""Generative networks have made it possible to generate meaningful signals such as images and texts from simple noise. Recently, generative methods based on GAN and VAE were developed for graphs and graph signals. However, the mathematical properties of these methods are unclear, and training good generative models is difficult. This work proposes a graph generation model that uses a recent adaptation of Mallat's scattering transform to graphs. The proposed model is naturally composed of an encoder and a decoder. The encoder is a Gaussianized graph scattering transform, which is robust to signal and graph manipulation. The decoder is a simple fully connected network that is adapted to specific tasks, such as link prediction, signal generation on graphs and full graph and signal generation. The training of our proposed system is efficient since it is only applied to the decoder and the hardware requirement is moderate. Numerical results demonstrate state-of-the-art performance of the proposed system for both link prediction and graph and signal generation. These results are in contrast to experience with Euclidean data, where it is difficult to form a generative scattering network that performs as well as state-of-the-art methods. We believe that this is because of the discrete and simpler nature of graph applications, unlike the more complex and high-frequency nature of Euclidean data, in particular, of some natural images. ""","""AR1 is concerned about the novelty and what are exact novel elements of the proposed approach. AR2 is worried about the novelty (combination of existing blocks) and lack of insights. AR3 is also concerned about the novelty, complexity and poor evaluations/lack of thorough comparisons with other baselines. After rebuttal, the reviewers remained unconvinced e.g. AR3 still would like to see why the proposed method would be any better than GAN-based approaches. With regret, at this point, the AC cannot accept this paper but AC encourages the authors to take all reviews into consideration and improve their manuscript accordingly. Matters such as complexity (perhaps scattering networks aren't the most friendly here), clear insights and strong comparisons to generative approaches are needed."""
190,"""FAST OBJECT LOCALIZATION VIA SENSITIVITY ANALYSIS""","['Internal Representations', 'Sensitivity Analysis', 'Object Detection']","""Deep Convolutional Neural Networks (CNNs) have been repeatedly shown to perform well on image classification tasks, successfully recognizing a broad array of objects when given sufficient training data. Methods for object localization, however, are still in need of substantial improvement. Common approaches to this problem involve the use of a sliding window, sometimes at multiple scales, providing input to a deep CNN trained to classify the contents of the window. In general, these approaches are time consuming, requiring many classification calculations. In this paper, we offer a fundamentally different approach to the localization of recognized objects in images. Our method is predicated on the idea that a deep CNN capable of recognizing an object must implicitly contain knowledge about object location in its connection weights. We provide a simple method to interpret classifier weights in the context of individual classified images. This method involves the calculation of the derivative of network generated activation patterns, such as the activation of output class label units, with regard to each in- put pixel, performing a sensitivity analysis that identifies the pixels that, in a local sense, have the greatest influence on internal representations and object recognition. These derivatives can be efficiently computed using a single backward pass through the deep CNN classifier, producing a sensitivity map of the image. We demonstrate that a simple linear mapping can be learned from sensitivity maps to bounding box coordinates, localizing the recognized object. Our experimental results, using real-world data sets for which ground truth localization information is known, reveal competitive accuracy from our fast technique.""","""This paper was reviewed by three experts. After the author response, R2 and R3 recommend rejecting this paper citing concerns of novelty and experimental evaluation. R1 assigns it a score of ""6"" but in comments agrees that the manuscript is not ready for ICLR. The AC finds no basis for accepting this paper in this state. """
191,"""Policy Generalization In Capacity-Limited Reinforcement Learning""","['reinforcement learning', 'generalization', 'capacity constraints', 'information theory']","""Motivated by the study of generalization in biological intelligence, we examine reinforcement learning (RL) in settings where there are information-theoretic constraints placed on the learner's ability to represent a behavioral policy. We first show that the problem of optimizing expected utility within capacity-limited learning agents maps naturally to the mathematical field of rate-distortion (RD) theory. Applying the RD framework to the RL setting, we develop a new online RL algorithm, Capacity-Limited Actor-Critic, that learns a policy that optimizes a tradeoff between utility maximization and information processing costs. Using this algorithm in a 2D gridworld environment, we demonstrate two novel empirical results. First, at high information rates (high channel capacity), the algorithm achieves faster learning and discovers better policies compared to the standard tabular actor-critic algorithm. Second, we demonstrate that agents with capacity-limited policy representations avoid 'overfitting' and exhibit superior transfer to modified environments, compared to policies learned by agents with unlimited information processing resources. Our work provides a principled framework for the development of computationally rational RL agents.""","""The paper studies RL from a rate-distortion (RD) theory perspective. A new actor-critic algorithm is developed and evaluated on a series of 2D grid worlds. The paper has some novel idea, and the connection of RL to RD is quite new. This seems like an interesting direction that is worth further investigation. On the other hand, all reviewers agreed there is a severe flaw in this work, casting a doubt where RD can be directly applied to an RL setting because the distribution is not fixed (unlike in standard RD). This issue could have been addressed empirically, by running controlled experiments, something the the paper might include in a future version."""
192,"""Differentiable Perturb-and-Parse: Semi-Supervised Parsing with a Structured Variational Autoencoder""","['differentiable dynamic programming', 'variational auto-encoder', 'dependency parsing', 'semi-supervised learning']","""Human annotation for syntactic parsing is expensive, and large resources are available only for a fraction of languages. A question we ask is whether one can leverage abundant unlabeled texts to improve syntactic parsers, beyond just using the texts to obtain more generalisable lexical features (i.e. beyond word embeddings). To this end, we propose a novel latent-variable generative model for semi-supervised syntactic dependency parsing. As exact inference is intractable, we introduce a differentiable relaxation to obtain approximate samples and compute gradients with respect to the parser parameters. Our method (Differentiable Perturb-and-Parse) relies on differentiable dynamic programming over stochastically perturbed edge scores. We demonstrate effectiveness of our approach with experiments on English, French and Swedish.""","""This paper proposes a method for unsupervised learning that uses a latent variable generative model for semi-supervised dependency parsing. The key learning method consists of making perturbations to the logits going into a parsing algorithm, to make it possible to sample within the variational auto-encoder framework. Significant gains are found through semi-supervised learning. The largest reviewer concern was that the baselines were potentially not strong enough, as significantly better numbers have been reported in previous work, which may have a result of over-stating the perceived utility. Overall though it seems that the reviewers appreciated the novel solution to an important problem, and in general would like to see the paper accepted."""
193,"""A NOVEL VARIATIONAL FAMILY FOR HIDDEN NON-LINEAR MARKOV MODELS""","['variational inference', 'time series', 'nonlinear dynamics', 'neuroscience']","""Latent variable models have been widely applied for the analysis and visualization of large datasets. In the case of sequential data, closed-form inference is possible when the transition and observation functions are linear. However, approximate inference techniques are usually necessary when dealing with nonlinear evolution and observations. Here, we propose a novel variational inference framework for the explicit modeling of time series, Variational Inference for Nonlinear Dynamics (VIND), that is able to uncover nonlinear observation and latent dynamics from sequential data. The framework includes a structured approximate posterior, and an algorithm that relies on the fixed-point iteration method to find the best estimate for latent trajectories. We apply the method to several datasets and show that it is able to accurately infer the underlying dynamics of these systems, in some cases substantially outperforming state-of-the-art methods.""","""The reviewers in general like the paper but has serous reservations regarding relation to other work (novelty) and clarity of presentation. Given non-linear state space models is a crowded field it is perhaps better that these points are dealt with first and then submitted elsewhere."""
194,"""Optimal Transport Maps For Distribution Preserving Operations on Latent Spaces of Generative Models""","['generative models', 'optimal transport', 'distribution preserving operations']","""Generative models such as Variational Auto Encoders (VAEs) and Generative Adversarial Networks (GANs) are typically trained for a fixed prior distribution in the latent space, such as uniform or Gaussian. After a trained model is obtained, one can sample the Generator in various forms for exploration and understanding, such as interpolating between two samples, sampling in the vicinity of a sample or exploring differences between a pair of samples applied to a third sample. However, the latent space operations commonly used in the literature so far induce a distribution mismatch between the resulting outputs and the prior distribution the model was trained on. Previous works have attempted to reduce this mismatch with heuristic modification to the operations or by changing the latent distribution and re-training models. In this paper, we propose a framework for modifying the latent space operations such that the distribution mismatch is fully eliminated. Our approach is based on optimal transport maps, which adapt the latent space operations such that they fully match the prior distribution, while minimally modifying the original operation. Our matched operations are readily obtained for the commonly used operations and distributions and require no adjustment to the training procedure.""","""This is a well-written paper that shows how to use optimal transport to perform smooth interpolation, between two random vectors sampled from the prior distribution of the latent space of a deep generative model. By encouraging the marginal of the interpolated vector to match the prior distribution, these interpolated distribution-preserving random vectors in the latent space are shown to result in better image interpolation quality for GANs. The problem is of interest to the community and the resulted solutions are simple to implement. As pointed out by Reviewer 1, the paper could be made clearly more convincing by showing that these distribution preservation operations also help perform interpolation in the latent space of VAEs, and the AC strongly encourages the authors to add these results if possible. The AC appreciates that the authors have added experiments to satisfactorily address his/her concern: ""Suppose z_1,z_2 are independent, and drawn from N(\mu,\Sigma), then t z_1 + (1-t)z_2 ~ N(\mu, (t^2+(1-t)^2)\Sigma). If one lets y | z_1, z_2 ~ N(t z_1 + (1-t)z_2, (1-t^2-(1-t)^2)\Sigma) as the latent space interpolation, then marginally we have y ~ N(\mu, \Sigma). This is an extremely simple and fast procedure to make sure that the latent space interpolation y is highly related to the linear interpolation t z_1 + (1-t)z_2 but also satisfies y ~ N(\mu, \Sigma)."" The AC strongly encourages the authors to add these new results into their revision, and highlight ""smooth interpolation"" as an important characteristic in addition to ""distribution preserving."" A potential suggestion is changing ""Distribution Preserving Operations"" in the title to ""Distribution Preserving Smooth Operations."" """
195,"""Uncovering Surprising Behaviors in Reinforcement Learning via Worst-case Analysis""","['Reinforcement learning', 'Adversarial examples', 'Navigation', 'Evaluation', 'Analysis']","""Reinforcement learning agents are typically trained and evaluated according to their performance averaged over some distribution of environment settings. But does the distribution over environment settings contain important biases, and do these lead to agents that fail in certain cases despite high average-case performance? In this work, we consider worst-case analysis of agents over environment settings in order to detect whether there are directions in which agents may have failed to generalize. Specifically, we consider a 3D first-person task where agents must navigate procedurally generated mazes, and where reinforcement learning agents have recently achieved human-level average-case performance. By optimizing over the structure of mazes, we find that agents can suffer from catastrophic failures, failing to find the goal even on surprisingly simple mazes, despite their impressive average-case performance. Additionally, we find that these failures transfer between different agents and even significantly different architectures. We believe our findings highlight an important role for worst-case analysis in identifying whether there are directions in which agents have failed to generalize. Our hope is that the ability to automatically identify failures of generalization will facilitate development of more general and robust agents. To this end, we report initial results on enriching training with settings causing failure.""","""The paper presents adversarial ""attacks"" to maze generation for RL agents trained to perform 2D navigation tasks in 3D environments (DM Lab). The paper is well written, and the rebuttal(s) and additional experiments (section 4) make the paper better. The approach itself is very interesting. However, there are a few limitations, and thus I am very borderline on this submission: - the analysis of why and how the navigation trained models fail, is rather succinct. Analyzing what happens on the model side (not just the features of the adversarial mazes vs. training mazes) would make the paper stronger. - (more importantly) Section 4: ""adapting the training distribution"" by incorporating adversarial mazes into training feels incomplete. That is a pithy as giving an adversarial attack for RL trained navigation agents would be much more complete of a contribution if at least the most obvious way to defend the attack was studied in depth. The authors themselves are honest about it and write ""Therefore, it is possible that many more training iterations are necessary for agents to learn to perform well in each adversarial setting."" (under 4.4 / Expensive Training). I would invite the authors to submit this version to the workshop track, and/or to finish the work started in Section 4 and make it a strong paper."""
196,"""Towards Metamerism via Foveated Style Transfer""","['Metamerism', 'foveation', 'perception', 'style transfer', 'psychophysics']","""The problem of visual metamerism is defined as finding a family of perceptually indistinguishable, yet physically different images. In this paper, we propose our NeuroFovea metamer model, a foveated generative model that is based on a mixture of peripheral representations and style transfer forward-pass algorithms. Our gradient-descent free model is parametrized by a foveated VGG19 encoder-decoder which allows us to encode images in high dimensional space and interpolate between the content and texture information with adaptive instance normalization anywhere in the visual field. Our contributions include: 1) A framework for computing metamers that resembles a noisy communication system via a foveated feed-forward encoder-decoder network We observe that metamerism arises as a byproduct of noisy perturbations that partially lie in the perceptual null space; 2) A perceptual optimization scheme as a solution to the hyperparametric nature of our metamer model that requires tuning of the image-texture tradeoff coefficients everywhere in the visual field which are a consequence of internal noise; 3) An ABX psychophysical evaluation of our metamers where we also find that the rate of growth of the receptive fields in our model match V1 for reference metamers and V2 between synthesized samples. Our model also renders metamers at roughly a second, presenting a 1000 speed-up compared to the previous work, which now allows for tractable data-driven metamer experiments.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The problem is well-motivated and related work is thoroughly discussed - The evaluation is compelling and extensive. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - Very dense. Clarity could be improved in some sections. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. No major points of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be accepted. """
197,"""Single Shot Neural Architecture Search Via Direct Sparse Optimization""","['Neural Architecture Search', 'Sparse Optimization']","""Recently Neural Architecture Search (NAS) has aroused great interest in both academia and industry, however it remains challenging because of its huge and non-continuous search space. Instead of applying evolutionary algorithm or reinforcement learning as previous works, this paper proposes a Direct Sparse Optimization NAS (DSO-NAS) method. In DSO-NAS, we provide a novel model pruning view to NAS problem. In specific, we start from a completely connected block, and then introduce scaling factors to scale the information flow between operations. Next, we impose sparse regularizations to prune useless connections in the architecture. Lastly, we derive an efficient and theoretically sound optimization method to solve it. Our method enjoys both advantages of differentiability and efficiency, therefore can be directly applied to large datasets like ImageNet. Particularly, On CIFAR-10 dataset, DSO-NAS achieves an average test error 2.84%, while on the ImageNet dataset DSO-NAS achieves 25.4% test error under 600M FLOPs with 8 GPUs in 18 hours.""","""This paper proposes Direct Sparse Optimization (DSO)-NAS to obtain neural architectures on specific problems at a reasonable computational cost. Regularization by sparsity is a neat idea, but similar idea has been discussed by many pruning papers. ""model pruning formulation for neural architecture search based on sparse optimization"" is claimed to be the main contribution, but it's debatable if such contribution is strong: worse accuracy, more computation, more #parameters than Mnas (less search time, but also worse search quality). The effect of each proposed technique is appropriately evaluated. However, the reviewers are concerned that the proposed method does not outperform the existing state-of-the-art methods in terms of classification accuracy. There's also some concerns about the search space of the proposed method. It is debatable about claim that ""the first NAS algorithm to perform direct search on ImageNet"" and ""the first method to perform direct search without block structure sharing"". Given the acceptance rate of ICLR should be <30%, I would say this paper is good but not outstanding. """
198,"""Filter Training and Maximum Response: Classification via Discerning""","['filter training', 'maximum response', 'multiple check', 'ensemble learning']","""This report introduces a training and recognition scheme, in which classification is realized via class-wise discerning. Trained with datasets whose labels are randomly shuffled except for one class of interest, a neural network learns class-wise parameter values, and remolds itself from a feature sorter into feature filters, each of which discerns objects belonging to one of the classes only. Classification of an input can be inferred from the maximum response of the filters. A multiple check with multiple versions of filters can diminish fluctuation and yields better performance. This scheme of discerning, maximum response and multiple check is a method of general viability to improve performance of feedforward networks, and the filter training itself is a promising feature abstraction procedure. In contrast to the direct sorting, the scheme mimics the classification process mediated by a series of one component picking.""","""This work examines how to deal with multiple classes. Unfortunately, as reviewers note, it fails to adequately ground its approach in previous work and show how the architecture relates to the considerable research that has examined the question beforehand."""
199,"""Modeling Uncertainty with Hedged Instance Embeddings""","['uncertainty', 'instance embedding', 'metric learning', 'probabilistic embedding']","""Instance embeddings are an efficient and versatile image representation that facilitates applications like recognition, verification, retrieval, and clustering. Many metric learning methods represent the input as a single point in the embedding space. Often the distance between points is used as a proxy for match confidence. However, this can fail to represent uncertainty which can arise when the input is ambiguous, e.g., due to occlusion or blurriness. This work addresses this issue and explicitly models the uncertainty by hedging the location of each input in the embedding space. We introduce the hedged instance embedding (HIB) in which embeddings are modeled as random variables and the model is trained under the variational information bottleneck principle (Alemi et al., 2016; Achille & Soatto, 2018). Empirical results on our new N-digit MNIST dataset show that our method leads to the desired behavior of hedging its bets across the embedding space upon encountering ambiguous inputs. This results in improved performance for image matching and classification tasks, more structure in the learned embedding space, and an ability to compute a per-exemplar uncertainty measure which is correlated with downstream performance.""","""This work presents a method to model embeddings as distributions, instead of points, to better quantify uncertainty. Evaluations are carried out on a new dataset created from mixtures of MNIST digits, including noise (certain probability of occlusions), that introduce ambiguity, using a small ""toy"" neural network that is incapable of perfectly fitting the data, because authors mention that performance difference lessens when the network is complex enough to almost perfectly fit the data. Reviewer assessment is unanimously accept, with the following points: Pros: + ""The topic of injecting uncertainty in neural networks should be of broad interest to the ICLR community."" + ""The paper is generally clear."" + ""The qualitative evaluation provides intuitive results."" Cons: - Requirement of drawing samples may add complexity. Authors reply that alternatives should be studied in future work. - No comparison to other uncertainty methods, such as dropout. Authors reply that dropout represents model uncertainty and not data uncertainty, but do not carry out an experiment to compare (i.e. sample from model leaving dropout activated during evaluation). - No evaluation in larger scale/dimensionality datasets. Authors mention method scales linearly, but how practical or effective this method is to use on, say, face recognition datasets, is unclear. As the general reviewer consensus is accept, Area Chair is recommending Accept; However, Area Chair has strong reservations because the method is evaluated on a very limited dataset, with a toy model designed to exaggerate differences between techniques. Essentially, the toy evaluation was designed to get the results the authors were looking for. A more thorough investigation would use more realistic sized network models on true datasets. """
200,"""Conscious Inference for Object Detection""","['consciousness', 'conscious inference', 'object detection', 'object pose estimation']","""Current Convolutional Neural Network (CNN)-based object detection models adopt strictly feedforward inference to predict the final detection results. However, the widely used one-way inference is agnostic to the global image context and the interplay between input image and task semantics. In this work, we present a general technique to improve off-the-shelf CNN-based object detection models in the inference stage without re-training, architecture modification or ground-truth requirements. We propose an iterative, bottom-up and top-down inference mechanism, which is named conscious inference, as it is inspired by prevalent models for human consciousness with top-down guidance and temporal persistence. While the downstream pass accumulates category-specific evidence over time, it subsequently affects the proposal calculation and the final detection. Feature activations are updated in line with no additional memory cost. Our approach advances the state of the art using popular detection models (Faster-RCNN, YOLOv2, YOLOv3) on 2D object detection and 6D object pose estimation.""","""The paper presents an interesting idea, but there are significant concerns about the presentation issues and experimental results (e.g., comparisons with baselines). Overall, it is not ready for publication. """
201,"""Differential Equation Networks""","['deep learning', 'activation function', 'differential equations']","""Most deep neural networks use simple, fixed activation functions, such as sigmoids or rectified linear units, regardless of domain or network structure. We introduce differential equation networks, an improvement to modern neural networks in which each neuron learns the particular nonlinear activation function that it requires. We show that enabling each neuron with the ability to learn its own activation function results in a more compact network capable of achieving comperable, if not superior performance when compared to much larger networks. We also showcase the capability of a differential equation neuron to learn behaviors, such as oscillation, currently only obtainable by a large group of neurons. The ability of differential equation networks to essentially compress a large neural network, without loss of overall performance makes them suitable for on-device applications, where predictions must be computed locally. Our experimental evaluation of real-world and toy datasets show that differential equation networks outperform fixed activatoin networks in several areas.""","""The reviewers unanimously agreed the paper did not meet the bar of acceptance for ICLR. They raised questions around the technical correctness of the paper, as well as the experimental setup. The authors did not address any reviewer concerns, or provide any response. Therefore, I recommend rejection."""
202,"""Human-Guided Column Networks: Augmenting Deep Learning with Advice""","['Knowledge-guided learning', 'Human advice', 'Column Networks', 'Knowledge-based relational deep model', 'Collective classification']","""While extremely successful in several applications, especially with low-level representations; sparse, noisy samples and structured domains (with multiple objects and interactions) are some of the open challenges in most deep models. Column Networks, a deep architecture, can succinctly capture such domain structure and interactions, but may still be prone to sub-optimal learning from sparse and noisy samples. Inspired by the success of human-advice guided learning in AI, especially in data-scarce domains, we propose Knowledge-augmented Column Networks that leverage human advice/knowledge for better learning with noisy/sparse samples. Our experiments demonstrate how our approach leads to either superior overall performance or faster convergence.""","""The paper considers the task of incorporating knowledge expressed as rules into column networks. The reviewers acknowledge the need for such techniques, like the flexibility of the proposed approach, and appreciate the improvements to convergence speed and accuracy afforded by the proposed work. The reviewers and the AC note the following as the primary concerns of the paper: (1) The primary concerned raised by the reviewers was that the evaluation is focused on whether KCLN can beat one with the knowledge, instead of measuring the efficacy of incorporating the knowledge itself (e.g. by comparing with other forms of incorporating knowledge, or by varying the quality of the rules that were introduced), (2) Even otherwise, the empirical results are not significant, offering slight improvements over the vanilla CLN (reviewer 1), (3) There are concerns that the rule-based gates are introduced but gradients are only computed on the final layer, which might lead to instability, and (4) There are a number of issues in the presentation, where the space is used on redundant information and description of datasets, instead of focusing on the proposed model. The comments by the authors address some of these concerns, in particular, clarifying that the forms of knowledge/rules are not limited, however, they focused on simple rules in the paper. However, the primary concerns in the evaluation still remain: (1) it seems to focus on comparing against Vanilla-CLN, instead of focusing on the source of the knowledge, or on the efficacy in incorporating it (see earlier work on examples of how to evaluate these), and (2) the results are not considerably better with the proposed work, making the reviewers doubtful about the significance of the proposed work. The reviewers agree that the paper is not ready for publication."""
203,"""Adaptive Estimators Show Information Compression in Deep Neural Networks""","['deep neural networks', 'mutual information', 'information bottleneck', 'noise', 'L2 regularization']","""To improve how neural networks function it is crucial to understand their learning process. The information bottleneck theory of deep learning proposes that neural networks achieve good generalization by compressing their representations to disregard information that is not relevant to the task. However, empirical evidence for this theory is conflicting, as compression was only observed when networks used saturating activation functions. In contrast, networks with non-saturating activation functions achieved comparable levels of task performance but did not show compression. In this paper we developed more robust mutual information estimation techniques, that adapt to hidden activity of neural networks and produce more sensitive measurements of activations from all functions, especially unbounded functions. Using these adaptive estimation techniques, we explored compression in networks with a range of different activation functions. With two improved methods of estimation, firstly, we show that saturation of the activation function is not required for compression, and the amount of compression varies between different activation functions. We also find that there is a large amount of variation in compression between different network initializations. Secondary, we see that L2 regularization leads to significantly increased compression, while preventing overfitting. Finally, we show that only compression of the last layer is positively correlated with generalization.""","""This paper suggests that noise-regularized estimators of mutual information in deep neural networks should be adaptive, in the sense that the variance of the regularization noise should be proportional to the range of the hidden activity. Two adaptive estimators are proposed: (1) an entropy-based adaptive binning (EBAB) estimator that chooses the bin boundaries such that each bin contains the same number of unique observed activation levels, and (2) an adaptive kernel density estimator (aKDE) that adds isotropic Gaussian noise, where the variance of the noise is proportional to the maximum activity value in a given layer. These estimators are then used to show that (1) ReLU networks can compress, but that compression may or may not occur depending on the specific weight initialization; (2) different nonsaturating noninearities exhibit different information plane behaviors over the course of training; and (3) L2 regularization in ReLU networks encourages compression. The paper also finds that only compression in the last (softmax) layer correlates with generalization performance. The reviewers liked the range of experiments and found the observations in the paper interesting, but had reservations about the lack of rigor in the paper (no theoretical analysis of the convergence of the proposed estimator), were worried that post-hoc addition of noise distorts the function of the network, and felt that there wasn't much insight provided on the cause of compression in deep neural networks. The AC shares these concerns, and considers them to be more significant than the reviewers do, but doesn't wish to override the reviewers' recommendation that the paper be accepted."""
204,"""Meta-Learning with Domain Adaptation for Few-Shot Learning under Domain Shift""","['Meta-Learning', 'Few-Shot Learning', 'Domain Adaptation']","""Few-Shot Learning (learning with limited labeled data) aims to overcome the limitations of traditional machine learning approaches which require thousands of labeled examples to train an effective model. Considered as a hallmark of human intelligence, the community has recently witnessed several contributions on this topic, in particular through meta-learning, where a model learns how to learn an effective model for few-shot learning. The main idea is to acquire prior knowledge from a set of training tasks, which is then used to perform (few-shot) test tasks. Most existing work assumes that both training and test tasks are drawn from the same distribution, and a large amount of labeled data is available in the training tasks. This is a very strong assumption which restricts the usage of meta-learning strategies in the real world where ample training tasks following the same distribution as test tasks may not be available. In this paper, we propose a novel meta-learning paradigm wherein a few-shot learning model is learnt, which simultaneously overcomes domain shift between the train and test tasks via adversarial domain adaptation. We demonstrate the efficacy the proposed method through extensive experiments.""","""This paper addresses the problem of few shot learning and then domain transfer. The proposed approach consists of combining a known few shot learning model, prototypical nets, together with image to image translation via CycleGAN for domain adaptation. Thus the algorithmic novelty is minor and amounts to combining two techniques to address a different problem statement. In addition, as mentioned by Reviewer 2, though meta learning could be a solution to learn with few examples, the solution being used in this work is not meta learning and so should not be in the title to avoid confusion. As this is a new problem statement the authors apply multiple existing works from few shot learning (and now adaptation) to their setting. The proposed approach does outperform prior work, however this is not surprising as the prior work was not designed for this task. Despite improvements during the rebuttal to address clarity the specific experimental setting is still unclear -- especially the setup of meta test data vs unsupervised da data. This paper is borderline. However, since the main contribution consists of proposing a new problem statement and suggesting a combination of prior techniques as a first solution, the paper needs a more thorough ablation of other possible combination of techniques as well as a clearly defined experimental setup before it is ready for publication."""
205,"""PeerNets: Exploiting Peer Wisdom Against Adversarial Attacks""","['peernet', 'peernets', 'graph', 'geometric deep learning', 'adversarial', 'perturbation', 'defense', 'peer regularization']","""Deep learning systems have become ubiquitous in many aspects of our lives. Unfortunately, it has been shown that such systems are vulnerable to adversarial attacks, making them prone to potential unlawful uses. Designing deep neural networks that are robust to adversarial attacks is a fundamental step in making such systems safer and deployable in a broader variety of applications (e.g. autonomous driving), but more importantly is a necessary step to design novel and more advanced architectures built on new computational paradigms rather than marginally building on the existing ones. In this paper we introduce PeerNets, a novel family of convolutional networks alternating classical Euclidean convolutions with graph convolutions to harness information from a graph of peer samples. This results in a form of non-local forward propagation in the model, where latent features are conditioned on the global structure induced by the graph, that is up to 3 times more robust to a variety of white- and black-box adversarial attacks compared to conventional architectures with almost no drop in accuracy.""","""The paper presents a novel with compelling experiments. Good paper, accept. """
206,"""Detecting Adversarial Examples Via Neural Fingerprinting""","['Adversarial Attacks', 'Deep Neural Networks']","""Deep neural networks are vulnerable to adversarial examples: input data that has been manipulated to cause dramatic model output errors. To defend against such attacks, we propose NeuralFingerprinting: a simple, yet effective method to detect adversarial examples that verifies whether model behavior is consistent with a set of fingerprints. These fingerprints are encoded into the model response during training and are inspired by the use of biometric and cryptographic signatures. In contrast to previous defenses, our method does not rely on knowledge of the adversary and can scale to large networks and input data. The benefits of our method are that 1) it is fast, 2) it is prohibitively expensive for an attacker to reverse-engineer which fingerprints were used, and 3) it does not assume knowledge of the adversary. In this work, we 1) theoretically analyze NeuralFingerprinting for linear models and 2) show that NeuralFingerprinting significantly improves on state-of-the-art detection mechanisms for deep neural networks, by detecting the strongest known adversarial attacks with 98-100% AUC-ROC scores on the MNIST, CIFAR-10 and MiniImagenet (20 classes) datasets. In particular, we consider several threat models, including the most conservative one in which the attacker has full knowledge of the defender's strategy. In all settings, the detection accuracy of NeuralFingerprinting generalizes well to unseen test-data and is robust over a wide range of hyperparameters.""","""* Strengths The paper proposes a novel and interesting method for detecting adversarial examples, which has the advantage of being based on general fingerprint statistics of a model and is not restricted to any specific threat model (in contrast to much of the work in the area which is restricted to adversarial examples in some L_p norm ball). The writing is clear and the experiments are extensive. * Weaknesses The experiments are thorough. However, they contain a subtle but important flaw. During discussion it was revealed that the attacks used to evaluate the method fail to reduce accuracy even at large values of epsilon where there are simple adversarial attacks that should reduce the accuracy to zero. This casts doubt on whether the attacks at small values of epsilon really are providing a good measure of the methods robustness. * Discussion There was substantial disagreement about the paper, with R1 feeling that the evaluation issues were serious enough to merit rejection and R3 feeling that they were not a large issue. In discussion with me, both R1 and R3 agreed that if an attack were demonstrated to break the method, that would be grounds for rejection. They also both agreed that there probably is an attack that breaks the method. A potential key difference is that R3 thinks this might be quite difficult to find and so merits publishing the paper to motivate stronger attacks. I ultimately agree with R1 that the evaluation issues are indeed serious. One reason for this is that there is by now a long record of adversarial defense papers posting impressive numbers that are often invalidated within a short period (often less than a month or so) of the paper being published. The Obfuscated Gradients paper of Athalye, Carlini, and Wagner suggests several basic sanity checks to help avoid this. One of the sanity checks (which the present paper fails) is to test that attacks work when epsilon is large. This is not an arbitrary test but gets at a key issue---any given attack provides only an *upper bound* on the worst-case accuracy of a method. For instance, if an attack only brings the accuracy of a method down to 80% at epsilon=1 (when we know the true accuracy should be 0%), then at epsilon=0.01 we know that the measured accuracy of the attack comes 80% from the over-optimistic accuracy at epsilon=1 and at most 20% from the true accuracy at epsilon=0.01. If the measured accuracy at epsilon=1 is close to 100%, then accuracy at lower values of epsilon provides basically no information. This means that the experiments as currently performed give no information about the true accuracy of the method, which is a serious issue that the authors should address before the paper can be accepted."""
207,"""Augment your batch: better training with larger batches""","['Large Batch Training', 'Augmentation', 'Deep Learning']","""Recently, there is regained interest in large batch training of neural networks, both of theory and practice. New insights and methods allowed certain models to be trained using large batches with no adverse impact on performance. Most works focused on accelerating wall clock training time by modifying the learning rate schedule, without introducing accuracy degradation. We propose to use large batch training to boost accuracy and accelerate convergence by combining it with data augmentation. Our method, ""batch augmentation"", suggests using multiple instances of each sample at the same large batch. We show empirically that this simple yet effective method improves convergence and final generalization accuracy. We further suggest possible reasons for its success.""","""The authors propose to use large batch training of neural networks, where each batch contains multiple augmentations of each sample. The experiments demonstrate that this leads to better performance compared to training with small batches. However, as noted by Reviewers 2 and 3, the experiments do not convincingly show where the improvement comes from. Considering that the described technique is very simplistic, having an extensive ablation study and comparison to the strong baselines is essential. The rebuttal didnt address the reviewers' concerns, and they argue for rejection."""
208,"""Causal Reasoning from Meta-reinforcement learning""","['meta-learning', 'causal reasoning', 'deep reinforcement learning', 'artificial intelligence']","""Discovering and exploiting the causal structure in the environment is a crucial challenge for intelligent agents. Here we explore whether modern deep reinforcement learning can be used to train agents to perform causal reasoning. We adopt a meta-learning approach, where the agent learns a policy for conducting experiments via causal interventions, in order to support a subsequent task which rewards making accurate causal inferences.We also found the agent could make sophisticated counterfactual predictions, as well as learn to draw causal inferences from purely observational data. Though powerful formalisms for causal reasoning have been developed, applying them in real-world domains can be difficult because fitting to large amounts of high dimensional data often requires making idealized assumptions. Our results suggest that causal reasoning in complex settings may benefit from powerful learning-based approaches. More generally, this work may offer new strategies for structured exploration in reinforcement learning, by providing agents with the ability to performand interpretexperiments.""","""The reviewers raised a number of concerns including insufficiently demonstrated benefits of the proposed methodology, lack of explanations, and the lack of thorough and convincing experimental evaluation. The authors rebuttal failed to alleviate these concerns fully. I agree with the main concerns raised and, although I also believe that the work can result eventually in a very interesting paper, I cannot suggest it at this stage for presentation at ICLR."""
209,"""A theoretical framework for deep and locally connected ReLU network""","['theoretical analysis', 'deep network', 'optimization', 'disentangled representation']","""Understanding theoretical properties of deep and locally connected nonlinear network, such as deep convolutional neural network (DCNN), is still a hard problem despite its empirical success. In this paper, we propose a novel theoretical framework for such networks with ReLU nonlinearity. The framework bridges data distribution with gradient descent rules, favors disentangled representations and is compatible with common regularization techniques such as Batch Norm, after a novel discovery of its projection nature. The framework is built upon teacher-student setting, by projecting the student's forward/backward pass onto the teacher's computational graph. We do not impose unrealistic assumptions (e.g., Gaussian inputs, independence of activation, etc). Our framework could help facilitate theoretical analysis of many practical issues, e.g. disentangled representations in deep networks. ""","""This paper studies the behavior of gradient descent on deep neural network architectures with spatial locality, under generic input data distributions, using a planted or ""teacher-student"" model. Whereas R1 was supportive of this work, R2 and R3 could not verify the main statements and the proofs due to a severe lack of clarity and mathematical rigor. The AC strongly aligns with the latter, and therefore recommends rejection at this time, encouraging the authors to address clarity and rigor issues and resubmit their work again. """
210,"""Meta-Learning Probabilistic Inference for Prediction""","['probabilistic models', 'approximate inference', 'few-shot learning', 'meta-learning']","""This paper introduces a new framework for data efficient and versatile learning. Specifically: 1) We develop ML-PIP, a general framework for Meta-Learning approximate Probabilistic Inference for Prediction. ML-PIP extends existing probabilistic interpretations of meta-learning to cover a broad class of methods. 2) We introduce \Versa{}, an instance of the framework employing a flexible and versatile amortization network that takes few-shot learning datasets as inputs, with arbitrary numbers of shots, and outputs a distribution over task-specific parameters in a single forward pass. \Versa{} substitutes optimization at test time with forward passes through inference networks, amortizing the cost of inference and relieving the need for second derivatives during training. 3) We evaluate \Versa{} on benchmark datasets where the method sets new state-of-the-art results, and can handle arbitrary number of shots, and for classification, arbitrary numbers of classes at train and test time. The power of the approach is then demonstrated through a challenging few-shot ShapeNet view reconstruction task.""","""The paper proposes a decision-theoretic framework for meta-learning. The ideas and analysis are interesting and well-motivated, and the experiments are thorough. The primary concerns of the reviewers have been addressed in new revisions of the paper. The reviewers all agree that the paper should be accepted. Hence, I recommend acceptance."""
211,"""Improving Composition of Sentence Embeddings through the Lens of Statistical Relational Learning""","['Statistical Relational Learning', 'Sentence Embedding', 'Composition functions', 'Natural Language Inference', 'InferSent', 'SentEval', 'ComplEx']","""Various NLP problems -- such as the prediction of sentence similarity, entailment, and discourse relations -- are all instances of the same general task: the modeling of semantic relations between a pair of textual elements. We call them textual relational problems. A popular model for textual relational problems is to embed sentences into fixed size vectors and use composition functions (e.g. difference or concatenation) of those vectors as features for the prediction. Meanwhile, composition of embeddings has been a main focus within the field of Statistical Relational Learning (SRL) whose goal is to predict relations between entities (typically from knowledge base triples). In this work, we show that textual relational models implicitly use compositions from baseline SRL models. We show that such compositions are not expressive enough for several tasks (e.g. natural language inference). We build on recent SRL models to address textual relational problems, showing that they are more expressive, and can alleviate issues from simpler compositions. The resulting models significantly improve the state of the art in both transferable sentence representation learning and relation prediction.""","""This paper offers a new angle through which to study the development of comparison functions for sentence pair classification tasks by drawing on the literature on statistical relational learning. All three reviewers seemed happy to see an attempt to unify these two closely related relation-learning problems. However, none of the reviewers were fully convinced that this attempt has yielded any substantial new knowledge: Many of the ideas that come out of this synthesis have already appeared in the sentence-pair modeling literature (in work cited in the paper under review), and the proposed new methods do not yield substantial improvements for the tasks they're tested on. I'm happy to accept the authors' arguments that sentence-to-vector models have practical value, and I'm not placing too much weight on the reviewer's comments about the choice to use that modeling framework. I am slightly concerned that the reviewers (especially R2) observed some overly broad statements in the paper, and I urge the authors to take those comments very seriously. I'm mostly concerned, though, about the lack of an impactful positive contribution: I'd have hoped for a paper of this kind to offer a a method with clear empirical advantages over prior work, or else a formal result which is more clearly new, and the reviewers are not convinced that this paper makes a contribution of either kind. """
212,"""Mixed Precision Quantization of ConvNets via Differentiable Neural Architecture Search""","['Neural Net Quantization', 'Neural Architecture Search']","""Recent work in network quantization has substantially reduced the time and space complexity of neural network inference, enabling their deployment on embedded and mobile devices with limited computational and memory resources. However, existing quantization methods often represent all weights and activations with the same precision (bit-width). In this paper, we explore a new dimension of the design space: quantizing different layers with different bit-widths. We formulate this problem as a neural architecture search problem and propose a novel differentiable neural architecture search (DNAS) framework to efficiently explore its exponential search space with gradient-based optimization. Experiments show we surpass the state-of-the-art compression of ResNet on CIFAR-10 and ImageNet. Our quantized models with 21.1x smaller model size or 103.9x lower computational cost can still outperform baseline quantized or even full precision models.""","""The paper proposes a quantization framework that learns a different bit width per layer. It is based on a differentiable objective where the Gumbel softmax approach is used with an annealing procedure. The objective trades off accuracy and model size. The reviewers generally thought the idea has merit. Quoting from discussion comments (R4): ""The paper cited by AnonReviewer 3 is indeed close to the current submission, but in my opinion the strongest contribution of this paper is the formulation from architecture search perspective."" The approach is general, and seems to be reasonably efficient (ResNet 18 took ""less than 5 hours"") The main negatives are the comparison to other methods. In the rebuttal, the authors suggested in multiple places that they would update the submission with additional experiments in response to reviewer comments. As of the decision deadline, these experiments do not appear to have been added to the document. In the discussion: R4: ""This paper seems novel enough to me, but I agree that the prior work should at least be cited and compared to. This is a general weakness in the paper, the comparison to relevant prior works is not sufficient."" R3: ""Not only novel, but more general han the prior work mentioned, but the discussion / experiments do not seem to capture this."" With a range of scores around the borderline threshold for acceptance at ICLR, this is a difficult case. On the balance, it appears that shortcomings in the experimental results are not resolved in time for ICLR 2019. The missing results include ablation studies (promised to R4) and a comparison to DARTS (promised to R3): ""We plan to perform the suggested experiments of comparing with exhaustive search and DARTS. The results will be hopefully updated before the revision deadline and the camera-ready if the paper is accepted."" These results are not present and could not be evaluated during the review/discussion phase."""
213,"""Learning Two-layer Neural Networks with Symmetric Inputs""","['Neural Network', 'Optimization', 'Symmetric Inputs', 'Moment-of-moments']","""We give a new algorithm for learning a two-layer neural network under a very general class of input distributions. Assuming there is a ground-truth two-layer network y = A \sigma(Wx) + \xi, where A, W are weight matrices, \xi represents noise, and the number of neurons in the hidden layer is no larger than the input or output, our algorithm is guaranteed to recover the parameters A, W of the ground-truth network. The only requirement on the input x is that it is symmetric, which still allows highly complicated and structured input. Our algorithm is based on the method-of-moments framework and extends several results in tensor decompositions. We use spectral algorithms to avoid the complicated non-convex optimization in learning neural networks. Experiments show that our algorithm can robustly learn the ground-truth neural network with a small number of samples for many symmetric input distributions.""","""Although the paper considers a somewhat limited problem of learning a neural network with a single hidden layer, it achieves a surprisingly strong result that such a network can be learned exactly (or well approximated under sampling) under weaker assumptions than recent work. The reviewers unanimously recommended the paper be accepted. The paper would be more impactful if the authors could clarify the barriers to extending the technique of pure neuron detection to deeper networks, as well as the barriers to incorporating bias to eliminate the symmetry assumption."""
214,"""Minimum Divergence vs. Maximum Margin: an Empirical Comparison on Seq2Seq Models""","['sequence to sequence', 'training criteria']","""Sequence to sequence (seq2seq) models have become a popular framework for neural sequence prediction. While traditional seq2seq models are trained by Maximum Likelihood Estimation (MLE), much recent work has made various attempts to optimize evaluation scores directly to solve the mismatch between training and evaluation, since model predictions are usually evaluated by a task specific evaluation metric like BLEU or ROUGE scores instead of perplexity. This paper puts this existing work into two categories, a) minimum divergence, and b) maximum margin. We introduce a new training criterion based on the analysis of existing work, and empirically compare models in the two categories. Our experimental results show that our new training criterion can usually work better than existing methods, on both the tasks of machine translation and sentence summarization. ""","""The reviewers agree that the paper is worthy of publication at ICLR, hence I recommend accept. Regarding section 4.3 of the submission and the claim that this paper presents the first insight for existing work from a divergence minimization perspective, as pointed out by R2, I went and checked the details of RAML and they have similar insights in their equations (5) and (8). Please make this clearer in the paper. Regarding evaluation using greedy search instead of beam search, please consider using beam search for reporting test performance as this is the standard setup in sequence prediction. Please take my comments and the reviews into account an prepare the final version."""
215,"""Hierarchical Bayesian Modeling for Clustering Sparse Sequences in the Context of Group Profiling""","['Hierarchical Bayesian Modeling', 'Sparse sequence clustering', 'Group profiling', 'User group modeling']","""This paper proposes a hierarchical Bayesian model for clustering sparse sequences.This is a mixture model and does not need the data to be represented by a Gaussian mixture and that gives significant modelling freedom.It also generates a very interpretable profile for the discovered latent groups.The data that was used for the work have been contributed by a restaurant loyalty program company. The data is a collection of sparse sequences where each entry of each sequence is the number of user visits of one week to some restaurant. This algorithm successfully clustered the data and calculated the expected user affiliation in each cluster.""","""All reviewers agree to reject. While there were many positive points to this work, reviewers believed that it was not yet ready for acceptance."""
216,"""A Biologically Inspired Visual Working Memory for Deep Networks""","['memory', 'visual attention', 'image classification', 'image reconstruction', 'latent representations']","""The ability to look multiple times through a series of pose-adjusted glimpses is fundamental to human vision. This critical faculty allows us to understand highly complex visual scenes. Short term memory plays an integral role in aggregating the information obtained from these glimpses and informing our interpretation of the scene. Computational models have attempted to address glimpsing and visual attention but have failed to incorporate the notion of memory. We introduce a novel, biologically inspired visual working memory architecture that we term the Hebb-Rosenblatt memory. We subsequently introduce a fully differentiable Short Term Attentive Working Memory model (STAWM) which uses transformational attention to learn a memory over each image it sees. The state of our Hebb-Rosenblatt memory is embedded in STAWM as the weights space of a layer. By projecting different queries through this layer we can obtain goal-oriented latent representations for tasks including classification and visual reconstruction. Our model obtains highly competitive classification performance on MNIST and CIFAR-10. As demonstrated through the CelebA dataset, to perform reconstruction the model learns to make a sequence of updates to a canvas which constitute a parts-based representation. Classification with the self supervised representation obtained from MNIST is shown to be in line with the state of the art models (none of which use a visual attention mechanism). Finally, we show that STAWM can be trained under the dual constraints of classification and reconstruction to provide an interpretable visual sketchpad which helps open the `black-box' of deep learning.""","""The paper received mixed and divergent reviews. As a paper of unusual topic in ICLR, the presentation of this work would need improvement. For example, it is difficult to understand what's the overall objective function, why a specific design choice was made, etc. It's nice to see that the authors somehow did quite a bit of engineering to make their model work for classification and drawing tasks, but (as an ML person) its difficult to get a clear rationale on why the method works other than that its biologically motivated. In addition, the proposed model (at a functional level) looks quite similar to Mnih et al.'s ""Recurrent Models of Visual Attention"" work (for classification) and Gregor et al's DRAW model (for generation) in that all these models use sequential/recurrent attention/glimpse mechanisms, but no direct comparisons are made. For classification, the method achieves strong performance on MNIST but this may be due to a better architecture choice compared to Mnih's model but not due to the difference of the memory mechanism. For image generation/reconstruction, the proposed method seems to achieve quite good results but they are not as good as those from DRAW method. Overall, the paper is on the borderline, and while this work has some merits and might be of interest to some subset of audience in ICLR, there are many issues to be addressed in terms of presentation and comparisons. Please see reviews for other detailed comments."""
217,"""A Modern Take on the Bias-Variance Tradeoff in Neural Networks""","['bias-variance tradeoff', 'deep learning theory', 'generalization', 'concentration']","""We revisit the bias-variance tradeoff for neural networks in light of modern empirical findings. The traditional bias-variance tradeoff in machine learning suggests that as model complexity grows, variance increases. Classical bounds in statistical learning theory point to the number of parameters in a model as a measure of model complexity, which means the tradeoff would indicate that variance increases with the size of neural networks. However, we empirically find that variance due to training set sampling is roughly constant (with both width and depth) in practice. Variance caused by the non-convexity of the loss landscape is different. We find that it decreases with width and increases with depth, in our setting. We provide theoretical analysis, in a simplified setting inspired by linear models, that is consistent with our empirical findings for width. We view bias-variance as a useful lens to study generalization through and encourage further theoretical explanation from this perspective.""","""The paper revisits the traditional bias-variance trade-off for the case of large capacity neural networks. Reviewers requested several clarifications on the experimental setting and underlying results. Authors provided some, but it was deemed not enough for the paper to be strong enough to be accepted. Reviewers discussed among themselved but think that given the paper is mostly experimental, it needs more experimental evidence to be acceptable. Overall, I found the paper borderline but concur with the reviewers to reject it in its current form."""
218,"""Neural Rendering Model: Joint Generation and Prediction for Semi-Supervised Learning""","['neural nets', 'generative models', 'semi-supervised learning', 'cross-entropy']","""Unsupervised and semi-supervised learning are important problems that are especially challenging with complex data like natural images. Progress on these problems would accelerate if we had access to appropriate generative models under which to pose the associated inference tasks. Inspired by the success of Convolutional Neural Networks (CNNs) for supervised prediction in images, we design the Neural Rendering Model (NRM), a new hierarchical probabilistic generative model whose inference calculations correspond to those in a CNN. The NRM introduces a small set of latent variables at each level of the model and enforces dependencies among all the latent variables via a conjugate prior distribution. The conjugate prior yields a new regularizer for learning based on the paths rendered in the generative model for training CNNsthe Rendering Path Normalization (RPN). We demonstrate that this regularizer improves generalization both in theory and in practice. Likelihood estimation in the NRM yields the new Max-Min cross entropy training loss, which suggests a new deep network architecturethe Max- Min networkwhich exceeds or matches the state-of-art for semi-supervised and supervised learning on SVHN, CIFAR10, and CIFAR100.""","""This paper introduced a Neural Rendering Model, whose inference calculation corresponded to those in a CNN. It derived losses for both supervised and unsupervised learning settings. Furthermore, the paper introduced Max-Min network derived from the proposed loss, and showed strong performance on semi-supervised learning tasks. All reviewers agreed this paper introduces a highly interesting research direction and could be very useful for probabilistic inference. However, all reviewers found this paper hard to follow. It was written in an overly condensed way and tried to explain several concepts within the page limit such as NRM, rendering path, max-min network. In the end, it was not able to explain key concepts sufficiently. I suggest the authors take a major revision on the paper writing and give a better explanation about main components of the proposed method. The reviewer also suggested splitting the paper into two conference submissions in order to explain the main ideas sufficiently under a conference page limit."""
219,"""CoDraw: Collaborative Drawing as a Testbed for Grounded Goal-driven Communication""","['CoDraw', 'collaborative drawing', 'grounded language']","""In this work, we propose a goal-driven collaborative task that contains language, vision, and action in a virtual environment as its core components. Specifically, we develop a Collaborative image-Drawing game between two agents, called CoDraw. Our game is grounded in a virtual world that contains movable clip art objects. The game involves two players: a Teller and a Drawer. The Teller sees an abstract scene containing multiple clip art pieces in a semantically meaningful configuration, while the Drawer tries to reconstruct the scene on an empty canvas using available clip art pieces. The two players communicate via two-way communication using natural language. We collect the CoDraw dataset of ~10K dialogs consisting of ~138K messages exchanged between human agents. We define protocols and metrics to evaluate the effectiveness of learned agents on this testbed, highlighting the need for a novel ""crosstalk"" condition which pairs agents trained independently on disjoint subsets of the training data for evaluation. We present models for our task, including simple but effective baselines and neural network approaches trained using a combination of imitation learning and goal-driven training. All models are benchmarked using both fully automated evaluation and by playing the game with live human agents.""","""The reviewers raise a number of concerns including no methodological novelty, limited experimental evaluation, and relatively uninteresting application with very limited real-world application. This set of facts has been assessed differently by the three reviewers, and the scores range from probable rejection to probable acceptance. I believe that the work as is would not result in a wide interest by the ICLR attendees, mainly because of no methodological novelty and relatively simplistic application. The authors rebuttal failed to address these issues and I cannot recommend this work for presentation at ICLR."""
220,"""Classification in the dark using tactile exploration""","['tactile sensing', 'multimodal representations', 'vision', 'object identification']","""Combining information from different sensory modalities to execute goal directed actions is a key aspect of human intelligence. Specifically, human agents are very easily able to translate the task communicated in one sensory domain (say vision) into a representation that enables them to complete this task when they can only sense their environment using a separate sensory modality (say touch). In order to build agents with similar capabilities, in this work we consider the problem of a retrieving a target object from a drawer. The agent is provided with an image of a previously unseen object and it explores objects in the drawer using only tactile sensing to retrieve the object that was shown in the image without receiving any visual feedback. Success at this task requires close integration of visual and tactile sensing. We present a method for performing this task in a simulated environment using an anthropomorphic hand. We hope that future research in the direction of combining sensory signals for acting will find the object retrieval from a drawer to be a useful benchmark problem""","""The paper describes the use of tactile sensors for exploration. An important topic which has been addressed in various previous publications, but is unsolved to date. The research and the paper are unfortunately in a raw state. Rejected unanimously by the reviewers, without rebuttal chances used by the authors."""
221,"""Wizard of Wikipedia: Knowledge-Powered Conversational Agents""","['dialogue', 'knowledge', 'language', 'conversation']","""In open-domain dialogue intelligent agents should exhibit the use of knowledge, however there are few convincing demonstrations of this to date. The most popular sequence to sequence models typically generate and hope generic utterances that can be memorized in the weights of the model when mapping from input utterance(s) to output, rather than employing recalled knowledge as context. Use of knowledge has so far proved difficult, in part because of the lack of a supervised learning benchmark task which exhibits knowledgeable open dialogue with clear grounding. To that end we collect and release a large dataset with conversations directly grounded with knowledge retrieved from Wikipedia. We then design architectures capable of retrieving knowledge, reading and conditioning on it, and finally generating natural responses. Our best performing dialogue models are able to conduct knowledgeable discussions on open-domain topics as evaluated by automatic metrics and human evaluations, while our new benchmark allows for measuring further improvements in this important research direction.""","""The paper proposes a new dataset for studying knowledge grounded conversations, that would be very useful in advancing this field. In addition to the details of the dataset and its collection, the paper also includes a framework for advancing the research in this area, that includes evaluation methods and baselines with a relatively new approach. The proposed approach for dialogue generation however is a simple extension of previous work by (Zhang et al) to user transformers, hence is not very interesting. The proposed approach is also not compared to many previous studies in the experimental results. One of the reviewers highlighted the weakness of the human evaluation performed in the paper. Moving on, it would be useful if further approaches are considered and included in the task evaluation. A poster presentation of the work would enable participants to ask detailed questions about the proposed dataset and evaluation, and hence may be more appropriate. """
222,"""On the Geometry of Adversarial Examples""","['adversarial examples', 'high-dimensional geometry']","""Adversarial examples are a pervasive phenomenon of machine learning models where seemingly imperceptible perturbations to the input lead to misclassifications for otherwise statistically accurate models. We propose a geometric framework, drawing on tools from the manifold reconstruction literature, to analyze the high-dimensional geometry of adversarial examples. In particular, we highlight the importance of codimension: for low-dimensional data manifolds embedded in high-dimensional space there are many directions off the manifold in which to construct adversarial examples. Adversarial examples are a natural consequence of learning a decision boundary that classifies the low-dimensional data manifold well, but classifies points near the manifold incorrectly. Using our geometric framework we prove (1) a tradeoff between robustness under different norms, (2) that adversarial training in balls around the data is sample inefficient, and (3) sufficient sampling conditions under which nearest neighbor classifiers and ball-based adversarial training are robust.""","""The paper gives a theoretical analysis highlighting the role of codimension on the pervasiveness of adversarial examples. The paper demonstrates that a single decision boundary cannot be robust in different norms. They further proved that it is insufficient to learn robust decision boundaries by training against adversarial examples drawn from balls around the training set. The main concern with the paper is that most of the theoretical results might have a very restrictive scope and the writing is difficult to follow. The authors expressed concerns about a review not being very constructive. In a nutshell, the review in question points out that the theory might be too restrictive, that the experimental section is not very strong, that there are other works on related topics, and that the writing of the paper could be improved. While I understand the disappointing of the authors, the main points here appear to be consistent with the other reviews, which also mention that the theoretical results in this paper are not very general, that the writing is a bit complicated or heavy in mathematics, and not easy to follow, or that it is not clear if the bounds can be useful or easily applied in other work. One reviewer rates the paper marginally above the acceptance threshold, while two other reviewers rate the paper below the acceptance threshold. """
223,"""ISA-VAE: Independent Subspace Analysis with Variational Autoencoders""","['representation learning', 'disentanglement', 'interpretability', 'variational autoencoders']","""Recent work has shown increased interest in using the Variational Autoencoder (VAE) framework to discover interpretable representations of data in an unsupervised way. These methods have focussed largely on modifying the variational cost function to achieve this goal. However, we show that methods like beta-VAE simplify the tendency of variational inference to underfit causing pathological over-pruning and over-orthogonalization of learned components. In this paper we take a complementary approach: to modify the probabilistic model to encourage structured latent variable representations to be discovered. Specifically, the standard VAE probabilistic model is unidentifiable: the likelihood of the parameters is invariant under rotations of the latent space. This means there is no pressure to identify each true factor of variation with a latent variable. We therefore employ a rich prior distribution, akin to the ICA model, that breaks the rotational symmetry. Extensive quantitative and qualitative experiments demonstrate that the proposed prior mitigates the trade-off introduced by modified cost functions like beta-VAE and TCVAE between reconstruction loss and disentanglement. The proposed prior allows to improve these approaches with respect to both disentanglement and reconstruction quality significantly over the state of the art.""","""The paper proposes to improve VAE by using a prior distribution that has been previously proposed for independent subspace analysis (ISA). The clarity of the paper could be improved by more clearly describing the proposed method and its implementation details. The originality is not that high, as the main change to VAE is replacing the usual isotropic Gaussian prior with an ISA prior. Moreover, the paper does not provide comparison to VAEs with other more sophisticated priors, such as the VampPrior, and it is unclear whether using the ISA prior makes it difficult to scale to high-dimensional observations. Therefore, it is difficult to evaluate the significance of ISA-VAE. The authors are encouraged to carefully revise their paper to address these concerns. """
224,"""MANIFOLDNET: A DEEP NEURAL NETWORK FOR MANIFOLD-VALUED DATA""",[],"""Developing deep neural networks (DNNs) for manifold-valued data sets has gained much interest of late in the deep learning research community. Examples of manifold-valued data include data from omnidirectional cameras on automobiles, drones etc., diffusion magnetic resonance imaging, elastography and others. In this paper, we present a novel theoretical framework for DNNs to cope with manifold-valued data inputs. In doing this generalization, we draw parallels to the widely popular convolutional neural networks (CNNs). We call our network the ManifoldNet. As in vector spaces where convolutions are equivalent to computing the weighted mean of functions, an analogous definition for manifold-valued data can be constructed involving the computation of the weighted Fr\'{e}chet Mean (wFM). To this end, we present a provably convergent recursive computation of the wFM of the given data, where the weights makeup the convolution mask, to be learned. Further, we prove that the proposed wFM layer achieves a contraction mapping and hence the ManifoldNet does not need the additional non-linear ReLU unit used in standard CNNs. Operations such as pooling in traditional CNN are no longer necessary in this setting since wFM is already a pooling type operation. Analogous to the equivariance of convolution in Euclidean space to translations, we prove that the wFM is equivariant to the action of the group of isometries admitted by the Riemannian manifold on which the data reside. This equivariance property facilitates weight sharing within the network. We present experiments, using the ManifoldNet framework, to achieve video classification and image reconstruction using an auto-encoder+decoder setting. Experimental results demonstrate the efficacy of ManifoldNet in the context of classification and reconstruction accuracy.""","""This manuscript proposes an extension of convolution operations for manifold-valued data. The primary contributions include the development and description of the approach and implementation and evaluation on real data. The reviewers and AC expressed concern about the clarity of the presentation, particularly for a general ICLR audience. Though the contributions are primarily conceptual/theoretical, reviewers expressed concern about the breadth and quality of the presented experimental results. Some additional concerns related to missing proofs and details were addressed in the rebuttal."""
225,"""Progressive Weight Pruning Of Deep Neural Networks Using ADMM""","['deep learning', 'model compression', 'optimization', 'ADMM', 'weight pruning']","""Deep neural networks (DNNs) although achieving human-level performance in many domains, have very large model size that hinders their broader applications on edge computing devices. Extensive research work have been conducted on DNN model compression or pruning. However, most of the previous work took heuristic approaches. This work proposes a progressive weight pruning approach based on ADMM (Alternating Direction Method of Multipliers), a powerful technique to deal with non-convex optimization problems with potentially combinatorial constraints. Motivated by dynamic programming, the proposed method reaches extremely high pruning rate by using partial prunings with moderate pruning rates. Therefore, it resolves the accuracy degradation and long convergence time problems when pursuing extremely high pruning ratios. It achieves up to 34 pruning rate for ImageNet dataset and 167 pruning rate for MNIST dataset, significantly higher than those reached by the literature work. Under the same number of epochs, the proposed method also achieves faster convergence and higher compression rates. The codes and pruned DNN models are released in the anonymous link pseudo-url.""","""The paper proposes a progressive pruning technique that achieves high pruning ratio. Reviewers have a consensus on rejection. Reviewer 1 pointed out that the experimental results are weak. Reviewer 2 is also concerned about the proposed method and experiments. Reviewer 3 is is concerned that this paper is incremental work. Overall, this paper does not meet the standard of ICLR. Recommend for rejection. """
226,"""Minimal Images in Deep Neural Networks: Fragile Object Recognition in Natural Images""",[],"""The human ability to recognize objects is impaired when the object is not shown in full. ""Minimal images"" are the smallest regions of an image that remain recognizable for humans. Ullman et al. (2016) show that a slight modification of the location and size of the visible region of the minimal image produces a sharp drop in human recognition accuracy. In this paper, we demonstrate that such drops in accuracy due to changes of the visible region are a common phenomenon between humans and existing state-of-the-art deep neural networks (DNNs), and are much more prominent in DNNs. We found many cases where DNNs classified one region correctly and the other incorrectly, though they only differed by one row or column of pixels, and were often bigger than the average human minimal image size. We show that this phenomenon is independent from previous works that have reported lack of invariance to minor modifications in object location in DNNs. Our results thus reveal a new failure mode of DNNs that also affects humans to a much lesser degree. They expose how fragile DNN recognition ability is in natural images even without adversarial patterns being introduced. Bringing the robustness of DNNs in natural images to the human level remains an open challenge for the community. ""","""This paper characterizes a particular kind of fragility in the image classification ability of deep networks: minimal image regions which are classified correctly, but for which neighboring regions shifted by one row or column of pixels are classified incorrectly. Comparisons are made to human vision. All three reviewers recommend acceptance. AnonReviewer1 places the paper marginally above threshold, due to limited originality over Ullman et al. 2016, and concerns about overall significance. """
227,"""Improved Gradient Estimators for Stochastic Discrete Variables""","['continuous relaxation', 'discrete stochastic variables', 'reparameterization trick', 'variational inference', 'discrete optimization', 'stochastic gradient estimation']","""In many applications we seek to optimize an expectation with respect to a distribution over discrete variables. Estimating gradients of such objectives with respect to the distribution parameters is a challenging problem. We analyze existing solutions including finite-difference (FD) estimators and continuous relaxation (CR) estimators in terms of bias and variance. We show that the commonly used Gumbel-Softmax estimator is biased and propose a simple method to reduce it. We also derive a simpler piece-wise linear continuous relaxation that also possesses reduced bias. We demonstrate empirically that reduced bias leads to a better performance in variational inference and on binary optimization tasks.""","""Strengths: This paper provides a useful review of some of the recent work on gradient estimators for discrete variables, and proposes both a computationally more efficient variant of one, and a new estimator based on piecewise linear functions. Weaknesses: Many new ideas are scattered throughout the paper. The notation is a bit dense. Comparisons to RELAX, which had better results than REBAR, are missing. Finally, it seems that REBAR was trained with a fixed temperature, instead of optimizing it during training, which is one of the main benefits of the method. Points of contention: Only R1 mentioned the omission of REBAR and RELAX. A discussion and a few comparisons to REBAR were added to the paper, but only in a few experiments. Consensus: This paper is borderline. I agree with R1: quality 6, clarity 8, originality 6, significance 4. All reviewers agreed that this was a decent paper but I think that R2 and R3 were relatively unfamiliar with the existing literature. Update for clarification: ===================== This section has been added to clarify the reasons for rejection. The abstract of the paper states: ""We show that the commonly used Gumbel-Softmax estimator is biased and propose a simple method to reduce it. We also derive a simpler piece-wise linear continuous relaxation that also possesses reduced bias. We demonstrate empirically that reduced bias leads to a better performance in variational inference and on binary optimization tasks."" The fact that Gumbel-Softmax is biased is well-known. Reducing its bias was the motivation for developing the _exactly_ unbiased REBAR method, which already has similar asymptotic complexity. A major side-benefit of using an exactly unbiased estimator is that the estimator's hyperparameters can be automatically tuned to reduce variance, as in REBAR and RELAX. This paper focuses on methods for reducing bias and variance, but hardly discusses related methods that already achieved its stated aims. This a major weakness of the paper. The experiments only compared with REBAR, and did not even tune the temperature to reduce variance (removing one of its major advantages). This reject decision is not made mainly on lack of experiments or state-of-the-art results. It's because the idea of reducing the bias of continuous-relaxation-based gradient estimators has already been fruitfully explored, and zero-bias CR estimators have been developed, but this work mostly ignores them. However, thorough experiments are always going to be necessary for a paper proposing biased estimators, because there are already many such estimators, and little theory to say which ones will work well in which situations. Suggestions to improve the paper: Run experiments on all methods that directly measure bias and variance. Incorporate discussion of REBAR throughout, not just in an appendix. Run comparisons against REBAR and RELAX without crippling their ability to reduce variance. Do more to characterize when different estimators will be expected to be effective."""
228,"""DelibGAN: Coarse-to-Fine Text Generation via Adversarial Network""","['unsupervised text generation', 'coarse-to-fine generator', 'multiple instance discriminator', 'GAN', 'DelibGAN']","""In this paper, we propose a novel adversarial learning framework, namely DelibGAN, for generating high-quality sentences without supervision. Our framework consists of a coarse-to-fine generator, which contains a first-pass decoder and a second-pass decoder, and a multiple instance discriminator. And we propose two training mechanisms DelibGAN-I and DelibGAN-II. The discriminator is used to fine-tune the second-pass decoder in DelibGAN-I and further evaluate the importance of each word and tune the first-pass decoder in DelibGAN-II. We compare our models with several typical and state-of-the-art unsupervised generic text generation models on three datasets (a synthetic dataset, a descriptive text dataset and a sentimental text dataset). Both qualitative and quantitative experimental results show that our models produce more realistic samples, and DelibGAN-II performs best.""","""All reviewers are in agreement for a rejection decision. Details below."""
229,"""Distribution-Interpolation Trade off in Generative Models""","['generative models', 'latent distribution', 'Cauchy distribution', 'interpolations']","""We investigate the properties of multidimensional probability distributions in the context of latent space prior distributions of implicit generative models. Our work revolves around the phenomena arising while decoding linear interpolations between two random latent vectors -- regions of latent space in close proximity to the origin of the space are oversampled, which restricts the usability of linear interpolations as a tool to analyse the latent space. We show that the distribution mismatch can be eliminated completely by a proper choice of the latent probability distribution or using non-linear interpolations. We prove that there is a trade off between the interpolation being linear, and the latent distribution having even the most basic properties required for stable training, such as finite mean. We use the multidimensional Cauchy distribution as an example of the prior distribution, and also provide a general method of creating non-linear interpolations, that is easily applicable to a large family of commonly used latent distributions.""","""All the reviewers and AC agrees that the main strength of the paper that it studies a rather important question of the validity of using linear interpolation in evaluating GANs. The paper gives concrete examples and theoretical and empirical analysis that shows linear interpolation is not a great idea. The potential weakness is that the paper doesn't provide a very convincing new evaluation to replace the linear interpolation. However, given that it's largely unclear what are the right evaluations for GANs, the AC thinks the ""negative result"" about linear interpolation already deserves an ICLR paper. """
230,"""Learning a Meta-Solver for Syntax-Guided Program Synthesis""","['Syntax-guided Synthesis', 'Context Free Grammar', 'Logical Specification', 'Representation Learning', 'Meta Learning', 'Reinforcement Learning']","""We study a general formulation of program synthesis called syntax-guided synthesis(SyGuS) that concerns synthesizing a program that follows a given grammar and satisfies a given logical specification. Both the logical specification and the grammar have complex structures and can vary from task to task, posing significant challenges for learning across different tasks. Furthermore, training data is often unavailable for domain specific synthesis tasks. To address these challenges, we propose a meta-learning framework that learns a transferable policy from only weak supervision. Our framework consists of three components: 1) an encoder, which embeds both the logical specification and grammar at the same time using a graph neural network; 2) a grammar adaptive policy network which enables learning a transferable policy; and 3) a reinforcement learning algorithm that jointly trains the embedding and adaptive policy. We evaluate the framework on 214 cryptographic circuit synthesis tasks. It solves 141 of them in the out-of-box solver setting, significantly outperforming a similar search-based approach but without learning, which solves only 31. The result is comparable to two state-of-the-art classical synthesis engines, which solve 129 and 153 respectively. In the meta-solver setting, the framework can efficiently adapt to unseen tasks and achieves speedup ranging from 2x up to 100x.""","""This paper presents an RL agent which progressively synthesis programs according to syntactic constraints, and can learn to solve problems with different DSLs, demonstrating some degree of transfer across program synthesis problems. Reviewers agreed that this was an exciting and important development in program synthesis and meta-learning (if that word still has any meaning to it), and were impressed with both the clarity of the paper and its evaluation. There were some concerns about missing baselines and benchmarks, some of which were resolved during the discussion period, although it would still be good to compare to out-of-the-box MCTS. Overall, everyone agrees this is a strong paper and that it belongs in the conference, so I have no hesitation in recommending it."""
231,"""Polar Prototype Networks""","['prototype networks', 'polar prototypes', 'output structure']","""This paper proposes a neural network for classification and regression, without the need to learn layout structures in the output space. Standard solutions such as softmax cross-entropy and mean squared error are effective but parametric, meaning that known inductive structures such as maximum margin separation and simplicity (Occam's Razor) need to be learned for the task at hand. Instead, we propose polar prototype networks, a class of networks that explicitly states the structure, \ie the layout, of the output. The structure is defined by polar prototypes, points on the hypersphere of the output space. For classification, each class is described by a single polar prototype and they are a priori distributed with maximal separation and equal shares on the hypersphere. Classes are assigned to prototypes randomly or based on semantic priors and training becomes a matter of minimizing angular distances between examples and their class prototypes. For regression, we show that training can be performed as a polar interpolation between two prototypes, arriving at a regression with higher-dimensional outputs. From empirical analysis, we find that polar prototype networks benefit from large margin separation and semantic class structure, while only requiring a minimal amount of output dimensions. While the structure is simple, the performance is on par with (classification) or better than (regression) standard network methods. Moreover, we show that we gain the ability to perform regression and classification jointly in the same space, which is disentangled and interpretable by design.""","""This work proposes a class of neural networks that can jointly perform classification and regression in the output space. The authors explore the concept of polar prototypes which are points on the hypersphere in the output space. For classification, each class is described by a single polar prototype and training is equivalent to minimizing angular distances between examples and their class prototypes. For regression, training can be performed as a polar interpolation between two prototypes. As rightly acknowledged by R3, it is nice to see an alternative to the dominant cross-entropy loss and l2 loss for deep classification and regression respectively, also the ability to tackle both at the same time. However, all reviewers and AC agreed that the current manuscript lacks convincing empirical evaluations that clearly show the benefits of the proposed approach. To strengthen the evaluation, (1) see R1s concern regarding the state-of-the-art performance on CIFAR-10; (2) see R3s suggestion to use more challenging datasets (e.g. ImageNet), stronger backbone networks (e.g. densenet), and also other applications (e.g. object recognition and pose estimation; face recognition and age estimation as classification and regression problems); (3) see R2s suggestions for more baselines to be compared to. Two other requests to further strengthen the manuscript are: (1) finding alternative ways to MC or evolutionary algorithms (R2); (2) exploring class correlation in the prototype space (R2). In the response, the authors acknowledged that their initial results were not aimed for state-of-the-art comparison, but to show that the proposed objective is comparable to minimizing softmax cross-entropy loss. The authors provide additional experiments using DenseNet as the base network and the results are still slightly inferior to state-of-the-art performance. The experiments using ImageNet dataset have been promised by the authors (in response to R3), but are not included in the current revision. AC suggests in its current state the manuscript is not ready for a publication. We hope the reviews are useful for improving and revising the paper. """
232,"""Disjoint Mapping Network for Cross-modal Matching of Voices and Faces""","['cross-modal matching', 'voices', 'faces']","""We propose a novel framework, called Disjoint Mapping Network (DIMNet), for cross-modal biometric matching, in particular of voices and faces. Different from the existing methods, DIMNet does not explicitly learn the joint relationship between the modalities. Instead, DIMNet learns a shared representation for different modalities by mapping them individually to their common covariates. These shared representations can then be used to find the correspondences between the modalities. We show empirically that DIMNet is able to achieve better performance than the current state-of-the-art methods, with the additional benefits of being conceptually simpler and less data-intensive.""","""All reviewers agree that the proposed method interesting and well presented. The authors' rebuttal addressed all outstanding raised issues. Two reviewers recommend clear accept and the third recommends borderline accept. I agree with this recommendation and believe that the paper will be of interest to the audience attending ICLR. I recommend accepting this work for a poster presentation at ICLR."""
233,"""Eidetic 3D LSTM: A Model for Video Prediction and Beyond""",[],"""Spatiotemporal predictive learning, though long considered to be a promising self-supervised feature learning method, seldom shows its effectiveness beyond future video prediction. The reason is that it is difficult to learn good representations for both short-term frame dependency and long-term high-level relations. We present a new model, Eidetic 3D LSTM (E3D-LSTM), that integrates 3D convolutions into RNNs. The encapsulated 3D-Conv makes local perceptrons of RNNs motion-aware and enables the memory cell to store better short-term features. For long-term relations, we make the present memory state interact with its historical records via a gate-controlled self-attention module. We describe this memory transition mechanism eidetic as it is able to effectively recall the stored memories across multiple time stamps even after long periods of disturbance. We first evaluate the E3D-LSTM network on widely-used future video prediction datasets and achieve the state-of-the-art performance. Then we show that the E3D-LSTM network also performs well on the early activity recognition to infer what is happening or what will happen after observing only limited frames of video. This task aligns well with video prediction in modeling action intentions and tendency.""","""Strengths: Strong results on future frame video prediction using a 3D convolutional network. Use of future video prediction to jointly learn auxiliary tasks shown to to increase performance. Good ablation study. Weaknesses: Comparisons with older action recognition methods. Some concerns about novelty, the main contribution is the E3D-LSTM architecture, which R1 characterized as an LSTM with an extra gate and attention mechanism. Contention: Authors point to novelty in 3D convolutions inside the RNN. Consensus: All reviewers give a final score of 7- well done experiments helped address concerns around novelty. Easy to recommend acceptance given the agreement. """
234,"""Gradient Descent Provably Optimizes Over-parameterized Neural Networks""","['theory', 'non-convex optimization', 'overparameterization', 'gradient descent']","""One of the mysteries in the success of neural networks is randomly initialized first order methods like gradient descent can achieve zero training loss even though the objective function is non-convex and non-smooth. This paper demystifies this surprising phenomenon for two-layer fully connected ReLU activated neural networks. For an pseudo-formula hidden node shallow neural network with ReLU activation and pseudo-formula training data, we show as long as pseudo-formula is large enough and no two inputs are parallel, randomly initialized gradient descent converges to a globally optimal solution at a linear convergence rate for the quadratic loss function. Our analysis relies on the following observation: over-parameterization and random initialization jointly restrict every weight vector to be close to its initialization for all iterations, which allows us to exploit a strong convexity-like property to show that gradient descent converges at a global linear rate to the global optimum. We believe these insights are also useful in analyzing deep models and other first order methods.""","""This paper proves that gradient descent with random initialization converges to global minima for a squared loss penalty over a two layer ReLU network and arbitrarily labeled data. The paper has several weakness such as, 1) assuming top layer is fixed, 2) large number of hidden units 'm', 3) analysis is for squared loss. Despite these weaknesses the paper makes a novel contribution to a relatively challenging problem, and is able to show convergence results without strong assumptions on the input data or the model. Reviewers find the results mostly interesting and have some concerns about the \lambda_0 requirement. I believe the authors have sufficiently addressed this issue in their response and I suggest acceptance. """
235,"""Pearl: Prototype lEArning via Rule Lists""","['rule list learning', 'prototype learning', 'interpretability', 'healthcare']","""Deep neural networks have demonstrated promising prediction and classification performance on many healthcare applications. However, the interpretability of those models are often lacking. On the other hand, classical interpretable models such as rule lists or decision trees do not lead to the same level of accuracy as deep neural networks and can often be too complex to interpret (due to the potentially large depth of rule lists). In this work, we present PEARL, Prototype lEArning via Rule Lists, which iteratively uses rule lists to guide a neural network to learn representative data prototypes. The resulting prototype neural network provides accurate prediction, and the prediction can be easily explained by prototype and its guiding rule lists. Thanks to the prediction power of neural networks, the rule lists from prototypes are more concise and hence provide better interpretability. On two real-world electronic healthcare records (EHR) datasets, PEARL consistently outperforms all baselines across both datasets, especially achieving performance improvement over conventional rule learning by up to 28% and over prototype learning by up to 3%. Experimental results also show the resulting interpretation of PEARL is simpler than the standard rule learning.""","""This paper presents an approach that combines rule lists with prototype-based neural models to learn accurate models that are also interpretable (both due to rules and the prototypes). This combination is quite novel, the reviewers and the AC are unaware of prior work that has combined them, and find it potentially impactful. The experiments on the healthcare application were appreciated, and it is clear that the proposed approach produces accurate models, with much fewer rules than existing rule learning approaches. The reviewers and AC note the following potential weaknesses: (1) there are substantial presentation issues, including the details of the approach, (2) unclear what the differences are from existing approaches, in particular, the benefits, and (3) The evaluation lacked in several important aspects, including user study on interpretability, and choice of benchmarks. The authors provided a revision to their paper that addresses some of the presentation issues in notation, and incorporates some of the evaluation considerations as appendices into the paper. However, the reviewer scores are unchanged since most of the presentation and evaluation concerns remain, requiring significant modifications to be addressed."""
236,"""Learning Disentangled Representations with Reference-Based Variational Autoencoders""","['Disentangled representations', 'Variational Autoencoders', 'Adversarial Learning', 'Weakly-supervised learning']","""Learning disentangled representations from visual data, where different high-level generative factors are independently encoded, is of importance for many computer vision tasks. Supervised approaches, however, require a significant annotation effort in order to label the factors of interest in a training set. To alleviate the annotation cost, we introduce a learning setting which we refer to as ""reference-based disentangling''. Given a pool of unlabelled images, the goal is to learn a representation where a set of target factors are disentangled from others. The only supervision comes from an auxiliary ""reference set'' that contains images where the factors of interest are constant. In order to address this problem, we propose reference-based variational autoencoders, a novel deep generative model designed to exploit the weak supervisory signal provided by the reference set. During training, we use the variational inference framework where adversarial learning is used to minimize the objective function. By addressing tasks such as feature learning, conditional image generation or attribute transfer, we validate the ability of the proposed model to learn disentangled representations from minimal supervision. ""","""This is a proposed method that studies learning of disentangled representations in a relatively specific setting, defined as follows: given two datasets, one unlabeled and another that has a particular factor of variation fixed, the method will disentangle the factor of variation from the others. The reviewers found the method promising, with interesting results (qual & quant). The weaknesses of the method as discussed in the reviews and after: - the quantitative results with weak supervision are not a big improvement over beta-vae-like methods or mathieu et al. - a red flag of sorts to me is that it is not very clear where the gains are coming from: the authors claim to have done a fair comparison with the various baselines, but they introduce an entirely new encoder/decoder architecture that was likely (involuntarily, but still) tuned more to their method than others. - the setup as presented is somewhat artificial and less general than it could be (however, this was not a major factor in my decision). It is easy to get confused by the kind of disentagled representations that this work is aiming to get. I think this has the potential to be a solid paper, but at this stage it's missing a number of ablation studies to truly understand what sets it apart from the previous work. At the very least, there is a number of architectural and training choices in Appendix D -- like the 0.25 dropout -- that require more explanation / empirical understanding and how they generalize to other datasets. Given all of this, at this point it is hard for me to recommend acceptance of this work. I encourage the authors to take all this feedback into account, extend their work to more domains (the artistic-style disentangling that they mention seems like a good idea) and provide more empirical evidence about their architectural choices and their effect on the results."""
237,"""EMI: Exploration with Mutual Information Maximizing State and Action Embeddings""","['reinforcement learning', 'exploration', 'representation learning']","""Policy optimization struggles when the reward feedback signal is very sparse and essentially becomes a random search algorithm until the agent stumbles upon a rewarding or the goal state. Recent works utilize intrinsic motivation to guide the exploration via generative models, predictive forward models, or more ad-hoc measures of surprise. We propose EMI, which is an exploration method that constructs embedding representation of states and actions that does not rely on generative decoding of the full observation but extracts predictive signals that can be used to guide exploration based on forward prediction in the representation space. Our experiments show the state of the art performance on challenging locomotion task with continuous control and on image-based exploration tasks with discrete actions on Atari.""","""This paper proposes a method to compute embeddings of states and actions that facilitate computing measures of surprise for intrinsic reward. Though some of the ideas are quite interesting, there are currently issues with the experiments and the motivation. The experiments have high variance across the 5 runs, with significant overlap of shaded regions representing just one standard deviation from the mean. It is hard to draw any conclusions about improved performance, and statements like the following are much too strong: ""For vision-based exploration tasks, our results in Figure 5 show that EMI achieves the state of the art performance on Freeway, Frostbite, Venture, and Montezumas Revenge in comparison to the baseline exploration methods."" Further, the proposed approach has three new hyperparameters (lambdas), without much understanding into how to set them or their effect on the results. Specific values are reported for the different game types, without explanation for how or why these values were chosen. Similarly strong claims, that are not well substantiated, are given for the proposed approach. This paper seems to suggest that this is a principled approach to using surprise for exploration, contrasted to other ad-hoc approaches (""Other approaches utilize more ad-hoc measures (Pathak et al., 2017; Tang et al., 2017) that aim to approximate surprise.""). Yet, the paper does not define surprise (say by citing work by Itti and Baldi on Bayesian surprise), and then proposes what is largely a intuitive approach to providing a good intrinsic reward related to surprise. For example, ""we show that imposing linear topology on the learned embedding representation space (such that the transitions are linear), thereby offloading most of the modeling burden onto the embedding function itself, provides an essential informative measure of surprise when visiting novel states."" This might be intuitively true, but I do not see a clear demonstration in Section 4.2 actually showing that this restriction provides a measure of surprise. Additionally, some of the choices in Section 4.2 are about estimating ""irreducible error under the linear dynamics model"", but irreducible error is about inherent uncertainty (due to stochasticity and partial observability), not due to the choice of modeling class. In general, many intuitive choices in the algorithm need to be better justified, and some claims disparaging other work for being ad-hoc should be toned down. Overall, this paper is as yet a bit preliminary, in terms of clarity and experiments. In a further iteration, with some improvements, it could be a useful contribution for exploration in image-based environments. """
238,"""Exploring the interpretability of LSTM neural networks over multi-variable data""","['Interpretability', 'recurrent neural network', 'attention']","""In learning a predictive model over multivariate time series consisting of target and exogenous variables, the forecasting performance and interpretability of the model are both essential for deployment and uncovering knowledge behind the data. To this end, we propose the interpretable multi-variable LSTM recurrent neural network (IMV-LSTM) capable of providing accurate forecasting as well as both temporal and variable level importance interpretation. In particular, IMV-LSTM is equipped with tensorized hidden states and update process, so as to learn variables-wise hidden states. On top of it, we develop a mixture attention mechanism and associated summarization methods to quantify the temporal and variable importance in data. Extensive experiments using real datasets demonstrate the prediction performance and interpretability of IMV-LSTM in comparison to a variety of baselines. It also exhibits the prospect as an end-to-end framework for both forecasting and knowledge extraction over multi-variate data. ""","""The reviewers appreciated the clarity of writing, and the importance of the problem being addressed. There was a moderate amount of discussion around the paper, but the two reviewers who responded to the author discussion were split in their opinion, with one slightly increasing their score to a 6, and the other remaining unconvinced. The scores overall are borderline for ICLR acceptance, and given that, no reviewer stepped forward to champion the paper."""
239,"""Effective and Efficient Batch Normalization Using Few Uncorrelated Data for Statistics' Estimation""","['batch normalization', 'acceleration', 'correlation', 'sampling']","""Deep Neural Networks (DNNs) thrive in recent years in which Batch Normalization (BN) plays an indispensable role. However, it has been observed that BN is costly due to the reduction operations. In this paper, we propose alleviating the BNs cost by using only a small fraction of data for mean & variance estimation at each iteration. The key challenge to reach this goal is how to achieve a satisfactory balance between normalization effectiveness and execution efficiency. We identify that the effectiveness expects less data correlation while the efficiency expects regular execution pattern. To this end, we propose two categories of approach: sampling or creating few uncorrelated data for statistics estimation with certain strategy constraints. The former includes Batch Sampling (BS) that randomly selects few samples from each batch and Feature Sampling (FS) that randomly selects a small patch from each feature map of all samples, and the latter is Virtual Dataset Normalization (VDN) that generates few synthetic random samples. Accordingly, multi-way strategies are designed to reduce the data correlation for accurate estimation and optimize the execution pattern for running acceleration in the meantime. All the proposed methods are comprehensively evaluated on various DNN models, where an overall training speedup by up to 21.7% on modern GPUs can be practically achieved without the support of any specialized libraries, and the loss of model accuracy and convergence rate are negligible. Furthermore, our methods demonstrate powerful performance when solving the well-known micro-batch normalization problem in the case of tiny batch size.""","""This paper proposes a faster approximation to batch norm, which avoids summing over the entire batch by subsampling either random examples or random image locations. It analyzes some of the tradeoffs of computation time vs. statistical efficiency of gradient estimation, and proposes schemes for decorrelating the samples to make good use of smaller numbers of samples. The proposal is a reasonable one, and seems to give a noticeable improvement in efficiency. However, it's not clear there is a substantial enough contribution for an ICLR paper. The idea of subsampling is fairly obvious, and various other methods have already been proposed which decouple the computation of BN statistics from the training batch. From a practical standpoint, it's not clear that the observed benefit is large enough to justify the considerable complexity of an efficient implementation. """
240,"""A Proposed Hierarchy of Deep Learning Tasks""","['Deep learning', 'scaling with data', 'computational complexity', 'learning curves', 'speech recognition', 'image recognition', 'machine translation', 'language modeling']","""As the pace of deep learning innovation accelerates, it becomes increasingly important to organize the space of problems by relative difficultly. Looking to other fields for inspiration, we see analogies to the Chomsky Hierarchy in computational linguistics and time and space complexity in theoretical computer science. As a complement to prior theoretical work on the data and computational requirements of learning, this paper presents an empirical approach. We introduce a methodology for measuring validation error scaling with data and model size and test tasks in natural language, vision, and speech domains. We find that power-law validation error scaling exists across a breadth of factors and that model size scales sublinearly with data size, suggesting that simple learning theoretic models offer insights into the scaling behavior of realistic deep learning settings, and providing a new perspective on how to organize the space of problems. We measure the power-law exponent---the ""steepness"" of the learning curve---and propose using this metric to sort problems by degree of difficulty. There is no data like more data, but some tasks are more effective at taking advantage of more data. Those that are more effective are easier on the proposed scale. Using this approach, we can observe that studied tasks in speech and vision domains scale faster than those in the natural language domain, offering insight into the observation that progress in these areas has proceeded more rapidly than in natural language.""","""This paper attempts at ranking of tasks handled by deep learning methods based on learning curves. A main premise of the paper is ""fitting learning curves to a power law, and then sorting tasks by empirical estimates of exponents"". The idea of the paper is quite interesting. However, the paper makes some bold claims which are a bit distant from the empirical study it conducts. It is hard to line up the order in Table 2 with the Chomsky hierarchy. Also, for various tasks, various different deep models are used (ResNets for image classification, LSTMs for LM, and so on). I was not convinced that the beta parameter is model-agnostic. Similar concerns are expressed by the reviewers, and they agree that the paper should address the criticism that they express in their feedback."""
241,"""Recovering the Lowest Layer of Deep Networks with High Threshold Activations""","['Deep Learning', 'Parameter Recovery', 'Non-convex optimization', 'high threshold activation']","""Giving provable guarantees for learning neural networks is a core challenge of machine learning theory. Most prior work gives parameter recovery guarantees for one hidden layer networks, however, the networks used in practice have multiple non-linear layers. In this work, we show how we can strengthen such results to deeper networks -- we address the problem of uncovering the lowest layer in a deep neural network under the assumption that the lowest layer uses a high threshold before applying the activation, the upper network can be modeled as a well-behaved polynomial and the input distribution is gaussian.""","""The reviewers reached a consense on that the paper is not quite ready for publication at ICRL. The main potential drawback include a) the exposition of the paper can be improved; b) it's not entirely clear that some of the assumptions (such as the threshold for the first layer, the polynomial approximation of higher layers) are meaningful , and it seems that the proof technique exploits heavily some of these assumptions and some of the key intermediate steps won't hold in practice. (see reviewer 3's comment for more details.) The authors clarify the writing and intuitions in the response, but overall the AC decided that the paper is not quite ready for publications at the moment. """
242,"""Layerwise Recurrent Autoencoder for General Real-world Traffic Flow Forecasting""","['traffic flow forecasting', 'spatiotemporal dependencies', 'deep learning', 'intelligent transportation system']","""Accurate spatio-temporal traffic forecasting is a fundamental task that has wide applications in city management, transportation area and financial domain. There are many factors that make this significant task also challenging, like: (1) maze-like road network makes the spatial dependency complex; (2) the traffic-time relationships bring non-linear temporal complication; (3) with the larger road network, the difficulty of flow forecasting grows. The prevalent and state-of-the-art methods have mainly been discussed on datasets covering relatively small districts and short time span, e.g., the dataset that is collected within a city during months. To forecast the traffic flow across a wide area and overcome the mentioned challenges, we design and propose a promising forecasting model called Layerwise Recurrent Autoencoder (LRA), in which a three-layer stacked autoencoder (SAE) architecture is used to obtain temporal traffic correlations and a recurrent neural networks (RNNs) model for prediction. The convolutional neural networks (CNNs) model is also employed to extract spatial traffic information within the transport topology for more accurate prediction. To the best of our knowledge, there is no general and effective method for traffic flow prediction in large area which covers a group of cities. The experiment is completed on such large scale real-world traffic datasets to show superiority. And a smaller dataset is exploited to prove universality of the proposed model. And evaluations show that our model outperforms the state-of-the-art baselines by 6% - 15%.""","""The paper proposes an interesting neural architecture for traffic flow forecasting, which is tested on a number of datasets. Unfortunately, the lack of clarity as well as precision in writing appears to be a big issue for this paper, which prevents it from being accepted for publication in its current form. However, the reviewers did provide valuable feedback regarding writing, explanation, presentation and structure, that the paper would benefit from. """
243,"""Maximum a Posteriori on a Submanifold: a General Image Restoration Method with GAN""",[],"""We propose a general method for various image restoration problems, such as denoising, deblurring, super-resolution and inpainting. The problem is formulated as a constrained optimization problem. Its objective is to maximize a posteriori probability of latent variables, and its constraint is that the image generated by these latent variables must be the same as the degraded image. We use a Generative Adversarial Network (GAN) as our density estimation model. Convincing results are obtained on MNIST dataset.""","""This paper proposes a framework of image restoration by searching for a MAP in a trained GAN subject to a degradation constraint. Experiments on MNIST show good performance in restoring the images under different types of degradation. The main problem as pointed out by R1 and R3 is that there has been rich literature of image restoration methods and also several recent works that also utilized GAN, but the authors failed to make comparison any of those baselines in the experiments. Additional experiments on natural images would provide more convincing evidence for the proposed algorithm. The authors argue that the restoration tasks in the experiments are too difficult for TV to work. It would be great to provide actual experiments to verify the claim."""
244,"""Learning to Adapt in Dynamic, Real-World Environments through Meta-Reinforcement Learning""","['meta-learning', 'reinforcement learning', 'meta reinforcement learning', 'online adaptation']","""Although reinforcement learning methods can achieve impressive results in simulation, the real world presents two major challenges: generating samples is exceedingly expensive, and unexpected perturbations or unseen situations cause proficient but specialized policies to fail at test time. Given that it is impractical to train separate policies to accommodate all situations the agent may see in the real world, this work proposes to learn how to quickly and effectively adapt online to new tasks. To enable sample-efficient learning, we consider learning online adaptation in the context of model-based reinforcement learning. Our approach uses meta-learning to train a dynamics model prior such that, when combined with recent data, this prior can be rapidly adapted to the local context. Our experiments demonstrate online adaptation for continuous control tasks on both simulated and real-world agents. We first show simulated agents adapting their behavior online to novel terrains, crippled body parts, and highly-dynamic environments. We also illustrate the importance of incorporating online adaptation into autonomous agents that operate in the real world by applying our method to a real dynamic legged millirobot: We demonstrate the agent's learned ability to quickly adapt online to a missing leg, adjust to novel terrains and slopes, account for miscalibration or errors in pose estimation, and compensate for pulling payloads.""","""The authors consider the use of MAML with model based RL and applied this to robotics tasks with very encouraging results. There was definite interest in the paper, but also some concerns over how the results were situated, particularly with respect to the related research in the robotics community. The authors are strongly encouraged to carefully consider this feedback, as they have been doing in their responses, and address this as well as possible in the final version. """
245,"""The Importance of Norm Regularization in Linear Graph Embedding: Theoretical Analysis and Empirical Demonstration""","['Graph Embedding', 'Generalization Analysis', 'Matrix Factorization']","""Learning distributed representations for nodes in graphs is a crucial primitive in network analysis with a wide spectrum of applications. Linear graph embedding methods learn such representations by optimizing the likelihood of both positive and negative edges while constraining the dimension of the embedding vectors. We argue that the generalization performance of these methods is not due to the dimensionality constraint as commonly believed, but rather the small norm of embedding vectors. Both theoretical and empirical evidence are provided to support this argument: (a) we prove that the generalization error of these methods can be bounded by limiting the norm of vectors, regardless of the embedding dimension; (b) we show that the generalization performance of linear graph embedding methods is correlated with the norm of embedding vectors, which is small due to the early stopping of SGD and the vanishing gradients. We performed extensive experiments to validate our analysis and showcased the importance of proper norm regularization in practice.""","""This paper provides a generalization analysis for graph embedding methods concluding with the observation that the norm of the embedding vectors provides an effective regularization, more so than dimensionality alone. The main theoretical result is backed up by several experiments. While the result appears to be correct, norm control, dimensionality reduction and early stopping during optimization are all very well studied in machine learning as effective regularizers, either operating alone or in conjunction. The regularization parameters, iteration count, embedding dimensionality is typically tuned for an application. The AC agrees with Reviewer 2 that the paper does not provide sufficiently interesting insights beyond this observation and is unlikely to influence practical applications of these methods. Both reviewer 2 and 3 have also raised points on the need for stronger empirical analysis."""
246,"""Better Generalization with On-the-fly Dataset Denoising""","['dataset denoising', 'supervised learning', 'implicit regularization']","""Memorization in over-parameterized neural networks can severely hurt generalization in the presence of mislabeled examples. However, mislabeled examples are to hard avoid in extremely large datasets. We address this problem using the implicit regularization effect of stochastic gradient descent with large learning rates, which we find to be able to separate clean and mislabeled examples with remarkable success using loss statistics. We leverage this to identify and on-the-fly discard mislabeled examples using a threshold on their losses. This leads to On-the-fly Data Denoising (ODD), a simple yet effective algorithm that is robust to mislabeled examples, while introducing almost zero computational overhead. Empirical results demonstrate the effectiveness of ODD on several datasets containing artificial and real-world mislabeled examples.""","""The paper aims to clean data samples with label noise in the training procedure. The reviewers and AC note the following potential weaknesses: (1) the assumption of uniform noise, which is not the case in practice, (2) marginal gains under real-world datasets and (3) highly empirical and ad-hoc approach. AC thinks the proposed method has potential and is interesting, but decided that the authors need more significant works to publish the work. """
247,"""BLISS in Non-Isometric Embedding Spaces""","['bilingual lexicon induction', 'semi-supervised methods', 'embeddings']","""Recent work on bilingual lexicon induction (BLI) has frequently depended either on aligned bilingual lexicons or on distribution matching, often with an assumption about the isometry of the two spaces. We propose a technique to quantitatively estimate this assumption of the isometry between two embedding spaces and empirically show that this assumption weakens as the languages in question become increasingly etymologically distant. We then propose Bilingual Lexicon Induction with Semi-Supervision (BLISS) --- a novel semi-supervised approach that relaxes the isometric assumption while leveraging both limited aligned bilingual lexicons and a larger set of unaligned word embeddings, as well as a novel hubness filtering technique. Our proposed method improves over strong baselines for 11 of 14 on the MUSE dataset, particularly for languages whose embedding spaces do not appear to be isometric. In addition, we also show that adding supervision stabilizes the learning procedure, and is effective even with minimal supervision.""","""This paper is very close to the decision boundary and the reviewers were split about whether it should be accepted or not. The authors updated the paper with additional experiments as request by the reviewers. The area chair acknowledges that there is some novelty that leads to (moderate) empirical gains but does not see these as sufficient to push the paper over the very competitive acceptance threshold. """
248,"""Complement Objective Training""","['optimization', 'entropy', 'image recognition', 'natural language understanding', 'adversarial attacks', 'deep learning']","""Learning with a primary objective, such as softmax cross entropy for classification and sequence generation, has been the norm for training deep neural networks for years. Although being a widely-adopted approach, using cross entropy as the primary objective exploits mostly the information from the ground-truth class for maximizing data likelihood, and largely ignores information from the complement (incorrect) classes. We argue that, in addition to the primary objective, training also using a complement objective that leverages information from the complement classes can be effective in improving model performance. This motivates us to study a new training paradigm that maximizes the likelihood of the ground-truth class while neutralizing the probabilities of the complement classes. We conduct extensive experiments on multiple tasks ranging from computer vision to natural language understanding. The experimental results confirm that, compared to the conventional training with just one primary objective, training also with the complement objective further improves the performance of the state-of-the-art models across all tasks. In addition to the accuracy improvement, we also show that models trained with both primary and complement objectives are more robust to single-step adversarial attacks. ""","""This paper proposes adding a second objective to the training of neural network classifiers that aims to make the distribution over incorrect labels as flat as possible for each training sample. The authors describe this as ""maximizing the complement entropy."" Rather than adding the cross-entropy objective and the (negative) complement entropy term (since the complement entropy should be maximized while the cross-entropy is minimized), this paper proposes an alternating optimization framework in which first a step is taken to reduce the cross-entropy, then a step is taken to maximize the complement entropy. Extensive experiments on image classification (CIFAR-10, CIFAR-100, SVHN, Tiny Imagenet, and Imagenet), neural machine translation (IWSLT 2015 English-Vietnamese task), and small-vocabulary isolated-word recognition (Google Commands), show that the proposed two-objective approach outperforms training only to minimize cross-entropy. Experiments on CIFAR-10 also show that models trained in this framework have somewhat better resistance to single-step adversarial attacks. Concerns about the presentation of the adversarial attack experiments were raised by anonymous commenters and one of the reviewers, but these concerns were addressed in the revision and discussion. The primary remaining concern is a lack of any theoretical guarantees that the alternating optimization converges, but the strong empirical results compensate for this problem."""
249,"""GraphSeq2Seq: Graph-Sequence-to-Sequence for Neural Machine Translation""","['Neural Machine Translation', 'Natural Language Generation', 'Graph Embedding', 'LSTM']","""Sequence-to-Sequence (Seq2Seq) neural models have become popular for text generation problems, e.g. neural machine translation (NMT) (Bahdanau et al.,2014; Britz et al., 2017), text summarization (Nallapati et al., 2017; Wang &Ling, 2016), and image captioning (Venugopalan et al., 2015; Liu et al., 2017). Though sequential modeling has been shown to be effective, the dependency graph among words contains additional semantic information and thus can be utilized for sentence modeling. In this paper, we propose a Graph-Sequence-to-Sequence(GraphSeq2Seq) model to fuse the dependency graph among words into the traditional Seq2Seq framework. For each sample, the sub-graph of each word is encoded to a graph representation, which is then utilized to sequential encoding. At last, a sequence decoder is leveraged for output generation. Since above model fuses different features by contacting them together to encode, we also propose a variant of our model that regards the graph representations as additional annotations in attention mechanism (Bahdanau et al., 2014) by separately encoding different features. Experiments on several translation benchmarks show that our models can outperform existing state-of-the-art methods, demonstrating the effectiveness of the combination of Graph2Seq and Seq2Seq.""","""This paper proposes a new method for graph representation in sequence-to-sequence models and validates its results on several tasks. The overall results are relatively strong. Overall, the reviewers thought this was a reasonable contribution if somewhat incremental. In addition, while the experimental comparison has greatly improved from the original version, there are still a couple of less satisfying points: notably the size of the training data is somewhat small. In addition, as far as I can tell all comparisons with other graph-based baselines actually aren't implemented in the same toolkit with the same hyperparameters, so it's a bit difficult to tell whether the gains are coming from the proposed method itself or from other auxiliary differences. I think this paper is very reasonable, and definitely on the borderline for acceptance, but given the limited number of slots available at ICLR this year I am leaning in favor of the other very good papers in my area."""
250,"""Directional Analysis of Stochastic Gradient Descent via von Mises-Fisher Distributions in Deep Learning""","['directional statistics', 'deep learning', 'SNR', 'gradient stochasticity', 'SGD', 'stochastic gradient', 'von Mises-Fisher', 'angle']","""Although stochastic gradient descent (SGD) is a driving force behind the recent success of deep learning, our understanding of its dynamics in a high-dimensional parameter space is limited. In recent years, some researchers have used the stochasticity of minibatch gradients, or the signal-to-noise ratio, to better characterize the learning dynamics of SGD. Inspired from these work, we here analyze SGD from a geometrical perspective by inspecting the stochasticity of the norms and directions of minibatch gradients. We propose a model of the directional concentration for minibatch gradients through von Mises-Fisher (VMF) distribution, and show that the directional uniformity of minibatch gradients increases over the course of SGD. We empirically verify our result using deep convolutional networks and observe a higher correlation between the gradient stochasticity and the proposed directional uniformity than that against the gradient norm stochasticity, suggesting that the directional statistics of minibatch gradients is a major factor behind SGD.""","""The paper presents a careful analysis of SGD by characterizing the stochastic gradient via von Mises-Fisher distributions. While the paper has good quality and clarity, and the authors' detailed response has further clarified several raised issues, some important concerns remain: Reviewer 1 would like to see careful discussions on related observations by other work in the literature, such as low rank Hessians in the over-parameterized regime, Reviewer 2 is concerned about the significance of the presented analysis and observations, and Reviewers 2 and 4 both would like to see how the presented theoretical analysis could be used to design improved algorithms. In the AC's opinion, while solid theoretical analysis of SGD is definitely valuable, it is highly desirable to demonstrate its practical value (considering that it does not provide clearly new insights about the learning dynamics of SGD)."""
251,"""BA-Net: Dense Bundle Adjustment Networks""","['Structure-from-Motion', 'Bundle Adjustment', 'Dense Depth Estimation']","""This paper introduces a network architecture to solve the structure-from-motion (SfM) problem via feature-metric bundle adjustment (BA), which explicitly enforces multi-view geometry constraints in the form of feature-metric error. The whole pipeline is differentiable, so that the network can learn suitable features that make the BA problem more tractable. Furthermore, this work introduces a novel depth parameterization to recover dense per-pixel depth. The network first generates several basis depth maps according to the input image, and optimizes the final depth as a linear combination of these basis depth maps via feature-metric BA. The basis depth maps generator is also learned via end-to-end training. The whole system nicely combines domain knowledge (i.e. hard-coded multi-view geometry constraints) and deep learning (i.e. feature learning and basis depth maps learning) to address the challenging dense SfM problem. Experiments on large scale real data prove the success of the proposed method.""","""The first reviewer summarizes the contribution well: This paper combines [a CNN that computes both a multi-scale feature pyramid and a depth prediction, which is expressed as a linear combination of ""depth bases""]. This is used to [define a dense re-projection error over the images, akin to that of dense or semi-dense methods]. [Then, this error is optimized with respect to the camera parameters and depth linear combination coefficients using Levenberg-Marquardt (LM). By unrolling 5 iterations of LM and expressing the dampening parameter lambda as the output of a MLP, the optimization process is made differentiable, allowing back-propagation and thus learning of the networks' parameters.] Strengths: While combining deep learning methods with bundle adjustment is not new, reviewers generally agree that the particular way in which that is achieved in this paper is novel and interesting. The authors accounted for reviewer feedback during the review cycle and improved the manuscript leading to an increased rating. Weaknesses: Weaknesses were addressed during the rebuttal including better evaluation of their predicted lambda and comparison with CodeSLAM. Contention: This paper was not particularly contentious, there was a score upgrade due to the efforts of the authors during the rebuttal period. Consensus: This paper addresses an interesting area of research at the intersection of geometric computer vision and deep learning and should be of considerable interest to many within the ICLR community. The discussion of the paper highlighted some important nuances of terminology regarding the characterization of different methods. This paper was also rated the highest in my batch. As such, I recommend this paper for an oral presentation. """
252,"""Hierarchical Generative Modeling for Controllable Speech Synthesis""","['speech synthesis', 'representation learning', 'deep generative model', 'sequence-to-sequence model']","""This paper proposes a neural end-to-end text-to-speech (TTS) model which can control latent attributes in the generated speech that are rarely annotated in the training data, such as speaking style, accent, background noise, and recording conditions. The model is formulated as a conditional generative model with two levels of hierarchical latent variables. The first level is a categorical variable, which represents attribute groups (e.g. clean/noisy) and provides interpretability. The second level, conditioned on the first, is a multivariate Gaussian variable, which characterizes specific attribute configurations (e.g. noise level, speaking rate) and enables disentangled fine-grained control over these attributes. This amounts to using a Gaussian mixture model (GMM) for the latent distribution. Extensive evaluation demonstrates its ability to control the aforementioned attributes. In particular, it is capable of consistently synthesizing high-quality clean speech regardless of the quality of the training data for the target speaker.""","""This is an ambitious paper tackling the important and timely problem of controlling non-annotated attributes in generated speech. The reviewers had mixed opinions about the results. R1 asks for more convincing exposition of results but, nevertheless, acknowledging that it is difficult to evaluate TTS systems systematically. Besides, R2 and R3 find the results good. Judging from the reviews and previous work, this paper does not seem to be very novel, although it certainly has intriguing new elements. Furthermore, it constitutes a mature piece of work. """
253,"""Minimal Random Code Learning: Getting Bits Back from Compressed Model Parameters""","['compression', 'neural networks', 'bits-back argument', 'Bayesian', 'Shannon', 'information theory']","""While deep neural networks are a highly successful model class, their large memory footprint puts considerable strain on energy consumption, communication bandwidth, and storage requirements. Consequently, model size reduction has become an utmost goal in deep learning. A typical approach is to train a set of deterministic weights, while applying certain techniques such as pruning and quantization, in order that the empirical weight distribution becomes amenable to Shannon-style coding schemes. However, as shown in this paper, relaxing weight determinism and using a full variational distribution over weights allows for more efficient coding schemes and consequently higher compression rates. In particular, following the classical bits-back argument, we encode the network weights using a random sample, requiring only a number of bits corresponding to the Kullback-Leibler divergence between the sampled variational distribution and the encoding distribution. By imposing a constraint on the Kullback-Leibler divergence, we are able to explicitly control the compression rate, while optimizing the expected loss on the training set. The employed encoding scheme can be shown to be close to the optimal information-theoretical lower bound, with respect to the employed variational family. Our method sets new state-of-the-art in neural network compression, as it strictly dominates previous approaches in a Pareto sense: On the benchmarks LeNet-5/MNIST and VGG-16/CIFAR-10, our approach yields the best test performance for a fixed memory budget, and vice versa, it achieves the highest compression rates for a fixed test performance.""","""This paper proposes a novel coding scheme for compressing neural network weights using Shannon-style coding and a variational distribution over weights. This approach is shown to improve over existing schemes for LeNet-5 on MNIST and VGG-16 on CIFAR-10, strictly dominating them in terms of compression/error rate tradeoffs. Comparing to more baselines would have been helpful. Theoretical analysis based on non-trivial extensions of prior work by Harsha et al. (2010) and Chatterjee & Diaconis (2018) is also presented. Overall, there was consensus among the reviewers that the paper makes a solid contribution and should be published. """
254,"""Backplay: 'Man muss immer umkehren'""","['Exploration', 'Games', 'Pommerman', 'Bomberman', 'AI', 'Reinforcement Learning', 'Machine Learning']","""Model-free reinforcement learning (RL) requires a large number of trials to learn a good policy, especially in environments with sparse rewards. We explore a method to improve the sample efficiency when we have access to demonstrations. Our approach, Backplay, uses a single demonstration to construct a curriculum for a given task. Rather than starting each training episode in the environment's fixed initial state, we start the agent near the end of the demonstration and move the starting point backwards during the course of training until we reach the initial state. Our contributions are that we analytically characterize the types of environments where Backplay can improve training speed, demonstrate the effectiveness of Backplay both in large grid worlds and a complex four player zero-sum game (Pommerman), and show that Backplay compares favorably to other competitive methods known to improve sample efficiency. This includes reward shaping, behavioral cloning, and reverse curriculum generation.""",""" -pros: - good, sensible idea - good evaluations on the domains considered - good analysis -cons: - novelty, broader evaluation I think this is a good and interesting paper and I appreciate the authors' engagment with the reviewers. I agree with the authors that it is not fair to compare their work to a blog post which hasn't been published and I have taken this into account. However, there is still concern among the reviewers about the strength of the technical contribution and the decision was made not to accept for ICLR this year. """
255,"""Variational Discriminator Bottleneck: Improving Imitation Learning, Inverse RL, and GANs by Constraining Information Flow""","['reinforcement learning', 'generative adversarial networks', 'imitation learning', 'inverse reinforcement learning', 'information bottleneck']","""Adversarial learning methods have been proposed for a wide range of applications, but the training of adversarial models can be notoriously unstable. Effectively balancing the performance of the generator and discriminator is critical, since a discriminator that achieves very high accuracy will produce relatively uninformative gradients. In this work, we propose a simple and general technique to constrain information flow in the discriminator by means of an information bottleneck. By enforcing a constraint on the mutual information between the observations and the discriminator's internal representation, we can effectively modulate the discriminator's accuracy and maintain useful and informative gradients. We demonstrate that our proposed variational discriminator bottleneck (VDB) leads to significant improvements across three distinct application areas for adversarial learning algorithms. Our primary evaluation studies the applicability of the VDB to imitation learning of dynamic continuous control skills, such as running. We show that our method can learn such skills directly from raw video demonstrations, substantially outperforming prior adversarial imitation learning methods. The VDB can also be combined with adversarial inverse reinforcement learning to learn parsimonious reward functions that can be transferred and re-optimized in new settings. Finally, we demonstrate that VDB can train GANs more effectively for image generation, improving upon a number of prior stabilization methods.""","""The paper proposes a simple and general technique based on the information bottleneck to constrain the information flow in the discriminator of adversarial models. It helps to train by maintaining informative gradients. While the information bottleneck is not novel, its application in adversarial learning to my knowledge is, and the empirical evaluation demonstrates impressive performance on a broad range of applications. Therefore, the paper should clearly be accepted. """
256,"""Learning Physics Priors for Deep Reinforcement Learing""","['Model-Based Reinforcement Learning', 'Intuitive Physics']","""While model-based deep reinforcement learning (RL) holds great promise for sample efficiency and generalization, learning an accurate dynamics model is challenging and often requires substantial interactions with the environment. Further, a wide variety of domains have dynamics that share common foundations like the laws of physics, which are rarely exploited by these algorithms. Humans often acquire such physics priors that allow us to easily adapt to the dynamics of any environment. In this work, we propose an approach to learn such physics priors and incorporate them into an RL agent. Our method involves pre-training a frame predictor on raw videos and then using it to initialize the dynamics prediction model on a target task. Our prediction model, SpatialNet, is designed to implicitly capture localized physical phenomena and interactions. We show the value of incorporating this prior through empirical experiments on two different domains a newly created PhysWorld and games from the Atari benchmark, outperforming competitive approaches and demonstrating effective transfer learning.""","""The paper suggests a new way to learn a physics prior, in an action-free way from raw frames. The idea is to ""learn the common rules of physics"" in some sense (from purely visual observations) and use that as pre-training. The authors made a number of experiments in response to the reviewer concerns, but the submission still fell short of their expectations. In the post-rebuttal discussion, the reviewers mentioned that it's not clear how SpatialNet is different from a ConvLSTM, mentioned the writing quality and the fact that the ""physics prior"" is really quite close to what others call video prediction in other baselines. """
257,"""An Alarm System for Segmentation Algorithm Based on Shape Model""","['segmentation evaluation', 'shape feature', 'variational auto-encoder']","""It is usually hard for a learning system to predict correctly on the rare events, and there is no exception for segmentation algorithms. Therefore, we hope to build an alarm system to set off alarms when the segmentation result is possibly unsatisfactory. One plausible solution is to project the segmentation results into a low dimensional feature space, and then learn classifiers/regressors in the feature space to predict the qualities of segmentation results. In this paper, we form the feature space using shape feature which is a strong prior information shared among different data, so it is capable to predict the qualities of segmentation results given different segmentation algorithms on different datasets. The shape feature of a segmentation result is captured using the value of loss function when the segmentation result is tested using a Variational Auto-Encoder(VAE). The VAE is trained using only the ground truth masks, therefore the bad segmentation results with bad shapes become the rare events for VAE and will result in large loss value. By utilizing this fact, the VAE is able to detect all kinds of shapes that are out of the distribution of normal shapes in ground truth (GT). Finally, we learn the representation in the one-dimensional feature space to predict the qualities of segmentation results. We evaluate our alarm system on several recent segmentation algorithms for the medical segmentation task. The segmentation algorithms perform differently on different datasets, but our system consistently provides reliable prediction on the qualities of segmentation results. ""","""The authors present a method using a VAE to model segmentation masks directly. Errors in reconstruction of masks by the VAE indicate that the mask may be outside the distribution of common mask shapes, and are used to predict poor quality segmentation scenarios that fall outside the distribution of common segmentations. Pros: + R2: Technical idea is interesting, and a number of baselines used to compare. + R1 & R4: Method is novel. Cons: - R3 & R4: The method ignores the original input in its prediction, making the method wholly reliant on shape priors. In situations where the shape prior is weak, the method may be expected to fail. Authors have confirmed this, but not added any experiments to quantify its effect. - R4: The baseline regressor method is missing key details, which makes it impossible to judge if the comparison is fair (i.e. at minimum, number of learned parameters for each model, number of convolutional layers, structure of network, etc.). Authors have not provided these details. Authors have not investigated datasets with weak shape prior to see how methods compare in this setting. - R2: GANs can be used as a baseline. Authors confirmed, but did not supply results. Reviewers generally agree that the idea is novel, but the value of the approach cannot be determined due to missing baseline experiments, and missing details of baselines. Recommend reject in current form, but encourage authors to complete experiments. """
258,"""Supervised Community Detection with Line Graph Neural Networks""","['community detection', 'graph neural networks', 'belief propagation', 'energy landscape', 'non-backtracking matrix']","""Community detection in graphs can be solved via spectral methods or posterior inference under certain probabilistic graphical models. Focusing on random graph families such as the stochastic block model, recent research has unified both approaches and identified both statistical and computational detection thresholds in terms of the signal-to-noise ratio. By recasting community detection as a node-wise classification problem on graphs, we can also study it from a learning perspective. We present a novel family of Graph Neural Networks (GNNs) for solving community detection problems in a supervised learning setting. We show that, in a data-driven manner and without access to the underlying generative models, they can match or even surpass the performance of the belief propagation algorithm on binary and multiclass stochastic block models, which is believed to reach the computational threshold in these cases. In particular, we propose to augment GNNs with the non-backtracking operator defined on the line graph of edge adjacencies. The GNNs are achieved good performance on real-world datasets. In addition, we perform the first analysis of the optimization landscape of using (linear) GNNs to solve community detection problems, demonstrating that under certain simplifications and assumptions, the loss value at any local minimum is close to the loss value at the global minimum/minima.""","""This paper introduces a new graph convolutional neural network, called LGNN, and applied it to solve the community detection problem. The reviewers think LGNN yields a nice and useful extension of graph CNN, especially in using the line graph of edge adjacencies and a non-backtracking operator. The empirical evaluation shows that the new method provides a useful tool for real datasets. The reviewers raised some issues in writing and reference, for which the authors have provided clarification and modified the papers accordingly. """
259,"""DeepTwist: Learning Model Compression via Occasional Weight Distortion""","['deep learning', 'model compression', 'pruning', 'quantization', 'SVD', 'regularization', 'framework']","""Model compression has been introduced to reduce the required hardware resources while maintaining the model accuracy. Lots of techniques for model compression, such as pruning, quantization, and low-rank approximation, have been suggested along with different inference implementation characteristics. Adopting model compression is, however, still challenging because the design complexity of model compression is rapidly increasing due to additional hyper-parameters and computation overhead in order to achieve a high compression ratio. In this paper, we propose a simple and efficient model compression framework called DeepTwist which distorts weights in an occasional manner without modifying the underlying training algorithms. The ideas of designing weight distortion functions are intuitive and straightforward given formats of compressed weights. We show that our proposed framework improves compression rate significantly for pruning, quantization, and low-rank approximation techniques while the efforts of additional retraining and/or hyper-parameter search are highly reduced. Regularization effects of DeepTwist are also reported.""","""The authors propose a framework for compressing neural network models which involves applying a weight distortion function periodically as part of training. The proposed approach is relatively simple to implement, and is shown to work for weight pruning, low-rank compression and quantization, without sacrificing accuracy. However, the reviewers had a number of concerns about the work. Broadly, the reviewers felt that the work was incremental. Further, if the proposed techniques are important to get the approach to work well in practice, then the paper would be significantly strengthened by further analyses. Finally, the reviewers noted that the paper does not consider whether the specific weight pruning strategies result in a reduction of computational resources beyond potential storage savings, which would be important if this method is to be used in practice. Overall, the AC tends to agree with the reviewers criticisms. The authors are encouraged to address some of these issues in future revisions of the work. """
260,"""Sample Efficient Adaptive Text-to-Speech""","['few shot', 'meta learning', 'text to speech', 'wavenet']","""We present a meta-learning approach for adaptive text-to-speech (TTS) with few data. During training, we learn a multi-speaker model using a shared conditional WaveNet core and independent learned embeddings for each speaker. The aim of training is not to produce a neural network with fixed weights, which is then deployed as a TTS system. Instead, the aim is to produce a network that requires few data at deployment time to rapidly adapt to new speakers. We introduce and benchmark three strategies: (i) learning the speaker embedding while keeping the WaveNet core fixed, (ii) fine-tuning the entire architecture with stochastic gradient descent, and (iii) predicting the speaker embedding with a trained neural network encoder. The experiments show that these approaches are successful at adapting the multi-speaker neural network to new speakers, obtaining state-of-the-art results in both sample naturalness and voice similarity with merely a few minutes of audio data from new speakers.""","""The paper benchmarks three strategies to adapt an existing TTS system (based on WaveNet) to new speakers. The paper is clearly written. The models and adaptation strategies are not very novel, but still a scientific contribution. Overall, the experimental results are detailed and convincing. The rebuttals addressed some of the concerns. This is a welcomed contribution to ICLR 2019."""
261,"""Coverage and Quality Driven Training of Generative Image Models""","['deep learning', 'generative modeling', 'unsupervised learning', 'maximum likelihood', 'adversarial learning', 'gan', 'vae']","""Generative modeling of natural images has been extensively studied in recent years, yielding remarkable progress. Current state-of-the-art methods are either based on maximum likelihood estimation or adversarial training. Both methods have their own drawbacks, which are complementary in nature. The first leads to over-generalization as the maximum likelihood criterion encourages models to cover the support of the training data by heavily penalizing small masses assigned to training data. Simplifying assumptions in such models limits their capacity and makes them spill mass on unrealistic samples. The second leads to mode-dropping since adversarial training encourages high quality samples from the model, but only indirectly enforces diversity among the samples. To overcome these drawbacks we make two contributions. First, we propose a model that extends variational autoencoders by using deterministic invertible transformation layers to map samples from the decoder to the image space. This induces correlations among the pixels given the latent variables, improving over factorial decoders commonly used in variational autoencoders. Second, we propose a unified training approach that leverages coverage and quality based criteria. Our models obtain likelihood scores competitive with state-of-the-art likelihood-based models, while achieving sample quality typical of adversarially trained networks. ""","""The overall view of the reviewers is that the paper is not quite good enough as it stands. The reviewers also appreciate the contributions so taking the comments into account and resubmit elsewhere is encouraged. """
262,"""Unsupervised Word Discovery with Segmental Neural Language Models""",[],"""We propose a segmental neural language model that combines the representational power of neural networks and the structure learning mechanism of Bayesian nonparametrics, and show that it learns to discover semantically meaningful units (e.g., morphemes and words) from unsegmented character sequences. The model generates text as a sequence of segments, where each segment is generated either character-by-character from a sequence model or as a single draw from a lexical memory that stores multi-character units. Its parameters are fit to maximize the marginal likelihood of the training data, summing over all segmentations of the input, and its hyperparameters are likewise set to optimize held-out marginal likelihood. To prevent the model from overusing the lexical memory, which leads to poor generalization and bad segmentation, we introduce a differentiable regularizer that penalizes based on the expected length of each segment. To our knowledge, this is the first demonstration of neural networks that have predictive distributions better than LSTM language models and also infer a segmentation into word-like units that are competitive with the best existing word discovery models.""","""a major issue or complaint from the reviewers seems to come from perhaps a wrong framing of this submission. i believe the framing of this work should have been a better language model (or translation model) with word discovery as an awesome side effect, which i carefully guess would've been a perfectly good story assuming that the perplexity result in Table 4 translates to text with blank spaces left in (it is not possible tell whether this is the case from the text alone.) even discounting R1, who i disagree with on quite a few points, the other reviewers also did not see much of the merit of this work, again probably due to the framing issue above. i highly encourage the authors to change the framing, evaluate it as a usual sequence model on various benchmarks and resubmit it to another venue."""
263,"""Learning Mixed-Curvature Representations in Product Spaces""","['embeddings', 'non-Euclidean geometry', 'manifolds', 'geometry of data']","""The quality of the representations achieved by embeddings is determined by how well the geometry of the embedding space matches the structure of the data. Euclidean space has been the workhorse for embeddings; recently hyperbolic and spherical spaces have gained popularity due to their ability to better embed new types of structured data---such as hierarchical data---but most data is not structured so uniformly. We address this problem by proposing learning embeddings in a product manifold combining multiple copies of these model spaces (spherical, hyperbolic, Euclidean), providing a space of heterogeneous curvature suitable for a wide variety of structures. We introduce a heuristic to estimate the sectional curvature of graph data and directly determine an appropriate signature---the number of component spaces and their dimensions---of the product manifold. Empirically, we jointly learn the curvature and the embedding in the product space via Riemannian optimization. We discuss how to define and compute intrinsic quantities such as means---a challenging notion for product manifolds---and provably learnable optimization functions. On a range of datasets and reconstruction tasks, our product space embeddings outperform single Euclidean or hyperbolic spaces used in previous works, reducing distortion by 32.55% on a Facebook social network dataset. We learn word embeddings and find that a product of hyperbolic spaces in 50 dimensions consistently improves on baseline Euclidean and hyperbolic embeddings, by 2.6 points in Spearman rank correlation on similarity tasks and 3.4 points on analogy accuracy. ""","""This paper proposes a novel framework for tractably learning non-eucliean embeddings that are product spaces formed by hyperbolic, spherical, and Euclidean components, providing a heterogenous mix of curvature properties. On several datasets, these product space embeddings outperform single Euclidean or hyperbolic spaces. The reviewers unanimously recommend acceptance."""
264,"""Downsampling leads to Image Memorization in Convolutional Autoencoders""","['Memorization in Deep Learning', 'Convolutional Autoencoders']","""Memorization of data in deep neural networks has become a subject of significant research interest. In this paper, we link memorization of images in deep convolutional autoencoders to downsampling through strided convolution. To analyze this mechanism in a simpler setting, we train linear convolutional autoencoders and show that linear combinations of training data are stored as eigenvectors in the linear operator corresponding to the network when downsampling is used. On the other hand, networks without downsampling do not memorize training data. We provide further evidence that the same effect happens in nonlinear networks. Moreover, downsampling in nonlinear networks causes the model to not only memorize just linear combinations of images, but individual training images. Since convolutional autoencoder components are building blocks of deep convolutional networks, we envision that our findings will shed light on the important phenomenon of memorization in over-parameterized deep networks. ""","""This paper studies the question of memorization within overparametrised neural networks. Specifically, the authors conjecture that memorization is linked to the downsampling operators present in many convolutional autoencoders. All reviewers agreed that this is an interesting question that deserves further analysis. However, they also agreed that in its current form, the paper lacks mathematical and experimental rigor. In particular, the paper does not follow the basic mathematical standards of proving any stated proposition/theorem, instead mixing empirical with mathematical proofs. The AC fully agrees with the points raised by reviewers, and therefore recommends rejection at this point, encouraging the authors to address these important points before resubmitting their work. """
265,"""Accumulation Bit-Width Scaling For Ultra-Low Precision Training Of Deep Networks""","['reduced precision floating-point', 'partial sum accumulation bit-width', 'deep learning', 'training']","""Efforts to reduce the numerical precision of computations in deep learning training have yielded systems that aggressively quantize weights and activations, yet employ wide high-precision accumulators for partial sums in inner-product operations to preserve the quality of convergence. The absence of any framework to analyze the precision requirements of partial sum accumulations results in conservative design choices. This imposes an upper-bound on the reduction of complexity of multiply-accumulate units. We present a statistical approach to analyze the impact of reduced accumulation precision on deep learning training. Observing that a bad choice for accumulation precision results in loss of information that manifests itself as a reduction in variance in an ensemble of partial sums, we derive a set of equations that relate this variance to the length of accumulation and the minimum number of bits needed for accumulation. We apply our analysis to three benchmark networks: CIFAR-10 ResNet 32, ImageNet ResNet 18 and ImageNet AlexNet. In each case, with accumulation precision set in accordance with our proposed equations, the networks successfully converge to the single precision floating-point baseline. We also show that reducing accumulation precision further degrades the quality of the trained network, proving that our equations produce tight bounds. Overall this analysis enables precise tailoring of computation hardware to the application, yielding area- and power-optimal systems.""","""The authors present a theoretical and practical study on low-precision training of neural networks. They introduce the notion of variance retention ratio (VRR) that determines the accumulation bit-width for precise tailoring of computation hardware. Empirically, the authors show that their theoretical result extends to practical implementation in three standard benchmarks. A criticism of the paper has been certain hyperparameters that a reviewer found to be chosen rather arbitrarily, but I think the reviewers do a reasonable job in rebutting it. Overall, there is consensus that the paper presents an interesting framework and does both practical and empirical analysis, and it should be accepted."""
266,"""Beyond Pixel Norm-Balls: Parametric Adversaries using an Analytically Differentiable Renderer""","['adversarial examples', 'norm-balls', 'differentiable renderer']","""Many machine learning image classifiers are vulnerable to adversarial attacks, inputs with perturbations designed to intentionally trigger misclassification. Current adversarial methods directly alter pixel colors and evaluate against pixel norm-balls: pixel perturbations smaller than a specified magnitude, according to a measurement norm. This evaluation, however, has limited practical utility since perturbations in the pixel space do not correspond to underlying real-world phenomena of image formation that lead to them and has no security motivation attached. Pixels in natural images are measurements of light that has interacted with the geometry of a physical scene. As such, we propose a novel evaluation measure, parametric norm-balls, by directly perturbing physical parameters that underly image formation. One enabling contribution we present is a physically-based differentiable renderer that allows us to propagate pixel gradients to the parametric space of lighting and geometry. Our approach enables physically-based adversarial attacks, and our differentiable renderer leverages models from the interactive rendering literature to balance the performance and accuracy trade-offs necessary for a memory-efficient and scalable adversarial data augmentation workflow.""","""The paper describes the use of differentiable physics based rendering schemes to generate adversarial perturbations that are constrained by physics of image formation. The paper puts forth a fairly novel approach to tackle an interesting question. However, some of the claims made regarding the ""believability"" of the adversarial examples produced by existing techniques are not fully supported. Also, the adversarial examples produced by the proposed techniques are not fully ""physical"" at least compared to how ""physical"" adversarial examples presented in some of the prior work were. Overall though this paper constitutes a valuable contribution. """
267,"""Greedy Attack and Gumbel Attack: Generating Adversarial Examples for Discrete Data""",['Adversarial Examples'],"""We present a probabilistic framework for studying adversarial attacks on discrete data. Based on this framework, we derive a perturbation-based method, Greedy Attack, and a scalable learning-based method, Gumbel Attack, that illustrate various tradeoffs in the design of attacks. We demonstrate the effectiveness of these methods using both quantitative metrics and human evaluation on various state-of-the-art models for text classification, including a word-based CNN, a character-based CNN and an LSTM. As an example of our results, we show that the accuracy of character-based convolutional networks drops to the level of random selection by modifying only five characters through Greedy Attack.""","""I appreciate the willingness of the authors to engage in vigorous discussion about their paper. Although several reviewers support accepting this submission, I do not find their arguments for acceptance convincing. The paper considers automated methods for finding errors in text classification models. I believe it is valuable to study the errors our models make in order to understand when they work well and how to improve them. Crucially, in the later case, we should demonstrate how to use the errors we find to close the loop and create better models. A paper about techniques to find errors for text models should make a sufficiently large contribution to be accepted. I view the following hypothetical contributions as the most salient in this specific case thus my decision reduces to determining if any of these conditions have been met. A paper need not achieve all of these things, any one of them would suffice: 1. Show that the errors found can be used to meaningfully improve the models. This requires building a better model than the one probed by the method and convincingly demonstrating that it is superior in an important way that is relevant to the original goals of the application. Ideally it would also consider alternative, simpler ways to improve the models (e.g. making them larger). 2. Show that errors are difficult to find, but that the proposed method is nonetheless capable of finding errors and that the method is non-obvious to a researcher in the field. This is not applicable here because errors are extremely easy to find on the test set and from labeling more data. If we demand an automated method, then the greedy algorithm does not qualify as sufficiently non-obvious and it seems to work fine, making the Gumbel method unnecessary. 3. Show that the particular specific errors found are qualitatively different from other errors in their implications and that they provide a unique and important insight. I do not believe this submission attempts to show this type of contribution. One example of this type of paper would be a paper that does a comparative study of the errors that different models make and finds something interesting (potentially yielding a path to improved models). 4. Generate a new, more difficult/interesting, dataset by finding errors of one or more trained models Given that the authors use human labelers to validate examples this is potentially another path. Here is an example of a paper using adversarial techniques in this way: pseudo-url However, I believe the paper would need to be rethought and rewritten to make this sort of contribution. Ultimately, the authors and reviews supporting acceptance must explain the contribution succinctly and convincingly. The reviewers most strongly advocating for accepting this submission seem to be saying that there is a valuable new method and probabilistic framework proposed here for finding model errors. I believe researchers in the field could have easily come up with the greedy algorithm (a standard approach to discrete optimization problems) proposed here without needing to read the paper. Furthermore, I believe the other more complicated Gumbel algorithm proposed is not necessary given the similarly effective and simpler greedy algorithm. If the authors believe that the Gumbel algorithm provides application-relevant advantages over the greedy algorithm, then they should specify how these errors will be used and rewrite the paper to make the greedy algorithm a baseline. However, I do not believe the experimental results support this idea. """
268,"""Estimating Information Flow in DNNs""","['information theory', 'representation learning', 'deep learning', 'differential entropy estimation']","""We study the evolution of internal representations during deep neural network (DNN) training, aiming to demystify the compression aspect of the information bottleneck theory. The theory suggests that DNN training comprises a rapid fitting phase followed by a slower compression phase, in which the mutual information I(X;T) between the input X and internal representations T decreases. Several papers observe compression of estimated mutual information on different DNN models, but the true I(X;T) over these networks is provably either constant (discrete X) or infinite (continuous X). This work explains the discrepancy between theory and experiments, and clarifies what was actually measured by these past works. To this end, we introduce an auxiliary (noisy) DNN framework for which I(X;T) is a meaningful quantity that depends on the network's parameters. This noisy framework is shown to be a good proxy for the original (deterministic) DNN both in terms of performance and the learned representations. We then develop a rigorous estimator for I(X;T) in noisy DNNs and observe compression in various models. By relating I(X;T) in the noisy DNN to an information-theoretic communication problem, we show that compression is driven by the progressive clustering of hidden representations of inputs from the same class. Several methods to directly monitor clustering of hidden representations, both in noisy and deterministic DNNs, are used to show that meaningful clusters form in the T space. Finally, we return to the estimator of I(X;T) employed in past works, and demonstrate that while it fails to capture the true (vacuous) mutual information, it does serve as a measure for clustering. This clarifies the past observations of compression and isolates the geometric clustering of hidden representations as the true phenomenon of interest.""","""This paper studies the compression aspect of the information bottleneck. It seeks to clarify a debate about the evolution of mutual information between inputs and representations during training in neural networks. The paper discusses numerous ideas and techniques and arrives at valuable conclusions. A concern is that parts of the paper (theoretical parts) are intended for a separate paper, and are included in the paper only for reference. This means that the actual contribution of the present paper is mostly on the experimental part. Nonetheless, the discussion derived from the theory and experiments seem valuable in the ongoing discussion of this topic. In any case, I encourage the authors to make efforts to obtain a transparent separation of the different pieces of work. A concern was raised that the current paper mainly addresses a discussion that originated in a paper that has not passed peer review. On the other hand, this discussion does occupy many researchers and justifies the analysis, even if the originating paper has not been published in a peer reviewed format. All reviewers are confident in their assessment. Two of them regard the paper positively and one of them regards the paper as ok, but not good enough, with main criticism in relation to the points discussed above. Although the paper is in any case very good, unfortunately it does not reach the very high bar for acceptance at this ICLR. """
269,"""Playing the Game of Universal Adversarial Perturbations""","['adversarial perturbations', 'universal adversarial perturbations', 'game theory', 'robust machine learning']","""We study the problem of learning classifiers robust to universal adversarial perturbations. While prior work approaches this problem via robust optimization, adversarial training, or input transformation, we instead phrase it as a two-player zero-sum game. In this new formulation, both players simultaneously play the same game, where one player chooses a classifier that minimizes a classification loss whilst the other player creates an adversarial perturbation that increases the same loss when applied to every sample in the training set. By observing that performing a classification (respectively creating adversarial samples) is the best response to the other player, we propose a novel extension of a game-theoretic algorithm, namely fictitious play, to the domain of training robust classifiers. Finally, we empirically show the robustness and versatility of our approach in two defence scenarios where universal attacks are performed on several image classification datasets -- CIFAR10, CIFAR100 and ImageNet.""","""Reviewers mostly recommended to reject after engaging with the authors, with one reviewer slightly suggesting to accept, but with confidence 1. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit."""
270,"""The Conditional Entropy Bottleneck""","['representation learning', 'information theory', 'uncertainty', 'out-of-distribution detection', 'adversarial example robustness', 'generalization', 'objective function']","""We present a new family of objective functions, which we term the Conditional Entropy Bottleneck (CEB). These objectives are motivated by the Minimum Necessary Information (MNI) criterion. We demonstrate the application of CEB to classification tasks. We show that CEB gives: well-calibrated predictions; strong detection of challenging out-of-distribution examples and powerful whitebox adversarial examples; and substantial robustness to those adversaries. Finally, we report that CEB fails to learn from information-free datasets, providing a possible resolution to the problem of generalization observed in Zhang et al. (2016).""","""This paper proposes a criterion for representation learning, minimum necessary information, which states that for a task defined by some joint probability distribution P(X,Y) and the goal of (for example) predicting Y from X, a learned representation of X, denoted Z, should satisfy the equality I(X;Y) = I(X;Z) = I(Y;Z). The authors then propose an objective function, the conditional entropy bottleneck (CEB), to ensure that a learned representation satisfies the minimum necessary information criterion, and a variational approximation to the conditional entropy bottleneck that can be parameterized using deep networks and optimized with standard methods such as stochastic gradient descent. The authors also relate the conditional entropy bottleneck to the information bottleneck Lagrangian proposed by Tishby, showing that the CEB corresponds to the information bottleneck with = 0.5. An important contribution of this work is that it gives a theoretical justification for selecting a specific value of rather than testing multiple values. Experiments on Fashion-MNIST show that, in comparison to a deterministic classifier and to variational information bottleneck models with in {0.01, 0.1, 0.5}, the CEB model achieves good accuracy and calibration, is competitive at detecting out-of-distribution inputs, and is more resistant to white-box adversarial attacks. Another experiment demonstrates that a model trained with the CEB criterion is *unable* to memorize a randomly labeled version of Fashion-MNIST. There was a strong difference of opinion between the reviewers on this paper. One reviewer (R1) dismissed the work as trivial. The authors rebutted this claim in their response and revision, and R1 failed to participate in the discussion, so the AC strongly discounted this review. The other two reviewers had some concerns about the paper, most of which were addressed by the revision. But, crucially, some concerns still remain. R4 would like more theoretical rigor in the paper, while R2 would like a direct comparison against MINE and CPC. In the end, the AC thinks that this paper needs just a bit more work to address these concerns. The authors are encouraged to revise this work and submit it to another machine learning venue."""
271,"""Rectified Gradient: Layer-wise Thresholding for Sharp and Coherent Attribution Maps""","['Interpretability', 'Attribution Method', 'Attribution Map']","""Saliency map, or the gradient of the score function with respect to the input, is the most basic means of interpreting deep neural network decisions. However, saliency maps are often visually noisy. Although several hypotheses were proposed to account for this phenomenon, there is no work that provides a rigorous analysis of noisy saliency maps. This may be a problem as numerous advanced attribution methods were proposed under the assumption that the existing hypotheses are true. In this paper, we identify the cause of noisy saliency maps. Then, we propose Rectified Gradient, a simple method that significantly improves saliency maps by alleviating that cause. Experiments showed effectiveness of our method and its superiority to other attribution methods. Codes and examples for the experiments will be released in public.""","""The main goal of the submission is to figure out a way to produce less ""noisy"" saliency maps. The RectGrad method uses some thresholding during backprop, like Guided Backprop. The visuals of the proposed method are good, but the reviewers rightfully point out that evaluating whether the proposed method is any good is not obvious. The ROAR/KAR results are perhaps not telling the whole story (and the authors claim that RectGrad is not expected to get a high ROAR score, but I would like to see this developed more in a further version of this work). Generally, I feel like there was a healthy back and forth between authors and R3 on the main concerns of this work. I agree that the mathematical justification for RectGrad seems not fully developed. Given all of these concerns, at this point I cannot support acceptance of this work at ICLR."""
272,"""Algorithmic Framework for Model-based Deep Reinforcement Learning with Theoretical Guarantees""","['model-based reinforcement learning', 'sample efficiency', 'deep reinforcement learning']","""Model-based reinforcement learning (RL) is considered to be a promising approach to reduce the sample complexity that hinders model-free RL. However, the theoretical understanding of such methods has been rather limited. This paper introduces a novel algorithmic framework for designing and analyzing model-based RL algorithms with theoretical guarantees. We design a meta-algorithm with a theoretical guarantee of monotone improvement to a local maximum of the expected reward. The meta-algorithm iteratively builds a lower bound of the expected reward based on the estimated dynamical model and sample trajectories, and then maximizes the lower bound jointly over the policy and the model. The framework extends the optimism-in-face-of-uncertainty principle to non-linear dynamical models in a way that requires no explicit uncertainty quantification. Instantiating our framework with simplification gives a variant of model-based RL algorithms Stochastic Lower Bounds Optimization (SLBO). Experiments demonstrate that SLBO achieves the state-of-the-art performance when only 1M or fewer samples are permitted on a range of continuous control benchmark tasks.""","""This paper proposes model-based reinforcement learning algorithms that have theoretical guarantees. These methods are shown to good results on Mujuco benchmark tasks. All of the reviewers have given a reasonable score to the paper, and the paper can be accepted."""
273,"""The Natural Language Decathlon: Multitask Learning as Question Answering""","['multitask learning', 'natural language processing', 'question answering', 'machine translation', 'relation extraction', 'semantic parsing', 'commensense reasoning', 'summarization', 'entailment', 'sentiment', 'dialog']","""Deep learning has improved performance on many natural language processing (NLP) tasks individually. However, general NLP models cannot emerge within a paradigm that focuses on the particularities of a single metric, dataset, and task. We introduce the Natural Language Decathlon (decaNLP), a challenge that spans ten tasks: question answering, machine translation, summarization, natural language inference, sentiment analysis, semantic role labeling, relation extraction, goal-oriented dialogue, semantic parsing, and commonsense pronoun resolution. We cast all tasks as question answering over a context. Furthermore, we present a new multitask question answering network (MQAN) that jointly learns all tasks in decaNLP without any task-specific modules or parameters more effectively than sequence-to-sequence and reading comprehension baselines. MQAN shows improvements in transfer learning for machine translation and named entity recognition, domain adaptation for sentiment analysis and natural language inference, and zero-shot capabilities for text classification. We demonstrate that the MQAN's multi-pointer-generator decoder is key to this success and that performance further improves with an anti-curriculum training strategy. Though designed for decaNLP, MQAN also achieves state of the art results on the WikiSQL semantic parsing task in the single-task setting. We also release code for procuring and processing data, training and evaluating models, and reproducing all experiments for decaNLP.""","""This paper presents a new multi-task training and evaluation set up called the Natural Language Decathlon, and evaluates models on it. While this AC is sympathetic to any work which introduces new datasets and evaluation tasks, the reviewers agreed amongst themselves that the paper is not quite ready for publication. The main concern is that multi-task learning should show benefits of transferring representations or other model components between tasks, demonstrating better generalisation and less task-specific overfitting, but that the results in the paper do not properly show this effect. A more thorough study of which tasks ""interact constructively"" and what model changes can properly exploit this needs to be done. With this further work, the AC has no doubt that this dataset and task suite, and associated models, will be very valuable to the NLP community. I should note that there were some issues during the review period which lead to AC-confidential communication between AC and authors, and AC and reviewers, to be leaked to the reviewers. It was due to an OpenReview bug, and no party is at fault. Through private discussion with the interested parties, we were able to resolve this matter, and through careful examination of the discussion, I am satisfied that the reviews and final recommendations of the reviewers were properly argued for and presented in good faith."""
274,"""Local Critic Training of Deep Neural Networks""","['inter-layer locking', 'local critic network', 'backpropagation', 'convolutional neural network', 'structural optimization', 'progress inference', 'ensemble inference']","""This paper proposes a novel approach to train deep neural networks by unlocking the layer-wise dependency of backpropagation training. The approach employs additional modules called local critic networks besides the main network model to be trained, which are used to obtain error gradients without complete feedforward and backward propagation processes. We propose a cascaded learning strategy for these local networks. In addition, the approach is also useful from multi-model perspectives, including structural optimization of neural networks, computationally efficient progressive inference, and ensemble classification for performance improvement. Experimental results show the effectiveness of the proposed approach and suggest guidelines for determining appropriate algorithm parameters.""","""This paper proposes a new training approach for deep neural interfaces. The idea is to bootstrap from critics of other layers instead of using the final loss as target. The method is evaluated of CIFAR-10 and CIFAR-100 and found to improve performance slightly upon Sobolev training while being simpler. The reviewers found the idea interesting but were concerned about the strength of the experimental results. The datasets are similar and the significance of the results is not clear. The revision submitted by the authors was only able to address some of these issues such as the evaluation protocol."""
275,"""Feature prioritization and regularization improve standard accuracy and adversarial robustness""","['adversarial robustness', 'feature prioritization', 'regularization']","""Adversarial training has been successfully applied to build robust models at a certain cost. While the robustness of a model increases, the standard classification accuracy declines. This phenomenon is suggested to be an inherent trade-off. We propose a model that employs feature prioritization by a nonlinear attention module and pseudo-formula feature regularization to improve the adversarial robustness and the standard accuracy relative to adversarial training. The attention module encourages the model to rely heavily on robust features by assigning larger weights to them while suppressing non-robust features. The regularizer encourages the model to extracts similar features for the natural and adversarial images, effectively ignoring the added perturbation. In addition to evaluating the robustness of our model, we provide justification for the attention module and propose a novel experimental strategy that quantitatively demonstrates that our model is almost ideally aligned with salient data characteristics. Additional experimental results illustrate the power of our model relative to the state of the art methods.""","""The paper proposes an attention mechanism to focus on robust features in the context of adversarial attacks. Reviewers asked for more intuition, more results, and more experiments with different attack/defense models. Authors have added experimental results and provided some intuition of their proposed approach. Overall, reviewers still think the novelty is too thin and recommend rejection. I concur with them."""
276,"""Generalizable Adversarial Training via Spectral Normalization""","['Adversarial attacks', 'adversarial training', 'spectral normalization', 'generalization guarantee']","""Deep neural networks (DNNs) have set benchmarks on a wide array of supervised learning tasks. Trained DNNs, however, often lack robustness to minor adversarial perturbations to the input, which undermines their true practicality. Recent works have increased the robustness of DNNs by fitting networks using adversarially-perturbed training samples, but the improved performance can still be far below the performance seen in non-adversarial settings. A significant portion of this gap can be attributed to the decrease in generalization performance due to adversarial training. In this work, we extend the notion of margin loss to adversarial settings and bound the generalization error for DNNs trained under several well-known gradient-based attack schemes, motivating an effective regularization scheme based on spectral normalization of the DNN's weight matrices. We also provide a computationally-efficient method for normalizing the spectral norm of convolutional layers with arbitrary stride and padding schemes in deep convolutional networks. We evaluate the power of spectral normalization extensively on combinations of datasets, network architectures, and adversarial training schemes.""","""Adversarial training has quickly become important for training robust neural networks. However this training generally results in poor generalization behavior. This paper proposes using margin loss with adversarial training for better generalization. The paper provides generalization bounds for this adversarial training setup motivating the use of spectral regularization. The experimental results using the spectral regularization with adversarial training are very promising and all the reviewers agree that they show non-trivial improvement. Even though the spectral regularization techniques have been tried in different settings, hence of limited novelty, the experimental results in the paper are encouraging and I believe will motivate further study on this topic. Reviewers also opined that the writing in the paper is currently not that great with limited explanation of the theoretical results. More discussions interpreting the theoretical results and their significance can help the readers appreciate the paper better."""
277,"""Discriminative Active Learning""","['Active Learning', 'Neural Networks']","""We propose a new batch mode active learning algorithm designed for neural networks and large query batch sizes. The method, Discriminative Active Learning (DAL), poses active learning as a binary classification task, attempting to choose examples to label in such a way as to make the labeled set and the unlabeled pool indistinguishable. Experimenting on image classification tasks, we empirically show our method to be on par with state of the art methods in medium and large query batch sizes, while being simple to implement and also extend to other domains besides classification tasks. Our experiments also show that none of the state of the art methods of today are clearly better than uncertainty sampling, negating some of the reported results in the recent literature.""","""This paper proposes a novel and interesting active learning approach, that trains a classifier to discriminate between the examples in the labeled and unlabeled data at each iteration. The top few samples that are most likely to be from the unlabeled set as per this classifier are selected to be labeled by an oracle, and are moved to the labeled training examples bin in the next iteration. The idea is simple and clear and is shown to have a principled basis and theoretical background, related to GANs and to previous results from the literature. Experiments performed on CIFAR-10 and MNIST benchmarks demonstrate good results in comparison to baselines. During the review period, authors considered most of the suggestions by the reviewers and updated the paper. Although the proposed method is similar to density-based active learning methods, as also suggested by the reviewers, baselines do not include such approaches in the comparison experiments."""
278,"""Adversarial Sampling for Active Learning""","['active learning', 'adversarial training', 'GAN']","""This paper proposes ASAL, a new pool based active learning method that generates high entropy samples. Instead of directly annotating the synthetic samples, ASAL searches similar samples from the pool and includes them for training. Hence, the quality of new samples is high and annotations are reliable. ASAL is particularly suitable for large data sets because it achieves a better run-time complexity (sub-linear) for sample selection than traditional uncertainty sampling (linear). We present a comprehensive set of experiments on two data sets and show that ASAL outperforms similar methods and clearly exceeds the established baseline (random sampling). In the discussion section we analyze in which situations ASAL performs best and why it is sometimes hard to outperform random sample selection. To the best of our knowledge this is the first adversarial active learning technique that is applied for multiple class problems using deep convolutional classifiers and demonstrates superior performance than random sample selection.""","""The paper proposes adversarial sampling for pool-based active learning. The reviewers and AC note the critical potential weaknesses on experimental results: it is far from being surprising the proposed method is better than random sampling. Ideally, one has to reduce the complexity under keeping the state-of-art performance. Otherwise, it is hard to claim the proposed method is fundamentally better than prior ones, although their targets might be different. AC thinks the proposed method has potential and is interesting, but decided that the authors need more works to publish."""
279,"""Feature Transformers: A Unified Representation Learning Framework for Lifelong Learning""","['continual learning', 'deep learning', 'lifelong learning', 'new task learning', 'representation learning']","""Despite the recent advances in representation learning, lifelong learning continues to be one of the most challenging and unconquered problems. Catastrophic forgetting and data privacy constitute two of the important challenges for a successful lifelong learner. Further, existing techniques are designed to handle only specific manifestations of lifelong learning, whereas a practical lifelong learner is expected to switch and adapt seamlessly to different scenarios. In this paper, we present a single, unified mathematical framework for handling the myriad variants of lifelong learning, while alleviating these two challenges. We utilize an external memory to store only the features representing past data and learn richer and newer representations incrementally through transformation neural networks - feature transformers. We define, simulate and demonstrate exemplary performance on a realistic lifelong experimental setting using the MNIST rotations dataset, paving the way for practical lifelong learners. To illustrate the applicability of our method in data sensitive domains like healthcare, we study the pneumothorax classification problem from X-ray images, achieving near gold standard performance. We also benchmark our approach with a number of state-of-the art methods on MNIST rotations and iCIFAR100 datasets demonstrating superior performance.""","""The paper proposes a framework for continual/lifelong learning that has potential to overcome the problems of catastrophic forgetting and data privacy. R1, R2 and AC agree that the proposed method is not suitable for lifelong learning in its current state as it linearly increases memory and computational cost over time (for storing features of all points in the past and increasing model capacity with new tasks) without account for budget constraints. The authors responded in their rebuttal that the data is not stored in the original form, but using feature representation (which is important for privacy issues). The main concern, however, was about the fact that one has to store information about all previous data points which is not feasible in lifelong learning. In the revision the authors have tried to address some of the R1s and R2s suggestion about taking into account the budget constraints. However more in-depth analysis is required to assess feasibility and advantage of the proposed approach. The authors motivate some of the key elements in their model as to protect privacy. However no actual study was conducted to show that this has been achieved. The comments from R3 were too brief and did not have a substantial impact on the decision. In conclusion, AC suggests that the authors prepare a major revision addressing suitability of the proposed approach for continual learning under budget constraints and for privacy preservation and resubmit for another round of reviews. """
280,"""Deep processing of structured data""","['structured data', 'representation learning', 'deep neural networks']","""We construct a general unified framework for learning representation of structured data, i.e. data which cannot be represented as the fixed-length vectors (e.g. sets, graphs, texts or images of varying sizes). The key factor is played by an intermediate network called SAN (Set Aggregating Network), which maps a structured object to a fixed length vector in a high dimensional latent space. Our main theoretical result shows that for sufficiently large dimension of the latent space, SAN is capable of learning a unique representation for every input example. Experiments demonstrate that replacing pooling operation by SAN in convolutional networks leads to better results in classifying images with different sizes. Moreover, its direct application to text and graph data allows to obtain results close to SOTA, by simpler networks with smaller number of parameters than competitive models.""","""The reviewers agree this paper is not good enough for ICLR."""
281,"""A fast quasi-Newton-type method for large-scale stochastic optimisation""","['optimisation', 'large-scale', 'stochastic']","""During recent years there has been an increased interest in stochastic adaptations of limited memory quasi-Newton methods, which compared to pure gradient-based routines can improve the convergence by incorporating second order information. In this work we propose a direct least-squares approach conceptually similar to the limited memory quasi-Newton methods, but that computes the search direction in a slightly different way. This is achieved in a fast and numerically robust manner by maintaining a Cholesky factor of low dimension. This is combined with a stochastic line search relying upon fulfilment of the Wolfe condition in a backtracking manner, where the step length is adaptively modified with respect to the optimisation progress. We support our new algorithm by providing several theoretical results guaranteeing its performance. The performance is demonstrated on real-world benchmark problems which shows improved results in comparison with already established methods.""","""The paper investigates a novel formulation of a stochastic, quasi-Newton optimization strategy based on the natural idea of relaxing the secant conditions. This is an interesting and promising idea, but unfortunately none of the reviewers recommended acceptance. The reviewers unanimously fixated on weaknesses in the paper's technical presentation. In particular, the reviewers expressed some dissatisfaction with many aspects, including: - Key details of the experimental evaluation were omitted (particularly concerning configuration of the baseline competitors), which is an essential aspect of reproducibility. One consequence is that the reviewers were not confident in the veracity of the experimental comparison. - The reviewers struggled with a lack of clarity and accurate rendering of some key technical details. An example is dissatisfaction with the non-symmetry of the inverse Hessian approximation, which was not fully alleviated by the author responses. - The proposed approach does not appear to possess any intrinsic advantage over standard methods from a computational complexity perspective. I think this is promising work, but a careful revision that strengthened the underlying technical claims appears necessary to make this a solid contribution."""
282,"""Deep Recurrent Gaussian Process with Variational Sparse Spectrum Approximation""","['Deep Gaussian Process Model', 'Recurrent Model', 'State-Space Model', 'Nonlinear system identification', 'Dynamical modeling']","""Modeling sequential data has become more and more important in practice. Some applications are autonomous driving, virtual sensors and weather forecasting. To model such systems, so called recurrent models are frequently used. In this paper we introduce several new Deep Recurrent Gaussian Process (DRGP) models based on the Sparse Spectrum Gaussian Process (SSGP) and the improved version, called Variational Sparse Spectrum Gaussian Process (VSSGP). We follow the recurrent structure given by an existing DRGP based on a specific variational sparse Nystrm approximation, the recurrent Gaussian Process (RGP). Similar to previous work, we also variationally integrate out the input-space and hence can propagate uncertainty through the Gaussian Process (GP) layers. Our approach can deal with a larger class of covariance functions than the RGP, because its spectral nature allows variational integration in all stationary cases. Furthermore, we combine the (Variational) Sparse Spectrum ((V)SS) approximations with a well known inducing-input regularization framework. For the DRGP extension of these combined approximations and the simple (V)SS approximations an optimal variational distribution exists. We improve over current state of the art methods in prediction accuracy for experimental data-sets used for their evaluation and introduce a new data-set for engine control, named Emission.""","""This paper is concerned with combining past approximation methods to obtain a variant of Deep Recurrent GPs. While this variant is new, 2/3 reviewers make very overlapping points about this extension being obtained from a straightforward combination of previous ideas. Furthermore, R3 is not convinced that the approach is well motivated, beyond filling the gap in the literature. All reviewers also pointed out that the paper is very hard to read. The authors have improved the manuscript during the rebuttal, but the AC believes that the paper is still written in an unnecessarily complicated way. Overall the AC believes that this paper needs some more work, specifically in (a) improving its presentation (b) providing more technical insights about the methods (as suggested by R2 and R3), which could be a means of boosting the novelty. """
283,"""Explicit Recall for Efficient Exploration""","['Exploration', 'goal-directed', 'deep reinforcement learning', 'explicit memory']","""In this paper, we advocate the use of explicit memory for efficient exploration in reinforcement learning. This memory records structured trajectories that have led to interesting states in the past, and can be used by the agent to revisit those states more effectively. In high-dimensional decision making problems, where deep reinforcement learning is considered crucial, our approach provides a simple, transparent and effective way that can be naturally combined with complex, deep learning models. We show how such explicit memory may be used to enhance existing exploration algorithms such as intrinsically motivated ones and count-based ones, and demonstrate our method's advantages in various simulated environments.""","""The paper presents an explicit memory that directly contributes to more efficient exploration. It stores trajectories to novel states, that serve as training data to learn to reach those states again (through iterative sub-goals). The description of the method is quite clear, the method is not completely novel but has some merit. Most weaknesses of the paper come from the experimental section: too specific environments/solutions, lack of points of comparisons, lacking some details. We strongly encourage the authors to add additional experimental evidence, and details. In its current form, the paper is not sufficient for publication at ICLR 2019. Reviewers wanted to note that the blog post from Uber (""Go-Explore"") did _not_ affect their evaluation of this paper."""
284,"""Remember and Forget for Experience Replay""","['reinforcement learning', 'experience replay', 'policy gradients']","""Experience replay (ER) is crucial for attaining high data-efficiency in off-policy deep reinforcement learning (RL). ER entails the recall of experiences obtained in past iterations to compute gradient estimates for the current policy. However, the accuracy of such updates may deteriorate when the policy diverges from past behaviors, possibly undermining the effectiveness of ER. Previous off-policy RL algorithms mitigated this issue by tuning their hyper-parameters in order to abate policy changes. We propose ReF-ER, a method for active management of experiences in the Replay Memory (RM). ReF-ER forgets experiences that would be too unlikely with the current policy and constrains policy changes within a trust region of the behaviors in the RM. We couple ReF-ER with Q-learning, deterministic policy gradient and off-policy gradient methods to show that ReF-ER reliably improves the performance of continuous-action off-policy RL. We complement ReF-ER with a novel off-policy actor-critic algorithm (RACER) for continuous-action control. RACER employs a computationally efficient closed-form approximation of the action values and is shown to be highly competitive with state-of-the-art algorithms on benchmark problems, while being robust to large hyper-parameter variations.""","""This paper introduces a novel idea, and demonstrates its utility in several simulated domains. The key parts of the algorithm are (a) to prefer keeping and using samples in the ER buffer where the corresponding rho_t, using the current policy pi_t, are not too big or small and (b) preventing the policy from changing too quickly, so that samples in the ER buffer are more on-policy. They key weakness is not better investigating the idea of making the ER buffer more on-policy, and the effect of doing so. The experiments compare to other algorithms, but do not sufficiently investigate the use of both Point 1 and Point 3. Further, the appendix contains an investigation into parameter sensitivity and gives some confidence intervals. However, the presentation of this is difficult to follow, and so it is difficult to gauge the sensitivity of Ref-ER. With a more thorough experimental section, better demonstrating the results (not necessarily running more things), the paper would be much stronger. For more context, the authors rightly mention ""It is commonly believed that off-policy methods (e.g. Q-learning) can handle the dissimilarity between off-policy and on-policy outcomes. We provide ample evidence that training from highly similar-policy experiences is essential to the success of off-policy continuous-action deep RL."" Q-learning can significantly suffer from changing the state-sampling distribution. However, adjusting sampling in the ER buffer using rho_t does not change the state-sampling distribution, and so that mismatch remains a problem. Changing the policy more slowly (Point 3) could help with this more. In general, however, these play two different roles that need to be better understood. The introduction more strongly focuses on classifying samples as more on or off-policy, to solve this problem, rather than the strategy used in Point 3. So, from the current pitch, its not clear which component is solving the issues claimed with off-policy updates. Overall, this paper has some interesting results and is well-written. With more clarity on the roles of the two components of Ref-ER and what they mean for making the ER buffer more on-policy, in terms of both action selection and state distribution, this paper would be a very useful contribution to stable control. """
285,"""An Automatic Operation Batching Strategy for the Backward Propagation of Neural Networks Having Dynamic Computation Graphs""","['Automatic Operation Batching', 'Dynamic Computation Graphs']","""Organizing the same operations in the computation graph of a neural network into batches is one of the important methods to improve the speed of training deep learning models and applications since it helps to execute operations with the same type in parallel and to make full use of the available hardware resources. This batching task is usually done by the developers manually and it becomes more dif- ficult when the neural networks have dynamic computation graphs because of the input data with varying structures or the dynamic flow control. Several automatic batching strategies were proposed and integrated into some deep learning toolkits so that the programmers dont have to be responsible for this task. These strategies, however, will miss some important opportunities to group the operations in the backward propagation of training neural networks. In this paper, we proposed a strategy which provides more efficient automatic batching and brings benefits to the memory access in the backward propagation. We also test our strategy on a variety of benchmarks with dynamic computation graphs. The result shows that it really brings further improvements in the training speed when our strategy is working with the existing automatic strategies.""","""This paper describes a new batching strategy for more efficient training of deep neural nets. The idea stems from the observation that some operations can only be batched more efficiently in the backward, suggesting that batching should be different between forward and backward. The results show that the proposed method improves upon existing batch strategies across three tasks. The reviewers find the work novel, but note that it does not properly address the trade-offs made by the technique - such as memory consumption. They also argue that the writing should be improved before acceptance at ICLR."""
286,"""Latent Domain Transfer: Crossing modalities with Bridging Autoencoders""","['Generative Model', 'Latent Space', 'Domain Transfer']","""Domain transfer is a exciting and challenging branch of machine learning because models must learn to smoothly transfer between domains, preserving local variations and capturing many aspects of variation without labels. However, most successful applications to date require the two domains to be closely related (ex. image-to-image, video-video), utilizing similar or shared networks to transform domain specific properties like texture, coloring, and line shapes. Here, we demonstrate that it is possible to transfer across modalities (ex. image-to-audio) by first abstracting the data with latent generative models and then learning transformations between latent spaces. We find that a simple variational autoencoder is able to learn a shared latent space to bridge between two generative models in an unsupervised fashion, and even between different types of models (ex. variational autoencoder and a generative adversarial network). We can further impose desired semantic alignment of attributes with a linear classifier in the shared latent space. The proposed variation autoencoder enables preserving both locality and semantic alignment through the transfer process, as shown in the qualitative and quantitative evaluations. Finally, the hierarchical structure decouples the cost of training the base generative models and semantic alignments, enabling computationally efficient and data efficient retraining of personalized mapping functions. ""","""This paper studies the problem of heterogeneous domain transfer, for example across different data modalities. The comments of the reviewers are overlapping to a great extent. On the one hand, the reviewers and AC agree that the problem considered is very interesting and deserves more attention. On the other hand, the reviewers have raised concerns about the amount of novelty contained in this manuscript, as well as convincingness of results. The AC understands the authors argument that a simple method can be a feature and not a flaw, however this work still does not feel complete. Even within a relatively simple framework, it would be desirable to examine the problem from multiple angles and ""disentangle"" the effects of the different hypotheses for example the reviewers have drawn attention to end-to-end training and comparison with other baselines. The points raised above, together with improving the manuscript (as commented by reviewers) would make this work more complete."""
287,"""The Nonlinearity Coefficient - Predicting Generalization in Deep Neural Networks""","['deep learning', 'neural networks', 'nonlinearity', 'activation functions', 'exploding gradients', 'vanishing gradients', 'neural architecture search']","""For a long time, designing neural architectures that exhibit high performance was considered a dark art that required expert hand-tuning. One of the few well-known guidelines for architecture design is the avoidance of exploding or vanishing gradients. However, even this guideline has remained relatively vague and circumstantial, because there exists no well-defined, gradient-based metric that can be computed {\it before} training begins and can robustly predict the performance of the network {\it after} training is complete. We introduce what is, to the best of our knowledge, the first such metric: the nonlinearity coefficient (NLC). Via an extensive empirical study, we show that the NLC, computed in the network's randomly initialized state, is a powerful predictor of test error and that attaining a right-sized NLC is essential for attaining an optimal test error, at least in fully-connected feedforward networks. The NLC is also conceptually simple, cheap to compute, and is robust to a range of confounders and architectural design choices that comparable metrics are not necessarily robust to. Hence, we argue the NLC is an important tool for architecture search and design, as it can robustly predict poor training outcomes before training even begins.""","""This paper proposes the NonLinearity Coefficient (NLC), a metric which aims to predicts test-time performance of neural networks at initialization. The idea is interesting and novel, and has clear practical implications. Reviewers unanimously agreed that the direction is a worthwhile one to pursue. However, several reviewers also raised concerns about how well-justified the method is: in particular, Reviewer 3 believes that a quantitative comparison to the related work is necessary, and takes issue with the motivation for being ad-hoc. Reviewer 2 also is concerned about the soundness of the coefficient in truly measuring nonlinearity. These concerns make it clear that the paper needs more work before it can be published. And, in particular, addressing the reviewers' concerns and providing proper comparison to related works will go a long way in that direction."""
288,"""Learning to Control Visual Abstractions for Structured Exploration in Deep Reinforcement Learning""","['exploration', 'deep reinforcement learning', 'intrinsic motivation', 'unsupervised learning']","""Exploration in environments with sparse rewards is a key challenge for reinforcement learning. How do we design agents with generic inductive biases so that they can explore in a consistent manner instead of just using local exploration schemes like epsilon-greedy? We propose an unsupervised reinforcement learning agent which learns a discrete pixel grouping model that preserves spatial geometry of the sensors and implicitly of the environment as well. We use this representation to derive geometric intrinsic reward functions, like centroid coordinates and area, and learn policies to control each one of them with off-policy learning. These policies form a basis set of behaviors (options) which allows us explore in a consistent way and use them in a hierarchical reinforcement learning setup to solve for extrinsically defined rewards. We show that our approach can scale to a variety of domains with competitive performance, including navigation in 3D environments and Atari games with sparse rewards.""","""The paper presents an unsupervised visual abstraction model, used for reinforcement learning tasks. It is trained through intrinsic rewards, generated from temporal differences of inputs. This is similar to ""learning to control pixels"". The method is tested in DM Lab (3D environment, 2D navigation tasks) and Atari (Montezuma's Revenge). The paper is at times hard to follow, and it seems the improvements accompanying the rebuttals did not convince reviewers to change their notes significantly. The experiments do not contain enough comparisons to other models, baselines, nor ablations, to sustain the claims. In its current form, this is not acceptable for publication at ICLR."""
289,"""Learning Multimodal Graph-to-Graph Translation for Molecule Optimization""","['graph-to-graph translation', 'graph generation', 'molecular optimization']","""We view molecule optimization as a graph-to-graph translation problem. The goal is to learn to map from one molecular graph to another with better properties based on an available corpus of paired molecules. Since molecules can be optimized in different ways, there are multiple viable translations for each input graph. A key challenge is therefore to model diverse translation outputs. Our primary contributions include a junction tree encoder-decoder for learning diverse graph translations along with a novel adversarial training method for aligning distributions of molecules. Diverse output distributions in our model are explicitly realized by low-dimensional latent vectors that modulate the translation process. We evaluate our model on multiple molecule optimization tasks and show that our model outperforms previous state-of-the-art baselines by a significant margin. ""","""The revisions made by the authors convinced the reviewers to all recommend accepting this paper. Therefore, I am recommending acceptance as well. I believe the revisions were important to make since I concur with several points in the initial reviews about additional baselines. It is all too easy to add confusion to the literature by not including enough experiments. """
290,"""GAN Dissection: Visualizing and Understanding Generative Adversarial Networks""","['GANs', 'representation', 'interpretability', 'causality']","""Generative Adversarial Networks (GANs) have recently achieved impressive results for many real-world applications, and many GAN variants have emerged with improvements in sample quality and training stability. However, visualization and understanding of GANs is largely missing. How does a GAN represent our visual world internally? What causes the artifacts in GAN results? How do architectural choices affect GAN learning? Answering such questions could enable us to develop new insights and better models. In this work, we present an analytic framework to visualize and understand GANs at the unit-, object-, and scene-level. We first identify a group of interpretable units that are closely related to object concepts with a segmentation-based network dissection method. Then, we quantify the causal effect of interpretable units by measuring the ability of interventions to control objects in the output. Finally, we examine the contextual relationship between these units and their surrounding by inserting the discovered object concepts into new images. We show several practical applications enabled by our framework, from comparing internal representations across different layers, models, and datasets, to improving GANs by locating and removing artifact-causing units, to interactively manipulating objects in the scene. We provide open source interpretation tools to help peer researchers and practitioners better understand their GAN models.""","""The paper proposes an interesting framework for visualizing and understanding GANs, that will be of clear help for understanding existing models and might provide insights for developing new ones. """
291,"""Multi-Domain Adversarial Learning""","['multi-domain learning', 'domain adaptation', 'adversarial learning', 'H-divergence', 'deep representation learning', 'high-content microscopy']","""Multi-domain learning (MDL) aims at obtaining a model with minimal average risk across multiple domains. Our empirical motivation is automated microscopy data, where cultured cells are imaged after being exposed to known and unknown chemical perturbations, and each dataset displays significant experimental bias. This paper presents a multi-domain adversarial learning approach, MuLANN, to leverage multiple datasets with overlapping but distinct class sets, in a semi-supervised setting. Our contributions include: i) a bound on the average- and worst-domain risk in MDL, obtained using the H-divergence; ii) a new loss to accommodate semi-supervised multi-domain learning and domain adaptation; iii) the experimental validation of the approach, improving on the state of the art on two standard image benchmarks, and a novel bioimage dataset, Cell.""","""This paper extends the single source H-divergence theory for domain adaptation to the case of multiple domains. Thus, drawing on the known connection between H-divergence and learning the domain classifier for adversarial adaptation, the authors propose a multi-domain adversarial learning algorithm. The approach builds upon the gradient reversal version of adversarial adaptation proposed by Ganin et al 2016. Overall, multi-domain learning and limiting the worst case performance on any single domain is an interesting problem which has been relatively underexplored. Though this work does not have the highest performance on all datasets across competing methods, as noted by reviewers, it proposes a useful theoretical result which future research may build on. I would encourage the reviewers to compare against and discuss the missing prior work cited by Rev 3. """
292,"""ProMP: Proximal Meta-Policy Search""","['Meta-Reinforcement Learning', 'Meta-Learning', 'Reinforcement-Learning']","""Credit assignment in Meta-reinforcement learning (Meta-RL) is still poorly understood. Existing methods either neglect credit assignment to pre-adaptation behavior or implement it naively. This leads to poor sample-efficiency during meta-training as well as ineffective task identification strategies. This paper provides a theoretical analysis of credit assignment in gradient-based Meta-RL. Building on the gained insights we develop a novel meta-learning algorithm that overcomes both the issue of poor credit assignment and previous difficulties in estimating meta-policy gradients. By controlling the statistical distance of both pre-adaptation and adapted policies during meta-policy search, the proposed algorithm endows efficient and stable meta-learning. Our approach leads to superior pre-adaptation policy behavior and consistently outperforms previous Meta-RL algorithms in sample-efficiency, wall-clock time, and asymptotic performance.""","""The paper studies the credit assignment problem in meta-RL, proposes a new algorithm that computes the right gradient, and demonstrates its superior empirical performance over others. The paper is well written, and all reviewers agree the work is a solid contribution to an important problem."""
293,"""Representing Formal Languages: A Comparison Between Finite Automata and Recurrent Neural Networks ""","['Language recognition', 'Recurrent Neural Networks', 'Representation Learning', 'deterministic finite automaton', 'automaton']","""We investigate the internal representations that a recurrent neural network (RNN) uses while learning to recognize a regular formal language. Specifically, we train a RNN on positive and negative examples from a regular language, and ask if there is a simple decoding function that maps states of this RNN to states of the minimal deterministic finite automaton (MDFA) for the language. Our experiments show that such a decoding function indeed exists, and that it maps states of the RNN not to MDFA states, but to states of an {\em abstraction} obtained by clustering small sets of MDFA states into ``''superstates''. A qualitative analysis reveals that the abstraction often has a simple interpretation. Overall, the results suggest a strong structural relationship between internal representations used by RNNs and finite automata, and explain the well-known ability of RNNs to recognize formal grammatical structure. ""","""This paper presents experiments showing that a linear mapping existing between the hidden states of RNNs trained to recognise (rather than model) formal languages, in the hope of at least partially elucidating the sort of representations this class of network architectures learns. This is important and timely work, fitting into a research programme begun by CL Giles in 92. Despite its relatively low overall score, I am concurring with the assessment made by reviewer 1, whose expertise in the topic I am aware of and respect. But more importantly, I feel the review process has failed the authors here: reviewers 2 and 3 had as chief concern that there were issues with the clarity of some aspects of the paper. The authors made a substantial and bona fide attempt in their response to address the points of concern raised by these reviewers. This is precisely what the discussion period of ICLR is for, and one would expect that clarity issues can be successfully remedied during this period. I am disappointed to have seen little timely engagement from these reviewers, or willingness to explain why they are stick by their assessment if not revisiting it. As far as I am concerned, the authors have done an appropriate job of addressing these concerns, and given reviewer 1's support for the paper, I am happy to add mine as well."""
294,"""ATTACK GRAPH CONVOLUTIONAL NETWORKS BY ADDING FAKE NODES""","['Graph Convolutional Network', 'adversarial attack', 'node classification']","""Graph convolutional networks (GCNs) have been widely used for classifying graph nodes in the semi-supervised setting. Previous works have shown that GCNs are vulnerable to the perturbation on adjacency and feature matrices of existing nodes. However, it is unrealistic to change the connections of existing nodes in many applications, such as existing users in social networks. In this paper, we investigate methods attacking GCNs by adding fake nodes. A greedy algorithm is proposed to generate adjacency and feature matrices of fake nodes, aiming to minimize the classification accuracy on the existing ones. In additional, we introduce a discriminator to classify fake nodes from real nodes, and propose a Greedy-GAN algorithm to simultaneously update the discriminator and the attacker, to make fake nodes indistinguishable to the real ones. Our non-targeted attack decreases the accuracy of GCN down to 0.10, and our targeted attack reaches a success rate of 0.99 for attacking the whole datasets, and 0.94 on average for attacking a single node.""","""While the main idea of the paper is nice, the reviewers are not satisfied with the clarity of the material and the execution."""
295,"""Transfer Value or Policy? A Value-centric Framework Towards Transferrable Continuous Reinforcement Learning""","['Reinforcement Learning', 'Transfer Learning', 'Control', 'Value function']","""Transferring learned knowledge from one environment to another is an important step towards practical reinforcement learning (RL). In this paper, we investigate the problem of transfer learning across environments with different dynamics while accomplishing the same task in the continuous control domain. We start by illustrating the limitations of policy-centric methods (policy gradient, actor- critic, etc.) when transferring knowledge across environments. We then propose a general model-based value-centric (MVC) framework for continuous RL. MVC learns a dynamics approximator and a value approximator simultaneously in the source domain, and makes decision based on both of them. We evaluate MVC against popular baselines on 5 benchmark control tasks in a training from scratch setting and a transfer learning setting. Our experiments demonstrate MVC achieves comparable performance with the baselines when it is trained from scratch, while it significantly surpasses them when it is used in the transfer setting. ""","""The paper studies whether the best strategy for transfer learning in RL is to transfer value estimates or policy probabilities. The paper also presents a model-based value-centric (MVC) framework for continuous RL. The reviewers raised concerns regarding (1) the coherence of the story, (2) the novelty and importance of the MVC framework and (3) the significance of the experiments. I encourage the authors to either focus on the algorithmic aspect or the transfer learning aspect and expand on the experimental results to make them more convincing. I appreciate the changes made to improve the paper, but in its current form the paper is still below the acceptance threshold at ICLR. PS: in my view one can think of value as (shifted and scaled) log of policy. Hence, it is a bit ambiguous to ask whether to transfer value or policy."""
296,"""A Kernel Random Matrix-Based Approach for Sparse PCA""","['Random Matrix Theory', 'Concentration of Measure', 'Sparse PCA', 'Covariance Thresholding']","""In this paper, we present a random matrix approach to recover sparse principal components from n p-dimensional vectors. Specifically, considering the large dimensional setting where n, p with p/n c (0, ) and under Gaussian vector observations, we study kernel random matrices of the type f (), where f is a three-times continuously differentiable function applied entry-wise to the sample covariance matrix of the data. Then, assuming that the principal components are sparse, we show that taking f in such a way that f'(0) = f''(0) = 0 allows for powerful recovery of the principal components, thereby generalizing previous ideas involving more specific f functions such as the soft-thresholding function.""","""The manuscript studies a random matrix approach to recover sparse principal components. This work extends prior work using soft thresholding of the sample covariance matrix to enable sparse PCA. In this light, the main contribution of the paper is a study of generalizing soft thresholding to a broader class of functions and showing that this improves performance. The contributions of this paper are primarily theoretical. The reviewers and AC note issues with the discussion that can be further improved to better illustrate contributions, and place this work in context. In particular, multiple reviewers assumed that ""kernel"" referred to the covariance matrix. The authors provide a satisfactory rebuttal addressing these issues. While not unanimous, overall the reviewers and AC have a positive opinion of this paper and recommend acceptance."""
297,"""W2GAN: RECOVERING AN OPTIMAL TRANSPORT MAP WITH A GAN""","['Optimal Transportation', 'Deep Learning', 'Generative Adversarial Networks', 'Wasserstein Distance']","""Understanding and improving Generative Adversarial Networks (GAN) using notions from Optimal Transport (OT) theory has been a successful area of study, originally established by the introduction of the Wasserstein GAN (WGAN). An increasing number of GANs incorporate OT for improving their discriminators, but that is so far the sole way for the two domains to cross-fertilize. In this work we address the converse question: is it possible to recover an optimal map in a GAN fashion? To achieve this, we build a new model relying on the second Wasserstein distance. This choice enables the use of many results from OT community. In particular, we may completely describe the dynamics of the generator during training. In addition, experiments show that practical uses of our model abide by the rule of evolution we describe. As an application, our generator may be considered as a new way of computing an optimal transport map. It is competitive in low-dimension with standard and deterministic ways to approach the same problem. In high dimension, the fact it is a GAN-style method makes it more powerful than other methods.""","""The paper introduces a W2GAN method for training GAN by minimizing 2-Wasserstein distance using by computing an optimal transport (OT) map between distributions. However, the difference of previous works is not significant or clearly clarified as pointed out some of the reviewers. The advantage of W2GAN over standard WGAN is also superficially explained, and did not supported by strong empirical evidence. """
298,"""Enabling Factorized Piano Music Modeling and Generation with the MAESTRO Dataset""","['music', 'piano transcription', 'transformer', 'wavnet', 'audio synthesis', 'dataset', 'midi']","""Generating musical audio directly with neural networks is notoriously difficult because it requires coherently modeling structure at many different timescales. Fortunately, most music is also highly structured and can be represented as discrete note events played on musical instruments. Herein, we show that by using notes as an intermediate representation, we can train a suite of models capable of transcribing, composing, and synthesizing audio waveforms with coherent musical structure on timescales spanning six orders of magnitude (~0.1 ms to ~100 s), a process we call Wave2Midi2Wave. This large advance in the state of the art is enabled by our release of the new MAESTRO (MIDI and Audio Edited for Synchronous TRacks and Organization) dataset, composed of over 172 hours of virtuosic piano performances captured with fine alignment (~3 ms) between note labels and audio waveforms. The networks and the dataset together present a promising approach toward creating new expressive and interpretable neural models of music.""","""All reviewers agree that the presented audio data augmentation is very interesting, well presented, and clearly advancing the state of the art in the field. The authors rebuttal clarified the remaining questions by the reviewers. All reviewers recommend strong acceptance (oral presentation) at ICLR. I would like to recommend this paper for oral presentation due to a number of reasons including the importance of the problem addressed (data augmentation is the only way forward in cases where we do not have enough of training data), the novelty and innovativeness of the model, and the clarity of the paper. The work will be of interest to the widest audience beyond ICLR."""
299,"""Neural Probabilistic Motor Primitives for Humanoid Control""","['Motor Primitives', 'Distillation', 'Reinforcement Learning', 'Continuous Control', 'Humanoid Control', 'Motion Capture', 'One-Shot Imitation']","""We focus on the problem of learning a single motor module that can flexibly express a range of behaviors for the control of high-dimensional physically simulated humanoids. To do this, we propose a motor architecture that has the general structure of an inverse model with a latent-variable bottleneck. We show that it is possible to train this model entirely offline to compress thousands of expert policies and learn a motor primitive embedding space. The trained neural probabilistic motor primitive system can perform one-shot imitation of whole-body humanoid behaviors, robustly mimicking unseen trajectories. Additionally, we demonstrate that it is also straightforward to train controllers to reuse the learned motor primitive space to solve tasks, and the resulting movements are relatively naturalistic. To support the training of our model, we compare two approaches for offline policy cloning, including an experience efficient method which we call linear feedback policy cloning. We encourage readers to view a supplementary video (pseudo-url ) summarizing our results.""","""Strengths: One-shot physics-based imitation at a scale and with efficiency not seen before. Clear video, paper, and related work. Weaknesses described include: the description of a secondary contribution (LFPC) takes up too much space (R1,4); results are not compelling (R1,4); prior art in graphics and robotics (R2,6); concerns about the potential limitations of the linearization used by LFPC. The original reviews are negative overall (6,3,4). The authors have posted detailed replies. R1 has posted a followup, standing by their score. We have not heard more from R2 and R3. The AC has read the paper, watched the video, and read all the reviews. Based on expertise in this area, the AC endorses the author's responses to R1 and R2. Being able to compare LFPC to more standard behavior cloning is a valuable data point for the community; there is value in testing simple and efficient models first. The AC identifies the following recent (Nov 2018) paper as being the closest work, which is not identified by the authors or the reviewers. The approach being proposed in the submitted paper demonstrates equal-or-better scalability, learning efficiency, and motion quality, and includes examples of learned high-level behaviors. An elaboration on HL/LL control: the DeepLoco work also learns mocap-based LL-control with learned HL behaviors. although with a more dedicated structure. Physics-based motion capture imitation with deep reinforcement learning pseudo-url Overall, the AC recommends this paper to be accepted as a paper of interest to ICLR. This does partially discount R3 and R1, who may not have worked as directly on these specific problems before. The AC requests is rating the confidence as ""not sure"" to flag this for the program committee chairs, in light of the fact that this discounts the R1 and R3 reviews. The AC is quite certain in terms of the technical contributions of the paper. """
300,"""Convolutional Neural Networks combined with Runge-Kutta Methods""",[],"""A convolutional neural network for image classification can be constructed mathematically since it can be regarded as a multi-period dynamical system. In this paper, a novel approach is proposed to construct network models from the dynamical systems view. Since a pre-activation residual network can be deemed an approximation of a time-dependent dynamical system using the forward Euler method, higher order Runge-Kutta methods (RK methods) can be utilized to build network models in order to achieve higher accuracy. The model constructed in such a way is referred to as the Runge-Kutta Convolutional Neural Network (RKNet). RK methods also provide an interpretation of Dense Convolutional Networks (DenseNets) and Convolutional Neural Networks with Alternately Updated Clique (CliqueNets) from the dynamical systems view. The proposed methods are evaluated on benchmark datasets: CIFAR-10/100, SVHN and ImageNet. The experimental results are consistent with the theoretical properties of RK methods and support the dynamical systems interpretation. Moreover, the experimental results show that the RKNets are superior to the state-of-the-art network models on CIFAR-10 and on par on CIFAR-100, SVHN and ImageNet.""","""The paper proposes a novel approach to neural net construction using dynamical systems approach, such as higher order Runge-Kutta method; this approach also allows a dynamical systems interpretation of DenseNets and CliqueNets. While all reviewers agree that this is an intersting a novel approach, along the lines of recent developments in the field on dynamical systems approaches to deep nets, they also suggest to further improve the writing/clarity of the paper and also strengthen the empirical results (currently, the method only provided advantage on CIFAR-10, while being somewhat suboptimal on other datasets, and more evidence for empirical advantages of the proposed approach would be great). Overall, this is a very interesting and promising work, and with a few more empirical demonstrations of the method's superiority as well as more polished wiriting the paper would make a nice contribution to ML community."""
301,"""Time-Agnostic Prediction: Predicting Predictable Video Frames""","['visual prediction', 'subgoal generation', 'bottleneck states', 'time-agnostic']","""Prediction is arguably one of the most basic functions of an intelligent system. In general, the problem of predicting events in the future or between two waypoints is exceedingly difficult. However, most phenomena naturally pass through relatively predictable bottlenecks---while we cannot predict the precise trajectory of a robot arm between being at rest and holding an object up, we can be certain that it must have picked the object up. To exploit this, we decouple visual prediction from a rigid notion of time. While conventional approaches predict frames at regularly spaced temporal intervals, our time-agnostic predictors (TAP) are not tied to specific times so that they may instead discover predictable ""bottleneck"" frames no matter when they occur. We evaluate our approach for future and intermediate frame prediction across three robotic manipulation tasks. Our predictions are not only of higher visual quality, but also correspond to coherent semantic subgoals in temporally extended tasks.""","""The paper introduces a new and convincing method for video frame prediction, by adding prediction uncertainty through VAEs. The results are convincing, and the reviewers are convinced. It's unfortunate however that the method is only evaluated on simulated data. Letting it loose on real data would cement the results and merit oral representation; in the current form, poster presentation is recommended."""
302,"""Learning agents with prioritization and parameter noise in continuous state and action space""","['reinforcement learning', 'continuous action space', 'prioritization', 'parameter', 'noise', 'policy gradients']","""Reinforcement Learning (RL) problem can be solved in two different ways - the Value function-based approach and the policy optimization-based approach - to eventually arrive at an optimal policy for the given environment. One of the recent breakthroughs in reinforcement learning is the use of deep neural networks as function approximators to approximate the value function or q-function in a reinforcement learning scheme. This has led to results with agents automatically learning how to play games like alpha-go showing better-than-human performance. Deep Q-learning networks (DQN) and Deep Deterministic Policy Gradient (DDPG) are two such methods that have shown state-of-the-art results in recent times. Among the many variants of RL, an important class of problems is where the state and action spaces are continuous --- autonomous robots, autonomous vehicles, optimal control are all examples of such problems that can lend themselves naturally to reinforcement based algorithms, and have continuous state and action spaces. In this paper, we adapt and combine approaches such as DQN and DDPG in novel ways to outperform the earlier results for continuous state and action space problems. We believe these results are a valuable addition to the fast-growing body of results on Reinforcement Learning, more so for continuous state and action space problems.""","""The authors take two algorithmic components that were proposed in the context of discrete-action RL - priority replay and parameter noise - and evaluate them with DDPG on continuous control tasks. The different approaches are nicely summarized by the authors, however the contribution of the paper is extremely limited. There is no novelty in the proposed approaches, the empirical evaluation is inconclusive and limited, and there is no analysis or additional insights or results. The AC and the reviewers agree that this paper is not strong enough for ICLR."""
303,"""Robust Conditional Generative Adversarial Networks""","['conditional GAN', 'unsupervised pathway', 'autoencoder', 'robustness']","""Conditional generative adversarial networks (cGAN) have led to large improvements in the task of conditional image generation, which lies at the heart of computer vision. The major focus so far has been on performance improvement, while there has been little effort in making cGAN more robust to noise. The regression (of the generator) might lead to arbitrarily large errors in the output, which makes cGAN unreliable for real-world applications. In this work, we introduce a novel conditional GAN model, called RoCGAN, which leverages structure in the target space of the model to address the issue. Our model augments the generator with an unsupervised pathway, which promotes the outputs of the generator to span the target manifold even in the presence of intense noise. We prove that RoCGAN share similar theoretical properties as GAN and experimentally verify that our model outperforms existing state-of-the-art cGAN architectures by a large margin in a variety of domains including images from natural scenes and faces.""","""The proposed method suggests a way to do robust conditional image generation with GANs. The premise is to make the image to image translation model resilient to noise by leveraging structure in the output space, with an unsupervised ""pathway"". In general, the qualitative results seem reasonable on a a number of datasets, including those suggested by reviewers. The method appears simple, novel and easy to try. The main concerns seem to be that the idea is maybe too simple, but I'm not particularly bothered by that. The authors showed it working well on a variety of tasks (synthetic and natural), provide SSIM numbers that look compelling (despite SSIM's short-comings) and otherwise give compelling arguments for the technical soundness of the approach. Thus, I recommend acceptance."""
304,"""Two-Timescale Networks for Nonlinear Value Function Approximation""","['Reinforcement learning', 'policy evaluation', 'nonlinear function approximation']","""A key component for many reinforcement learning agents is to learn a value function, either for policy evaluation or control. Many of the algorithms for learning values, however, are designed for linear function approximation---with a fixed basis or fixed representation. Though there have been a few sound extensions to nonlinear function approximation, such as nonlinear gradient temporal difference learning, these methods have largely not been adopted, eschewed in favour of simpler but not sound methods like temporal difference learning and Q-learning. In this work, we provide a two-timescale network (TTN) architecture that enables linear methods to be used to learn values, with a nonlinear representation learned at a slower timescale. The approach facilitates the use of algorithms developed for the linear setting, such as data-efficient least-squares methods, eligibility traces and the myriad of recently developed linear policy evaluation algorithms, to provide nonlinear value estimates. We prove convergence for TTNs, with particular care given to ensure convergence of the fast linear component under potentially dependent features provided by the learned representation. We empirically demonstrate the benefits of TTNs, compared to other nonlinear value function approximation algorithms, both for policy evaluation and control. ""","""The paper proposes a new method to approximate the nonlinear value function by estimating it as a sum of linear and nonlinear terms. The nonlinear term is updated much slower than the linear term, and the paper proposes to use a fast least-square algorithm to update the linear term. Convergence results are also discussed and empirical evidence is provided. As reviewers have pointed out, the novelty of the paper is limited, but the ideas are interesting and could be useful for the community. I strongly recommend taking reviewers comments into account for the camera ready and also add a discussion on the relationship with the existing work. Overall, I think this paper is interesting and I recommend acceptance. """
305,"""Automatic generation of object shapes with desired functionalities""","['automated design', 'affordance learning']","""3D objects (artefacts) are made to fulfill functions. Designing an object often starts with defining a list of functionalities that it should provide, also known as functional requirements. Today, the design of 3D object models is still a slow and largely artisanal activity, with few Computer-Aided Design (CAD) tools existing to aid the exploration of the design solution space. The purpose of the study is to explore the possibility of shape generation conditioned on desired functionalities. To accelerate the design process, we introduce an algorithm for generating object shapes with desired functionalities. We follow the principle form follows function, and assume that the form of a structure is correlated to its function. First, we use an artificial neural network to learn a function-to-form mapping by analysing a dataset of objects labeled with their functionalities. Then, we combine forms providing one or more desired functions, generating an object shape that is expected to provide all of them. Finally, we verify in simulation whether the generated object possesses the desired functionalities, by defining and executing functionality tests on it.""","""The paper presents a novel problem formulation, that of generating 3D object shapes based on their functionality. They use a dataset of 3d shapes annotated with functionalities to learn a voxel generative network that conditions on the desired functionality to generate a voxel occupancy grid. However, the fact that the results are not very convincing -resulting 3D shapes are very coarse- raises questions regarding the usefulness of the proposed problem formulation. Thus, the problem formulation novelty alone is not enough for acceptance. Combined with a motivating application to demonstrate the usefulness of the problem formulation and results, would make this paper a much stronger submission. Furthermore, the authors have greatly improved the writing of the manuscript during the discussion phase."""
306,"""Curiosity-Driven Experience Prioritization via Density Estimation""","['Curiosity-Driven', 'Experience Prioritization', 'Hindsight Experience', 'Reinforcement Learning']","""In Reinforcement Learning (RL), an agent explores the environment and collects trajectories into the memory buffer for later learning. However, the collected trajectories can easily be imbalanced with respect to the achieved goal states. The problem of learning from imbalanced data is a well-known problem in supervised learning, but has not yet been thoroughly researched in RL. To address this problem, we propose a novel Curiosity-Driven Prioritization (CDP) framework to encourage the agent to over-sample those trajectories that have rare achieved goal states. The CDP framework mimics the human learning process and focuses more on relatively uncommon events. We evaluate our methods using the robotic environment provided by OpenAI Gym. The environment contains six robot manipulation tasks. In our experiments, we combined CDP with Deep Deterministic Policy Gradient (DDPG) with or without Hindsight Experience Replay (HER). The experimental results show that CDP improves both performance and sample-efficiency of reinforcement learning agents, compared to state-of-the-art methods.""","""The manuscript describes a procedure for prioritizing the contents of an experience replay buffer in a UVFA setting based on a density model of the trajectory of the achieved goal states. A rank-based transformation of densities is used to stochastically prioritize the replay memory. Reducing the sample complexity of RL is a worthy goal and reviewers found the overall approach is interesting, if somewhat arbitrary in the implementation details. Concerns were raised about clarity and justification, and the restriction of experiments to fully deterministic environments. After personally reading the updated manuscript I found clarity to still be lacking. Statements like ""... uses the ranking number (starting from zero) directly as the probability for sampling"" -- this is not true (it is normalized, as confusingly laid out in equation 2 with the same symbol used for the unnormalized and normalized densities), and also implies that the least likely trajectory under the model is never sampled, which doesn't seem like a desirable property. Schaul's ""prioritized experience replay"" is cited for the choice of rank-based distribution, but the distribution employed in that work has rather different form. The related work section is also very poor given the existing body of literature on curiosity in a reinforcement learning context, and the new ""importance sampling perspective"" section serves little explicatory purpose given that an importance re-weighting is not performed. Overall, I concur most strongly with AnonReviewer1 that more work is needed to motivate the method and prove its robustness applicability, as well as to polish the presentation."""
307,"""DEEP HIERARCHICAL MODEL FOR HIERARCHICAL SELECTIVE CLASSIFICATION AND ZERO SHOT LEARNING""","['deep learning', 'large-scale classificaion', 'heirarchical classification', 'zero-shot learning']","""Object recognition in real-world image scenes is still an open problem. With the growing number of classes, the similarity structures between them become complex and the distinction between classes blurs, which makes the classification problem particularly challenging. Standard N-way discrete classifiers treat all classes as disconnected and unrelated, and therefore unable to learn from their semantic relationships. In this work, we present a hierarchical inter-class relationship model and train it using a newly proposed probability-based loss function. Our hierarchical model provides significantly better semantic generalization ability compared to a regular N-way classifier. We further proposed an algorithm where given a probabilistic classification model it can return the input corresponding super-group based on classes hierarchy without any further learning. We deploy it in two scenarios in which super-group retrieval can be useful. The first one, selective classification, deals with the problem of low-confidence classification, wherein a model is unable to make a successful exact classification. The second, zero-shot learning problem deals with making reasonable inferences on novel classes. Extensive experiments with the two scenarios show that our proposed hierarchical model yields more accurate and meaningful super-class predictions compared to a regular N-way classifier because of its significantly better semantic generalization ability.""","""The paper proposes to take into accunt the label structure for classification tasks, instead of a flat N-way softmax. This also lead to a zero-shot setting to consider novel classes. Reviewers point to a lack of reference to prior work and comparisons. Authors have tried to justify their choices, but the overall sentiment is that it lacks novelty with respect to previous approaches. All reviewers recommend to reject, and so do I."""
308,"""An analytic theory of generalization dynamics and transfer learning in deep linear networks""","['Generalization', 'Theory', 'Transfer', 'Multi-task', 'Linear']","""Much attention has been devoted recently to the generalization puzzle in deep learning: large, deep networks can generalize well, but existing theories bounding generalization error are exceedingly loose, and thus cannot explain this striking performance. Furthermore, a major hope is that knowledge may transfer across tasks, so that multi-task learning can improve generalization on individual tasks. However we lack analytic theories that can quantitatively predict how the degree of knowledge transfer depends on the relationship between the tasks. We develop an analytic theory of the nonlinear dynamics of generalization in deep linear networks, both within and across tasks. In particular, our theory provides analytic solutions to the training and testing error of deep networks as a function of training time, number of examples, network size and initialization, and the task structure and SNR. Our theory reveals that deep networks progressively learn the most important task structure first, so that generalization error at the early stopping time primarily depends on task structure and is independent of network size. This suggests any tight bound on generalization error must take into account task structure, and explains observations about real data being learned faster than random data. Intriguingly our theory also reveals the existence of a learning algorithm that proveably out-performs neural network training through gradient descent. Finally, for transfer learning, our theory reveals that knowledge transfer depends sensitively, but computably, on the SNRs and input feature alignments of pairs of tasks.""","""The authors provide a new analysis of generalization in deep linear networks, provide new insight through the role of ""task structure"". Empirical findings are used to cast light on the general case. This work seems interesting and worthy of publication."""
309,"""Hyperbolic Attention Networks""","['Hyperbolic Geometry', 'Attention Methods', 'Reasoning on Graphs', 'Relation Learning', 'Scale Free Graphs', 'Transformers', 'Power Law']","""Recent approaches have successfully demonstrated the benefits of learning the parameters of shallow networks in hyperbolic space. We extend this line of work by imposing hyperbolic geometry on the embeddings used to compute the ubiquitous attention mechanisms for different neural networks architectures. By only changing the geometry of embedding of object representations, we can use the embedding space more efficiently without increasing the number of parameters of the model. Mainly as the number of objects grows exponentially for any semantic distance from the query, hyperbolic geometry --as opposed to Euclidean geometry-- can encode those objects without having any interference. Our method shows improvements in generalization on neural machine translation on WMT'14 (English to German), learning on graphs (both on synthetic and real-world graph tasks) and visual question answering (CLEVR) tasks while keeping the neural representations compact.""","""Reviewers all agree that this is a strong submission. I also believe it is interesting that only by changing the geometry of embeddings, they can use the space more efficiently without increasing the number of parameters."""
310,"""Psychophysical vs. learnt texture representations in novelty detection""","['novelty detection', 'learnt texture representation', 'one-class neural network', 'human-vision-inspired anomaly detection']","""Parametric texture models have been applied successfully to synthesize artificial images. Psychophysical studies show that under defined conditions observers are unable to differentiate between model-generated and original natural textures. In industrial applications the reverse case is of interest: a texture analysis system should decide if human observers are able to discriminate between a reference and a novel texture. For example, in case of inspecting decorative surfaces the de- tection of visible texture anomalies without any prior knowledge is required. Here, we implemented a human-vision-inspired novelty detection approach. Assuming that the features used for texture synthesis are important for human texture percep- tion, we compare psychophysical as well as learnt texture representations based on activations of a pretrained CNN in a novelty detection scenario. Additionally, we introduce a novel objective function to train one-class neural networks for novelty detection and compare the results to standard one-class SVM approaches. Our experiments clearly show the differences between human-vision-inspired texture representations and learnt features in detecting visual anomalies. Based on a dig- ital print inspection scenario we show that psychophysical texture representations are able to outperform CNN-encoded features.""","""This paper focuses on the problem of detecting visual anomalies within textures. For that purpose, the authors consider several parametric texture models and train anomaly detection models on the corresponding outputs. Reviewers were generally positive about the topic under study, but were unanimous in signaling a severe weaknesses in the experimental setup. In particular, in R2 words, ""my main concern is that the performance evaluation is not suitable to achieve meaningful results"", and ""showing quantitative results from only two textures does not feel like a very comprehensive analysis"". Moreover, the authors did not respond to reviewers feedback. Therefore, the AC recommends rejection at this time."""
311,"""An investigation of model-free planning""",[],"""The field of reinforcement learning (RL) is facing increasingly challenging domains with combinatorial complexity. For an RL agent to address these challenges, it is essential that it can plan effectively. Prior work has typically utilized an explicit model of the environment, combined with a specific planning algorithm (such as tree search). More recently, a new family of methods have been proposed that learn how to plan, by providing the structure for planning via an inductive bias in the function approximator (such as a tree structured neural network), trained end-to-end by a model-free RL algorithm. In this paper, we go even further, and demonstrate empirically that an entirely model-free approach, without special structure beyond standard neural network components such as convolutional networks and LSTMs, can learn to exhibit many of the hallmarks that we would typically associate with a model-based planner. We measure our agent's effectiveness at planning in terms of its ability to generalize across a combinatorial and irreversible state space, its data efficiency, and its ability to utilize additional thinking time. We find that our agent has the characteristics that one might expect to find in a planning algorithm. Furthermore, it exceeds the state-of-the-art in challenging combinatorial domains such as Sokoban and outperforms other model-free approaches that utilize strong inductive biases toward planning.""","""The paper studies a convolutional LSTM (ConvLSTM) based model (DRC: Deep Repeated ConvLSTM) trained through reinforcement, and shows that it performs better than other model-free approaches, in particular in term of generalization. The ability to generalize is attributed to being able to plan. This last part is not completely convincing. The paper is clearly written, the experiments are in 4 limited domains: Sokoban, Boxworld, MiniPacman, Gridworld. While diverse, tasks are still all similarly navigation in top-down (2D) grid worlds. It is unclear what are the limits of the reach of this study. The experimental evidence presented here could also be interpreted as: local best-response recognition of shapes (Conv) and memory of such patterns and associated actions (LSTM) are sufficient for all those environments. Overall, this is an interesting direction, but it falls slightly short of being acceptable for publication at ICLR."""
312,"""Local SGD Converges Fast and Communicates Little""","['optimization', 'communication', 'theory', 'stochastic gradient descent', 'SGD', 'mini-batch', 'local SGD', 'parallel restart SGD', 'distributed training']","""Mini-batch stochastic gradient descent (SGD) is state of the art in large scale distributed training. The scheme can reach a linear speed-up with respect to the number of workers, but this is rarely seen in practice as the scheme often suffers from large network delays and bandwidth limits. To overcome this communication bottleneck recent works propose to reduce the communication frequency. An algorithm of this type is local SGD that runs SGD independently in parallel on different workers and averages the sequences only once in a while. This scheme shows promising results in practice, but eluded thorough theoretical analysis. We prove concise convergence rates for local SGD on convex problems and show that it converges at the same rate as mini-batch SGD in terms of number of evaluated gradients, that is, the scheme achieves linear speed-up in the number of workers and mini-batch size. The number of communication rounds can be reduced up to a factor of T^{1/2}---where T denotes the number of total steps---compared to mini-batch SGD. This also holds for asynchronous implementations. Local SGD can also be used for large scale training of deep learning models. The results shown here aim serving as a guideline to further explore the theoretical and practical aspects of local SGD in these applications.""","""This paper analyzes local SGD optimization for strongly convex functions, and proves that local SGD enjoys a linear speedup (in the number of workers and minibatch size) over vanilla SGD, while also communicating less than distributed mini-batch SGD. A similar analysis is also provided for the asynchronous case, and limited empirical confirmation of the theory is provided. The main weakness of the current revision is that it does not yet properly relate this work to two prior publications: Dekel et al., 2012 (pseudo-url) and Jain et al., 2016 (pseudo-url). It is critical that these references and suitable discussion be added in the camera-ready paper, since this issue was the subject of considerable discussion and the authors promised to include the references and discussion in the final paper."""
313,"""Dynamic Graph Representation Learning via Self-Attention Networks""","['Graph Representation Learning', 'Dynamic Graphs', 'Attention', 'Self-Attention', 'Deep Learning']","""Learning latent representations of nodes in graphs is an important and ubiquitous task with widespread applications such as link prediction, node classification, and graph visualization. Previous methods on graph representation learning mainly focus on static graphs, however, many real-world graphs are dynamic and evolve over time. In this paper, we present Dynamic Self-Attention Network (DySAT), a novel neural architecture that operates on dynamic graphs and learns node representations that capture both structural properties and temporal evolutionary patterns. Specifically, DySAT computes node representations by jointly employing self-attention layers along two dimensions: structural neighborhood and temporal dynamics. We conduct link prediction experiments on two classes of graphs: communication networks and bipartite rating networks. Our experimental results show that DySAT has a significant performance gain over several different state-of-the-art graph embedding baselines.""","""This paper proposes a self-attention based approach for learning representations for the vertices of a dynamic graph, where the topology of the edges may change. The attention focuses on representing the interaction of vertices that have connections. Experimental results for the link prediction task on multiple datasets demonstrate the benefits of the approach. The idea of attention or its computation is not novel, however its application for estimating embeddings for dynamic graph vertices is new. The original version of the paper did not have strong baselines as noted by multiple reviewers, but the paper was revised during the review period. However, some of these suggestions, for example, experiments with larger graph sizes and other related work i.e., similar work on static graphs are left as a future work."""
314,"""Projective Subspace Networks For Few-Shot Learning""","['few-shot', 'one-shot', 'semi-supervised', 'meta-learning']","""Generalization from limited examples, usually studied under the umbrella of meta-learning, equips learning techniques with the ability to adapt quickly in dynamical environments and proves to be an essential aspect of lifelong learning. In this paper, we introduce the Projective Subspace Networks (PSN), a deep learning paradigm that learns non-linear embeddings from limited supervision. In contrast to previous studies, the embedding in PSN deems samples of a given class to form an affine subspace. We will show that such modeling leads to robust solutions, yielding competitive results on supervised and semi-supervised few-shot classification. Moreover, our PSN approach has the ability of end-to-end learning. In contrast to previous works, our projective subspace can be thought of as a richer representation capturing higher-order information datapoints for modeling new concepts.""","""The reviewers all like the idea, and though the performance is a little better when compared to prototypical networks, the reviewers felt that the contribution over and above prototypical networks was marginal and none of them was willing to champion the paper. There is merit in that there is increased robustness to outliers, and future iterations of the paper should work to strengthen this aspect. As a quick nitpick: based on my reading, and on Figure 3, it looks like there might be a typo in the definition of X_k (bottom of page 4). Right now it is defined in terms of the original data space x, when I think it should be defined in terms of the embedding space f(x). Overall this paper is a good contribution to the few-shot learning area."""
315,"""Small steps and giant leaps: Minimal Newton solvers for Deep Learning""",['deep learning'],"""We propose a fast second-order method that can be used as a drop-in replacement for current deep learning solvers. Compared to stochastic gradient descent (SGD), it only requires two additional forward-mode automatic differentiation operations per iteration, which has a computational cost comparable to two standard forward passes and is easy to implement. Our method addresses long-standing issues with current second-order solvers, which invert an approximate Hessian matrix every iteration exactly or by conjugate-gradient methods, procedures that are much slower than a SGD step. Instead, we propose to keep a single estimate of the gradient projected by the inverse Hessian matrix, and update it once per iteration with just two passes over the network. This estimate has the same size and is similar to the momentum variable that is commonly used in SGD. No estimate of the Hessian is maintained. We first validate our method, called CurveBall, on small problems with known solutions (noisy Rosenbrock function and degenerate 2-layer linear networks), where current deep learning solvers struggle. We then train several large models on CIFAR and ImageNet, including ResNet and VGG-f networks, where we demonstrate faster convergence with no hyperparameter tuning. We also show our optimiser's generality by testing on a large set of randomly-generated architectures.""","""The proposal is a scheme for using implicit matrix-vector products to exploit curvature information for neural net optimization, roughly based on the adaptive learning rate and momentum tricks from Martens and Grosse (2015). The paper is well-written, and the proposed method seems like a reasonable thing to try. I don't see any critical flaws in the methods. While there was a long discussion between R1 and the authors on many detailed points, most of the points R1 raises seem very minor, and authors' response to the conceptual points seems satisfactory. In terms of novelty, the method is mostly a remixing of ideas that have already appeared in the neural net optimization literature. There is sufficient novelty to justify acceptance if there were strong experimental results, but in my opinion not enough for the conceptual contributions to stand on their own. There is not much evidence of a real optimization improvement. The per-epoch improvement over SGD is fairly small, and (as the reviewers point out) probably outweighed by the factor-of-2 computational overhead, so it's likely there is no wall-clock improvement. Other details of the experimental setup seem concerning; e.g., if I understand right, the SGD training curve flatlines because the SGD parameters were tuned for validation accuracy rather than training accuracy (as is reported). The only comparison to another second-order method is to K-FAC on an MNIST MLP, even though K-FAC and other methods have been applied to much larger-scale models. I think there's a promising idea here which could make a strong paper if the theory or experiments were further developed. But I can't recommend acceptance in its current form. """
316,"""Novel positional encodings to enable tree-structured transformers""","['program translation', 'tree structures', 'transformer']","""With interest in program synthesis and similarly avored problems rapidly increasing, neural models optimized for tree-domain problems are of great value. In the sequence domain, transformers can learn relationships across arbitrary pairs of positions with less bias than recurrent models. Under the intuition that a similar property would be beneficial in the tree domain, we propose a method to extend transformers to tree-structured inputs and/or outputs. Our approach abstracts transformer's default sinusoidal positional encodings, allowing us to substitute in a novel custom positional encoding scheme that represents node positions within a tree. We evaluated our model in tree-to-tree program translation and sequence-to-tree semantic parsing settings, achieving superior performance over the vanilla transformer model on several tasks. ""","""This paper extends the transformer model of Vashwani et al. by replacing the sine/cosine positional encodings with information reflecting the tree stucture of appropriately parsed data. According to the reviews, the paper, while interesting, does not make the cut. My concern here is that the quality of the reviews, in particular those of reviewers 2 and 3, is very sub par. They lack detail (or, in the case of R2, did so until 05 Dec(!!)), and the reviewers did not engage much (or at all) in the subsequent discussion period despite repeated reminders. Infuriatingly, this puts a lot of work squarely in the lap of the AC: if the review process fails the authors, I cannot make a decision on the basis of shoddy reviews and inexistent discussion! Clearly, as this is not the fault of the authors, the best I can offer is to properly read through the paper and reviews, and attempt to make a fair assessment. Having done so, I conclude that while interesting, I agree with the sentiment expressed in the reviews that the paper is very incremental. In particular, the points of comparison are quite limited and it would have been good to see a more thorough comparison across a wider range of tasks with some more contemporary baselines. Papers like Melis et al. 2017 have shown us that an endemic issue throughout language modelling (and certainly also other evaluation areas) is that complex model improvements are offered without comparison against properly tuned baselines and benchmarks, failing to offer assurances that the baselines would not match performance of the proposed model with proper regularisation. As some of the reviewers, the scope of comparison to prior art in this paper is extremely limited, as is the bibliography, which opens up this concern I've just outlined that it's difficult to take the results with the confidence they require. In short, my assessment, on the basis of reading the paper and reviews, is that the main failing of this paper is the lack of breadth and depth of evaluation, not that it is incremental (as many good ideas are). I'm afraid this paper is not ready for publication at this time, and am sorry the authors will have had a sub-par review process, but I believe it's in the best interest of this work to encourage the authors to further evaluate their approach before publishing it in conference proceedings."""
317,"""Super-Resolution via Conditional Implicit Maximum Likelihood Estimation""",['super-resolution'],"""Single-image super-resolution (SISR) is a canonical problem with diverse applications. Leading methods like SRGAN produce images that contain various artifacts, such as high-frequency noise, hallucinated colours and shape distortions, which adversely affect the realism of the result. In this paper, we propose an alternative approach based on an extension of the method of Implicit Maximum Likelihood Estimation (IMLE). We demonstrate greater effectiveness at noise reduction and preservation of the original colours and shapes, yielding more realistic super-resolved images. ""","""The main novelty of the paper lies in using multiple noise vectors to reconstruct the high resolution image in multiple ways. Then, the reconstruction with minimal loss is selected and updated to improve the fit against the target image. The most important control experiment in my opinion should compare this approach against the same architecture with only with m=1 noise vector (i.e., using a constant noise vector all the time). Unfortunately, the paper does not include such a comparison, which means the main hypothesis of the paper is not tested. Please include this experiment in the revised version of the paper. PS: There is another high level concern regarding the use of PSNR or SSIM for evaluation of super-resolution methods. As shown by ""Pixel recursive super resolution (Dahl et al.)"" and others, PSNR and SSIM metrics are only relevant in the low magnification regime, in which techniques based on MSE (mean squared error) are very competitive. Maybe you need to consider large magnification regime in which GAN and normalized flow-based models are more relevant."""
318,"""Modeling the Long Term Future in Model-Based Reinforcement Learning""","['model-based reinforcement learning', 'variation inference']","""In model-based reinforcement learning, the agent interleaves between model learning and planning. These two components are inextricably intertwined. If the model is not able to provide sensible long-term prediction, the executed planer would exploit model flaws, which can yield catastrophic failures. This paper focuses on building a model that reasons about the long-term future and demonstrates how to use this for efficient planning and exploration. To this end, we build a latent-variable autoregressive model by leveraging recent ideas in variational inference. We argue that forcing latent variables to carry future information through an auxiliary task substantially improves long-term predictions. Moreover, by planning in the latent space, the planner's solution is ensured to be within regions where the model is valid. An exploration strategy can be devised by searching for unlikely trajectories under the model. Our methods achieves higher reward faster compared to baselines on a variety of tasks and environments in both the imitation learning and model-based reinforcement learning settings. ""","""This paper explores the use of multi-step latent variable models of the dynamics in imitation learning, planning, and finding sub-goals. The reviewers found the approach to be interesting. The initial experiments were a main weakpoint in the initial submission. However, the authors updated the experimental results to address these concerns to a significant degree. The reviewers all agree that the paper is above the bar for acceptance. I recommend accept."""
319,"""Adversarial Audio Super-Resolution with Unsupervised Feature Losses""",[],"""Neural network-based methods have recently demonstrated state-of-the-art results on image synthesis and super-resolution tasks, in particular by using variants of generative adversarial networks (GANs) with supervised feature losses. Nevertheless, previous feature loss formulations rely on the availability of large auxiliary classifier networks, and labeled datasets that enable such classifiers to be trained. Furthermore, there has been comparatively little work to explore the applicability of GAN-based methods to domains other than images and video. In this work we explore a GAN-based method for audio processing, and develop a convolutional neural network architecture to perform audio super-resolution. In addition to several new architectural building blocks for audio processing, a key component of our approach is the use of an autoencoder-based loss that enables training in the GAN framework, with feature losses derived from unlabeled data. We explore the impact of our architectural choices, and demonstrate significant improvements over previous works in terms of both objective and perceptual quality.""","""The paper presents an algorithm for audio super-resolution using adversarial models along with additional losses, e.g. using auto-encoders and reconstruction losses, to improve the generation process. Strengths - Proposes audio super resolution based on GANs, extending some of the techniques proposed for vision / image to audio. - The authors improved the paper during the review process by including results from a user study and ablation analysis. Weaknesses - Although the paper presents an interesting application of GANs for the audio task, overall novelty is limited since the setup closely follows what has been done for vision and related tasks, and the baseline system. This is also not the first application of GANs for audio tasks. - Performance improvement over previously proposed (U-Net) models is small. It would have been useful to also include UNet4 in user-study, as one of the reviewers pointed out, since it sounds better in a few cases. - It is not entirely clear if the method would be an improvement of state-of-the-art audio generative models like Wavenet. Reviewers agree that the general direction of this work is interesting, but the results are not compelling enough at the moment for the paper to be accepted to ICLR. Given these review comments, the recommendation is to reject the paper."""
320,"""Unsupervised Hyper-alignment for Multilingual Word Embeddings""",[],"""We consider the problem of aligning continuous word representations, learned in multiple languages, to a common space. It was recently shown that, in the case of two languages, it is possible to learn such a mapping without supervision. This paper extends this line of work to the problem of aligning multiple languages to a common space. A solution is to independently map all languages to a pivot language. Unfortunately, this degrades the quality of indirect word translation. We thus propose a novel formulation that ensures composable mappings, leading to better alignments. We evaluate our method by jointly aligning word vectors in eleven languages, showing consistent improvement with indirect mappings while maintaining competitive performance on direct word translation.""","""This paper provides a simple and intuitive method for learning multilingual word embeddings that makes it possible to softly encourage the model to align the spaces of non-English language pairs. The results are better than learning just pairwise embeddings with English. The main remaining concern (in my mind) after the author response is that the method is less accurate empirically than Chen and Cardie (2018). I think however that given that these two works are largely contemporaneous, the methods are appreciably different, and the proposed method also has advantages with respect to speed, that the paper here is still a reasonably candidate for acceptance at ICLR. However, I would like to request that in the final version the authors feature Chen and Cardie (2018) more prominently in the introduction and discuss the theoretical and empirical differences between the two methods. This will make sure that readers get the full picture of the two works and understand their relative differences and advantages/disadvantages."""
321,"""On the Convergence of A Class of Adam-Type Algorithms for Non-Convex Optimization""","['nonconvex optimization', 'Adam', 'convergence analysis']","""This paper studies a class of adaptive gradient based momentum algorithms that update the search directions and learning rates simultaneously using past gradients. This class, which we refer to as the ''``Adam-type'', includes the popular algorithms such as Adam, AMSGrad, AdaGrad. Despite their popularity in training deep neural networks (DNNs), the convergence of these algorithms for solving non-convex problems remains an open question. In this paper, we develop an analysis framework and a set of mild sufficient conditions that guarantee the convergence of the Adam-type methods, with a convergence rate of order pseudo-formula for non-convex stochastic optimization. Our convergence analysis applies to a new algorithm called AdaFom (AdaGrad with First Order Momentum). We show that the conditions are essential, by identifying concrete examples in which violating the conditions makes an algorithm diverge. Besides providing one of the first comprehensive analysis for Adam-type methods in the non-convex setting, our results can also help the practitioners to easily monitor the progress of algorithms and determine their convergence behavior. ""","""This paper analysis the convergence properties of a family of 'Adam-Type' optimization algorithms, such as Adam, Amsgrad and AdaGrad, in the non-convex setting. The paper provides of the first comprehensive analyses of such algorithms in the non-convex setting. In addition, the results can help practitioners with monitoring convergence in experiments. Since Adam is a widely used method, the results have a potentially large impact. The reviewers agree that the paper is well-written, provides interesting new insights, and that is results are of sufficient interest to the ICLR community to be worthy of publication."""
322,"""Competitive experience replay""","['reinforcement learning', 'sparse reward', 'goal-based learning']","""Deep learning has achieved remarkable successes in solving challenging reinforcement learning (RL) problems when dense reward function is provided. However, in sparse reward environment it still often suffers from the need to carefully shape reward function to guide policy optimization. This limits the applicability of RL in the real world since both reinforcement learning and domain-specific knowledge are required. It is therefore of great practical importance to develop algorithms which can learn from a binary signal indicating successful task completion or other unshaped, sparse reward signals. We propose a novel method called competitive experience replay, which efficiently supplements a sparse reward by placing learning in the context of an exploration competition between a pair of agents. Our method complements the recently proposed hindsight experience replay (HER) by inducing an automatic exploratory curriculum. We evaluate our approach on the tasks of reaching various goal locations in an ant maze and manipulating objects with a robotic arm. Each task provides only binary rewards indicating whether or not the goal is achieved. Our method asymmetrically augments these sparse rewards for a pair of agents each learning the same task, creating a competitive game designed to drive exploration. Extensive experiments demonstrate that this method leads to faster converge and improved task performance.""","""The paper proposes a new method to improve exploration in sparse reward problems, by having two agents competing with each other to generate shaping reward that relies on how novel a newly visited state is. The idea is nice and simple, and the results are promising. The authors implemented more baselines suggested in initial reviews, which was also helpful. On the other hand, the approach appears somewhat ad hoc. It is not always clear why (and when) the method works, although some intuitions are given. One reviewer gave a nice suggestion of obtaining further insights by running experiments in less complex environments. Overall, this work is an interesting contribution."""
323,"""PRUNING WITH HINTS: AN EFFICIENT FRAMEWORK FOR MODEL ACCELERATION""","['model acceleration', 'mimic', 'knowledge distillation', 'channel pruning']","""In this paper, we propose an efficient framework to accelerate convolutional neural networks. We utilize two types of acceleration methods: pruning and hints. Pruning can reduce model size by removing channels of layers. Hints can improve the performance of student model by transferring knowledge from teacher model. We demonstrate that pruning and hints are complementary to each other. On one hand, hints can benefit pruning by maintaining similar feature representations. On the other hand, the model pruned from teacher networks is a good initialization for student model, which increases the transferability between two networks. Our approach performs pruning stage and hints stage iteratively to further improve the performance. Furthermore, we propose an algorithm to reconstruct the parameters of hints layer and make the pruned model more suitable for hints. Experiments were conducted on various tasks including classification and pose estimation. Results on CIFAR-10, ImageNet and COCO demonstrate the generalization and superiority of our framework.""","""This paper proposes a new framework which combines pruning and model distillation techniques for model acceleration. The reviewers have a consensus on rejection due to limited novelty."""
324,"""Episodic Curiosity through Reachability""","['deep learning', 'reinforcement learning', 'curiosity', 'exploration', 'episodic memory']","""Rewards are sparse in the real world and most of today's reinforcement learning algorithms struggle with such sparsity. One solution to this problem is to allow the agent to create rewards for itself - thus making rewards dense and more suitable for learning. In particular, inspired by curious behaviour in animals, observing something novel could be rewarded with a bonus. Such bonus is summed up with the real task reward - making it possible for RL algorithms to learn from the combined reward. We propose a new curiosity method which uses episodic memory to form the novelty bonus. To determine the bonus, the current observation is compared with the observations in memory. Crucially, the comparison is done based on how many environment steps it takes to reach the current observation from those in memory - which incorporates rich information about environment dynamics. This allows us to overcome the known ""couch-potato"" issues of prior work - when the agent finds a way to instantly gratify itself by exploiting actions which lead to hardly predictable consequences. We test our approach in visually rich 3D environments in ViZDoom, DMLab and MuJoCo. In navigational tasks from ViZDoom and DMLab, our agent outperforms the state-of-the-art curiosity method ICM. In MuJoCo, an ant equipped with our curiosity module learns locomotion out of the first-person-view curiosity only. The code is available at pseudo-url.""",""" The authors present a novel method for tackling exploration and exploitation that yields promising results on some hard navigation-like domains. The reviewers were impressed by the contribution and had some suggestions for improvement that should be addressed in the camera ready version. """
325,"""Learning to Reinforcement Learn by Imitation""","['meta-learning', 'reinforcement learning', 'imitation learning']","""Meta-reinforcement learning aims to learn fast reinforcement learning (RL) procedures that can be applied to new tasks or environments. While learning fast RL procedures holds promise for allowing agents to autonomously learn a diverse range of skills, existing methods for learning efficient RL are impractical for real world settings, as they rely on slow reinforcement learning algorithms for meta-training, even when the learned procedures are fast. In this paper, we propose to learn a fast reinforcement learning procedure through supervised imitation of an expert, such that, after meta-learning, an agent can quickly learn new tasks through trial-and-error. Through our proposed method, we show that it is possible to learn fast RL using demonstrations, rather than relying on slow RL, where expert agents can be trained quickly by using privileged information or off-policy RL methods. Our experimental evaluation on a number of complex simulated robotic domains demonstrates that our method can effectively learn to learn from spare rewards and is significantly more efficient than prior meta reinforcement learning algorithms.""","""This paper proposes a meta-learning algorithm for reinforcement learning that incorporates expert demonstrations. The objective is to improve sample efficiency, which is an important problem. The referees find the approach well-motivated and pertinent, but the theoretical and practical contributions of the paper too slim. A concern was also raised in regard to reproducibility of the results, missing details about the implementation and comparisons with previous results. The authors did not respond to the reviews. The four referees are not convinced by this paper, with ratings from strong reject to ok, but not good enough. """
326,"""A Direct Approach to Robust Deep Learning Using Adversarial Networks""","['deep learning', 'adversarial learning', 'generative adversarial networks']","""Deep neural networks have been shown to perform well in many classical machine learning problems, especially in image classification tasks. However, researchers have found that neural networks can be easily fooled, and they are surprisingly sensitive to small perturbations imperceptible to humans. Carefully crafted input images (adversarial examples) can force a well-trained neural network to provide arbitrary outputs. Including adversarial examples during training is a popular defense mechanism against adversarial attacks. In this paper we propose a new defensive mechanism under the generative adversarial network~(GAN) framework. We model the adversarial noise using a generative network, trained jointly with a classification discriminative network as a minimax game. We show empirically that our adversarial network approach works well against black box attacks, with performance on par with state-of-art methods such as ensemble adversarial training and adversarial training with projected gradient descent. ""","""The paper proposed a GAN approach to robust learning against adversarial examples, where a generator produces adversarial examples as perturbations and a discriminator is used to distinguish between adversarial and raw images. The performance on MNIST, SVHN, and CIFAR10 demonstrate the effectiveness of the approach, and in general, the performance is on par with carefully crafted algorithms for such task. The architecture of GANs used in the paper is standard, yet the defensive performance seems good. The reviewers wonder the reason behind this good mechanism and the novelty compared with other works in similar spirits. In response, the authors add some insights on discussing the mechanism as well as comparisons with other works mentioned by the reviewers. The reviewers all think that the paper presents a simple scheme for robust deep learning based on GANs, which shows its effectiveness in experiments. The understanding on why it works may need further explorations. Thus the paper is proposed to be borderline lean accept. """
327,"""Link Prediction in Hypergraphs using Graph Convolutional Networks""","['Graph convolution', 'hypergraph', 'hyperlink prediction']","""Link prediction in simple graphs is a fundamental problem in which new links between nodes are predicted based on the observed structure of the graph. However, in many real-world applications, there is a need to model relationships among nodes which go beyond pairwise associations. For example, in a chemical reaction, relationship among the reactants and products is inherently higher-order. Additionally, there is need to represent the direction from reactants to products. Hypergraphs provide a natural way to represent such complex higher-order relationships. Even though Graph Convolutional Networks (GCN) have recently emerged as a powerful deep learning-based approach for link prediction over simple graphs, their suitability for link prediction in hypergraphs is unexplored -- we fill this gap in this paper and propose Neural Hyperlink Predictor (NHP). NHP adapts GCNs for link prediction in hypergraphs. We propose two variants of NHP --NHP-U and NHP-D -- for link prediction over undirected and directed hypergraphs, respectively. To the best of our knowledge, NHP-D is the first method for link prediction over directed hypergraphs. Through extensive experiments on multiple real-world datasets, we show NHP's effectiveness.""","""The paper describes a method for the link prediction problem in both directed and undirected hypergraphs. While the problem discussed in the paper is clearly importnant and interesting, all reviewers agree that the novelty of the proposed approach is somewhat limited given the prior art."""
328,"""Selective Self-Training for semi-supervised Learning""","['deep learning', 'image recognition', 'semi-supervised learning']","""Semi-supervised learning (SSL) is a study that efficiently exploits a large amount of unlabeled data to improve performance in conditions of limited labeled data. Most of the conventional SSL methods assume that the classes of unlabeled data are included in the set of classes of labeled data. In addition, these methods do not sort out useless unlabeled samples and use all the unlabeled data for learning, which is not suitable for realistic situations. In this paper, we propose an SSL method called selective self-training (SST), which selectively decides whether to include each unlabeled sample in the training process. It is also designed to be applied to a more real situation where classes of unlabeled data are different from the ones of the labeled data. For the conventional SSL problems which deal with data where both the labeled and unlabeled samples share the same class categories, the proposed method not only performs comparable to other conventional SSL algorithms but also can be combined with other SSL algorithms. While the conventional methods cannot be applied to the new SSL problems where the separated data do not share the classes, our method does not show any performance degradation even if the classes of unlabeled data are different from those of the labeled data.""","""Reviewers have concerns about poor writing of the paper, lack of technical novelty, and the methodology taken by the paper not being very principled. """
329,"""Measuring and regularizing networks in function space""","['function space', 'Hilbert space', 'empirical characterization', 'multitask learning', 'catastrophic forgetting', 'optimization', 'natural gradient']","""To optimize a neural network one often thinks of optimizing its parameters, but it is ultimately a matter of optimizing the function that maps inputs to outputs. Since a change in the parameters might serve as a poor proxy for the change in the function, it is of some concern that primacy is given to parameters but that the correspondence has not been tested. Here, we show that it is simple and computationally feasible to calculate distances between functions in a pseudo-formula Hilbert space. We examine how typical networks behave in this space, and compare how parameter pseudo-formula distances compare to function pseudo-formula distances between various points of an optimization trajectory. We find that the two distances are nontrivially related. In particular, the pseudo-formula ratio decreases throughout optimization, reaching a steady value around when test error plateaus. We then investigate how the pseudo-formula distance could be applied directly to optimization. We first propose that in multitask learning, one can avoid catastrophic forgetting by directly limiting how much the input/output function changes between tasks. Secondly, we propose a new learning rule that constrains the distance a network can travel through pseudo-formula -space in any one update. This allows new examples to be learned in a way that minimally interferes with what has previously been learned. These applications demonstrate how one can measure and regularize function distances directly, without relying on parameters or local approximations like loss curvature.""","""This paper proposes to regularize neural network in function space rather than in parameter space, a proposal which makes sense and is also different than the natural gradient approach. After discussion and considering the rebuttal, all reviewers argue for acceptance. The AC does agree that this direction of research is an important one for deep learning, and while the paper could benefit from revision and tightening the story (and stronger experiments); these do not preclude publishing in its current state. Side comment: the visualization of neural networks in function space was done profusely when the effect of unsupervised pre-training on neural networks was investigated (among others). See e.g. Figure 7 in Erhan et al. AISTATS 2009 ""The Difficulty of Training Deep Architectures and the Effect of Unsupervised Pre-Training"". This literature should be cited (and it seems that tSNE might be a more appropriate visualization techniques for non-linear functions than MDS)."""
330,"""Transfer Learning for Sequences via Learning to Collocate""","['transfer learning', 'recurrent neural network', 'attention', 'natural language processing']","""Transfer learning aims to solve the data sparsity for a specific domain by applying information of another domain. Given a sequence (e.g. a natural language sentence), the transfer learning, usually enabled by recurrent neural network (RNN), represent the sequential information transfer. RNN uses a chain of repeating cells to model the sequence data. However, previous studies of neural network based transfer learning simply transfer the information across the whole layers, which are unfeasible for seq2seq and sequence labeling. Meanwhile, such layer-wise transfer learning mechanisms also lose the fine-grained cell-level information from the source domain. In this paper, we proposed the aligned recurrent transfer, ART, to achieve cell-level information transfer. ART is in a recurrent manner that different cells share the same parameters. Besides transferring the corresponding information at the same position, ART transfers information from all collocated words in the source domain. This strategy enables ART to capture the word collocation across domains in a more flexible way. We conducted extensive experiments on both sequence labeling tasks (POS tagging, NER) and sentence classification (sentiment analysis). ART outperforms the state-of-the-arts over all experiments. ""","""This paper presents a method for transferring source information via the hidden states of recurrent networks. The transfer happens via an attention mechanism that operates between the target and the source. Results on two tasks are strong. I found this paper similar in spirit to Hypernetworks (David Ha, Andrew Dai, Quoc V Le, ICLR 2016) since there too there is a dynamic weight generation for network given another network, although this method did not use an attention mechanism. However, reviewers thought that there is merit in this paper (albeit pointed the authors to other related work) and the empirical results are solid. """
331,"""An Exhaustive Analysis of Lazy vs. Eager Learning Methods for Real-Estate Property Investment""","['applied machine learning', 'housing analytics', 'eager learning', 'lazy learning', 'rent prediction']","""Accurate rent prediction in real estate investment can help in generating capital gains and guaranty a financial success. In this paper, we carry out a comprehensive analysis and study of eleven machine learning algorithms for rent prediction, including Linear Regression, Multilayer Perceptron, Random Forest, KNN, ML-KNN, Locally Weighted Learning, SMO, SVM, J48, lazy Decision Tree (i.e., lazy DT), and KStar algorithms. Our contribution in this paper is twofold: (1) We present a comprehensive analysis of internal and external attributes of a real-estate housing dataset and their correlation with rental prices. (2) We use rental prediction as a platform to study and compare the performance of eager vs. lazy machine learning methods using myriad of ML algorithms. We train our rent prediction models using a Zillow data set of 4K real estate properties in Virginia State of the US, including three house types of single-family, townhouse, and condo. Each data instance in the dataset has 21 internal attributes (e.g., area space, price, number of bed/bath, rent, school rating, so forth). In addition to Zillow data, external attributes like walk/transit score, and crime rate are collected from online data sources. A subset of the collected features - determined by the PCA technique- are selected to tune the parameters of the prediction models. We employ a hierarchical clustering approach to cluster the data based on two factors of house type, and average rent estimate of zip codes. We evaluate and compare the efficacy of the tuned prediction models based on two metrics of R-squared and Mean Absolute Error, applied on unseen data. Based on our study, lazy models like KStar lead to higher accuracy and lower prediction error compared to eager methods like J48 and LR. However, it is not necessarily found to be an overarching conclusion drawn from the comparison between all the lazy and eager methods in this work. ""","""The paper evaluates several off-the-shelf algorithms for predicting the return on real-estate property investment. The problem is interesting, but there is a consensus that the paper contains little technical novelty, and the empirical study on a fairly small dataset is also not convincing. """
332,"""NUTS: Network for Unsupervised Telegraphic Summarization""","['nlp', 'summarization', 'unsupervised learning', 'deep learning']","""Extractive summarization methods operate by ranking and selecting the sentences which best encapsulate the theme of a given document. They do not fare well in domains like fictional narratives where there is no central theme and core information is not encapsulated by a small set of sentences. For the purpose of reducing the size of the document while conveying the idea expressed by each sentence, we need more sentence specific methods. Telegraphic summarization, which selects short segments across several sentences, is better suited for such domains. Telegraphic summarization captures the plot better by retaining shorter versions of each sentence while not really concerning itself with grammatically linking these segments. In this paper, we propose an unsupervised deep learning network (NUTS) to generate telegraphic summaries. We use multiple encoder-decoder networks and learn to drop portions of the text that are inferable from the chosen segments. The model is agnostic to both sentence length and style. We demonstrate that the summaries produced by our model show significant quantitative and qualitative improvement over those produced by existing methods and baselines.""","""This paper presents methods for telegraphic summarization, a task that generates extremely short summaries. There are concerns about the utility of the task in general, and also the novelty of the modeling framework. There is overall consensus between reviewers regarding the paper's assessment the feedback is lukewarm."""
333,"""A comprehensive, application-oriented study of catastrophic forgetting in DNNs""","['incremental learning', 'deep neural networks', 'catatrophic forgetting', 'sequential learning']","""We present a large-scale empirical study of catastrophic forgetting (CF) in modern Deep Neural Network (DNN) models that perform sequential (or: incremental) learning. A new experimental protocol is proposed that takes into account typical constraints encountered in application scenarios. As the investigation is empirical, we evaluate CF behavior on the hitherto largest number of visual classification datasets, from each of which we construct a representative number of Sequential Learning Tasks (SLTs) in close alignment to previous works on CF. Our results clearly indicate that there is no model that avoids CF for all investigated datasets and SLTs under application conditions. We conclude with a discussion of potential solutions and workarounds to CF, notably for the EWC and IMM models.""","""This paper has two main contributions. The first is that it proposes a specific framework for measuring catastrophic forgetting in deep neural networks that incorporates three application-oriented constraints: (1) a low memory footprint, which implies that data from prior tasks cannot be retained; (2) causality, meaning that data from future tasks cannot be used in any way, including hyperparameter optimization and model selection; and (3) update complexity for new tasks that is moderate and also independent of the number of previously learned tasks, which precludes replay strategies. The second contribution is an extensive study of catastrophic forgetting, using different sequential learning tasks derived from 9 different datasets and examining 7 different models. The key conclusions from the study are that (1) permutation-based tasks are comparatively easy and should not be relied on to measure catastrophic forgetting; (2) with the application-oriented contraints in effect, all of the examined models suffer from catastrophic forgetting (a result that is contrary to a number of other recent papers); (3) elastic weight consolidation provides some protection against catastrophic forgetting for simple sequential learning tasks, but fails for more complex tasks; and (4) IMM is effective, but only if causality is violated in the selection of the IMM balancing parameter. The reviewer scores place this paper close to the decision boundary. The most negative reviewer (R2) had concerns about the novelty of the framework and its application-oriented constraints. The authors contend that recent papers on catastrophic forgetting fail to apply these quite natural constraints, leading to the deceptive conclusion that catastrophic forgetting may not be as big of a problem as it once was. The AC read a number of the papers mentioned by the authors and agrees with them: these constraints have been, at least at times, ignored in the literature, and they shouldn't be ignored. The other two reviewers appreciated the scope and rigor of the empirical study. On the balance, the AC thinks this is an important contribution and that it should appear at ICLR."""
334,"""SENSE: SEMANTICALLY ENHANCED NODE SEQUENCE EMBEDDING""","['Semantic', 'Graph', 'Sequence', 'Embeddings']","""Effectively capturing graph node sequences in the form of vector embeddings is critical to many applications. We achieve this by (i) first learning vector embeddings of single graph nodes and (ii) then composing them to compactly represent node sequences. Specifically, we propose SENSE-S (Semantically Enhanced Node Sequence Embedding - for Single nodes), a skip-gram based novel embedding mechanism, for single graph nodes that co-learns graph structure as well as their textual descriptions. We demonstrate that SENSE-S vectors increase the accuracy of multi-label classification tasks by up to 50% and link-prediction tasks by up to 78% under a variety of scenarios using real datasets. Based on SENSE-S, we next propose generic SENSE to compute composite vectors that represent a sequence of nodes, where preserving the node order is important. We prove that this approach is efficient in embedding node sequences, and our experiments on real data confirm its high accuracy in node order decoding.""","""The paper can also improved thorough a more thorough evaluation. """
335,"""h-detach: Modifying the LSTM Gradient Towards Better Optimization""","['LSTM', 'Optimization', 'Long term dependencies', 'Back-propagation through time']","""Recurrent neural networks are known for their notorious exploding and vanishing gradient problem (EVGP). This problem becomes more evident in tasks where the information needed to correctly solve them exist over long time scales, because EVGP prevents important gradient components from being back-propagated adequately over a large number of steps. We introduce a simple stochastic algorithm (\textit{h}-detach) that is specific to LSTM optimization and targeted towards addressing this problem. Specifically, we show that when the LSTM weights are large, the gradient components through the linear path (cell state) in the LSTM computational graph get suppressed. Based on the hypothesis that these components carry information about long term dependencies (which we show empirically), their suppression can prevent LSTMs from capturing them. Our algorithm\footnote{Our code is available at pseudo-url.} prevents gradients flowing through this path from getting suppressed, thus allowing the LSTM to capture such dependencies better. We show significant improvements over vanilla LSTM gradient based training in terms of convergence speed, robustness to seed and learning rate, and generalization using our modification of LSTM gradient on various benchmark datasets.""","""This paper presents a method for preventing exploding and vanishing gradients in LSTMs by stochastically blocking some paths of the information flow (but not others). Experiments show improved training speed and robustness to hyperparameter settings. I'm concerned about the quality of R2, since (as the authors point out) some of the text is copied verbatim from the paper. The other two reviewers are generally positive about the paper, with scores of 6 and 7, and R1 in particular points out that this work has already had noticeable impact in the field. While the reviewers pointed out some minor concerns with the experiments, there don't seem to be any major flaws. I think the paper is above the bar for acceptance. """
336,"""Synthnet: Learning synthesizers end-to-end""","['audio', 'synthesizers', 'music', 'convolutional neural networks', 'generative models', 'autoregressive models']","""Learning synthesizers and generating music in the raw audio domain is a challenging task. We investigate the learned representations of convolutional autoregressive generative models. Consequently, we show that mappings between musical notes and the harmonic style (instrument timbre) can be learned based on the raw audio music recording and the musical score (in binary piano roll format). Our proposed architecture, SynthNet uses minimal training data (9 minutes), is substantially better in quality and converges 6 times faster than the baselines. The quality of the generated waveforms (generation accuracy) is sufficiently high that they are almost identical to the ground truth. Therefore, we are able to directly measure generation error during training, based on the RMSE of the Constant-Q transform. Mean opinion scores are also provided. We validate our work using 7 distinct harmonic styles and also provide visualizations and links to all generated audio.""","""The paper describes a WaveNet-like model for MIDI-conditional music audio generation. As noted by all reviewers, the major limitation of the paper is that the method is evaluated on a synthetic dataset. The rebuttal and post-rebuttal discussion didn't change the reviewers' opinion."""
337,"""DVOLVER: Efficient Pareto-Optimal Neural Network Architecture Search""","['architecture search', 'Pareto optimality', 'multi-objective', 'optimization', 'cnn', 'deep learning']","""Automatic search of neural network architectures is a standing research topic. In addition to the fact that it presents a faster alternative to hand-designed architectures, it can improve their efficiency and for instance generate Convolutional Neural Networks (CNN) adapted for mobile devices. In this paper, we present a multi-objective neural architecture search method to find a family of CNN models with the best accuracy and computational resources tradeoffs, in a search space inspired by the state-of-the-art findings in neural search. Our work, called Dvolver, evolves a population of architectures and iteratively improves an approximation of the optimal Pareto front. Applying Dvolver on the model accuracy and on the number of floating points operations as objective functions, we are able to find, in only 2.5 days 1 , a set of competitive mobile models on ImageNet. Amongst these models one architecture has the same Top-1 accuracy on ImageNet as NASNet-A mobile with 8% less floating point operations and another one has a Top-1 accuracy of 75.28% on ImageNet exceeding by 0.28% the best MobileNetV2 model for the same computational resources.""","""The paper describes an architecture search method which optimises multiple objectives using a genetic algorithm. All reviewers agree on rejection due to limited novelty compared to the prior art; while the results are solid, they are not ground-breaking to justify acceptance based on results alone. """
338,"""Training for Faster Adversarial Robustness Verification via Inducing ReLU Stability""","['verification', 'adversarial robustness', 'adversarial examples', 'stability', 'deep learning', 'regularization']","""We explore the concept of co-design in the context of neural network verification. Specifically, we aim to train deep neural networks that not only are robust to adversarial perturbations but also whose robustness can be verified more easily. To this end, we identify two properties of network models - weight sparsity and so-called ReLU stability - that turn out to significantly impact the complexity of the corresponding verification task. We demonstrate that improving weight sparsity alone already enables us to turn computationally intractable verification problems into tractable ones. Then, improving ReLU stability leads to an additional 4-13x speedup in verification times. An important feature of our methodology is its ""universality,"" in the sense that it can be used with a broad range of training procedures and verification approaches. ""","""This paper introduced a concept called ReLU stability to motivate regularization and enable fast verification. Most of the analysis was presented empirically on two simple datasets and with low-performing models. I feel theoretical analysis and more comprehensive and realistic empirical studies would make the paper stronger. In general, the contribution of this paper is original and interesting. """
339,"""Preferences Implicit in the State of the World""","['Preference learning', 'Inverse reinforcement learning', 'Inverse optimal stochastic control', 'Maximum entropy reinforcement learning', 'Apprenticeship learning']","""Reinforcement learning (RL) agents optimize only the features specified in a reward function and are indifferent to anything left out inadvertently. This means that we must not only specify what to do, but also the much larger space of what not to do. It is easy to forget these preferences, since these preferences are already satisfied in our environment. This motivates our key insight: when a robot is deployed in an environment that humans act in, the state of the environment is already optimized for what humans want. We can therefore use this implicit preference information from the state to fill in the blanks. We develop an algorithm based on Maximum Causal Entropy IRL and use it to evaluate the idea in a suite of proof-of-concept environments designed to show its properties. We find that information from the initial state can be used to infer both side effects that should be avoided as well as preferences for how the environment should be organized. Our code can be found at pseudo-url.""","""The paper proposes to take advantage of implicit preferential information in a single state, to design auxiliary reward functions that can be combined with the standard RL reward function. The motivation is to use the implicit information to infer signals that might not have been included in the reward function. The paper has some nice ideas and is quite novel. A new algorithm is developed, and is supported by proof-of-concept experiments. Overall, the paper is a nice and novel contribution. But reviewers point out several limitations. The biggest one seems to be related to the problem setup: how to combine inferred reward and the given reward, especially when they are in conflict with each other. A discussion of multi-objective RL might be in place."""
340,"""Point Cloud GAN""","['Point Cloud', 'GAN']","""Generative Adversarial Networks (GAN) can achieve promising performance on learning complex data distributions on different types of data. In this paper, we first show that a straightforward extension of an existing GAN algorithm is not applicable to point clouds, because the constraint required for discriminators is undefined for set data. We propose a two fold modification to a GAN algorithm to be able to generate point clouds (PC-GAN). First, we combine ideas from hierarchical Bayesian modeling and implicit generative models by learning a hierarchical and interpretable sampling process. A key component of our method is that we train a posterior inference network for the hidden variables. Second, PC-GAN defines a generic framework that can incorporate many existing GAN algorithms. We further propose a sandwiching objective, which results in a tighter Wasserstein distance estimate than the commonly used dual form in WGAN. We validate our claims on the ModelNet40 benchmark dataset and observe that PC- GAN trained by the sandwiching objective achieves better results on test data than existing methods. We also conduct studies on several tasks, including generalization on unseen point clouds, latent space interpolation, classification, and image to point clouds transformation, to demonstrate the versatility of the proposed PC-GAN algorithm.""","""Reviewers mostly recommended to reject after engaging with the authors, however since not all author answers have been acknowledged by reviewers, I am not sure if there are any remaining issues with the submission. I thus lean to recommend to reject and resubmit. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit. """
341,"""Universal Transformers""","['sequence-to-sequence', 'rnn', 'transformer', 'machine translation', 'language understanding', 'learning to execute']","""Recurrent neural networks (RNNs) sequentially process data by updating their state with each new data point, and have long been the de facto choice for sequence modeling tasks. However, their inherently sequential computation makes them slow to train. Feed-forward and convolutional architectures have recently been shown to achieve superior results on some sequence modeling tasks such as machine translation, with the added advantage that they concurrently process all inputs in the sequence, leading to easy parallelization and faster training times. Despite these successes, however, popular feed-forward sequence models like the Transformer fail to generalize in many simple tasks that recurrent models handle with ease, e.g. copying strings or even simple logical inference when the string or formula lengths exceed those observed at training time. We propose the Universal Transformer (UT), a parallel-in-time self-attentive recurrent sequence model which can be cast as a generalization of the Transformer model and which addresses these issues. UTs combine the parallelizability and global receptive field of feed-forward sequence models like the Transformer with the recurrent inductive bias of RNNs. We also add a dynamic per-position halting mechanism and find that it improves accuracy on several tasks. In contrast to the standard Transformer, under certain assumptions UTs can be shown to be Turing-complete. Our experiments show that UTs outperform standard Transformers on a wide range of algorithmic and language understanding tasks, including the challenging LAMBADA language modeling task where UTs achieve a new state of the art, and machine translation where UTs achieve a 0.9 BLEU improvement over Transformers on the WMT14 En-De dataset.""","""This paper presents Universal Transformers that generalizes Transformers with recurrent connections. The goal of Universal Transformers is to combine the strength of feed-forward convolutional architectures (parallelizability and global receptive fields) with the strength of recurrent neural networks (sequential inductive bias). In addition, the paper investigates a dynamic halting scheme (by adapting Adaptive Computation Time (ACT) of Graves 2016) to allow each individual subsequence to stop recurrent computation dynamically. Pros: The paper presents a new generalized architecture that brings a reasonable novelty over the previous Transformers when combined with the dynamic halting scheme. Empirical results are reasonably comprehensive and the codebase is publicly available. Cons: Unlike RNNs, the network recurs T times over the entire sequence of length M, thus it is not a literal combination of Transformers with RNNs, but only inspired by RNNs. Thus the proposed architecture does not precisely replicate the sequential inductive bias of RNNs. Furthermore, depending on how one views it, the network architecture is not entirely novel in that it is reminiscent of the previous memory network extensions with multi-hop reasoning (--- a point raised by R1 and R2). While several datasets are covered in the empirical study, the selected datasets may be biased toward simpler/easier tasks (--- R1). Verdict: While key ideas might not be entirely novel (R1/R2), the novelty comes from the fact that these ideas have not been combined and experimented in this exact form of Universal Transformers (with optional dynamic halting/ACT), and that the empirical results are reasonably broad and strong, while not entirely impressive (R1). Sufficient novelty and substance overall, and no issues that are dealbreakers. """
342,"""Combining Learned Representations for Combinatorial Optimization""","['Generative Models', 'Restricted Boltzmann Machines', 'Transfer Learning', 'Compositional Learning']","""We propose a new approach to combine Restricted Boltzmann Machines (RBMs) that can be used to solve combinatorial optimization problems. This allows synthesis of larger models from smaller RBMs that have been pretrained, thus effectively bypassing the problem of learning in large RBMs, and creating a system able to model a large, complex multi-modal space. We validate this approach by using learned representations to create ``invertible boolean logic'', where we can use Markov chain Monte Carlo (MCMC) approaches to find the solution to large scale boolean satisfiability problems and show viability towards other combinatorial optimization problems. Using this method, we are able to solve 64 bit addition based problems, as well as factorize 16 bit numbers. We find that these combined representations can provide a more accurate result for the same sample size as compared to a fully trained model. ""","""Dear authors, Thank you for submitting your work to ICLR. The original goal of using smaller models to train a bigger one is definitely interesting and has been the topic of a lot of works. However, the reviewers had two major complaints: the first one is about the clarity of the paper and the second one is about the significance of the tasks on which the algorith is tested. For the latter point, your rebuttal uses arguments which are little known in the ML community and so should be expanded in a future submission."""
343,"""Tinkering with black boxes: counterfactuals uncover modularity in generative models""","['generatice models', 'causality', 'disentangled representations']","""Deep generative models such as Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs) are important tools to capture and investigate the properties of complex empirical data. However, the complexity of their inner elements makes their functionment challenging to assess and modify. In this respect, these architectures behave as black box models. In order to better understand the function of such networks, we analyze their modularity based on the counterfactual manipulation of their internal variables. Our experiments on the generation of human faces with VAEs and GANs support that modularity between activation maps distributed over channels of generator architectures is achieved to some degree, can be used to better understand how these systems operate and allow meaningful transformations of the generated images without further training. erate and edit the content of generated images.""","""This paper explores an interpretation of generative models in terms of interventions on their latent variables. The overall set of ideas seems novel and potentially useful, but the presentation is unclear, the goal of the method seems poorly defined, and the qualitative results (including the videos) are unconvincing. I recommend you put work into factoring the ideas in this paper into smaller ones. For instance, definition 1 is a mess. I would also recommend the use of algorithm boxes."""
344,"""Deep Online Learning Via Meta-Learning: Continual Adaptation for Model-Based RL""","['meta-learning', 'model-based', 'reinforcement learning', 'online learning', 'adaptation']","""Humans and animals can learn complex predictive models that allow them to accurately and reliably reason about real-world phenomena, and they can adapt such models extremely quickly in the face of unexpected changes. Deep neural network models allow us to represent very complex functions, but lack this capacity for rapid online adaptation. The goal in this paper is to develop a method for continual online learning from an incoming stream of data, using deep neural network models. We formulate an online learning procedure that uses stochastic gradient descent to update model parameters, and an expectation maximization algorithm with a Chinese restaurant process prior to develop and maintain a mixture of models to handle non-stationary task distributions. This allows for all models to be adapted as necessary, with new models instantiated for task changes and old models recalled when previously seen tasks are encountered again. Furthermore, we observe that meta-learning can be used to meta-train a model such that this direct online adaptation with SGD is effective, which is otherwise not the case for large function approximators. We apply our method to model-based reinforcement learning, where adapting the predictive model is critical for control; we demonstrate that our online learning via meta-learning algorithm outperforms alternative prior methods, and enables effective continuous adaptation in non-stationary task distributions such as varying terrains, motor failures, and unexpected disturbances.""","""The reviewers appreciated this contribution, particularly its ability to tackle nonstationary domains which are common in real-world tasks. """
345,"""On the loss landscape of a class of deep neural networks with no bad local valleys""","['loss landscape', 'local minima', 'deep neural networks']","""We identify a class of over-parameterized deep neural networks with standard activation functions and cross-entropy loss which provably have no bad local valley, in the sense that from any point in parameter space there exists a continuous path on which the cross-entropy loss is non-increasing and gets arbitrarily close to zero. This implies that these networks have no sub-optimal strict local minima.""","""This paper introduces a class of deep neural nets that provably have no bad local valleys. By constructing a new class of network this paper avoids having to rely on unrealistic assumptions and manages to provide a relatively concise proof that the network family has no strict local minima. Furthermore, it is demonstrated that this type of network yields reasonable experimental results on some benchmarks. The reviewers identified issues such as missing measurements of the training loss, which is the actual quantity studied in the theoretical results, as well as some issues with the presentation of the results. After revisions the reviewers are satisfied that their comments have been addressed. This paper continues an interesting line of theoretical research and brings it closer to practice and so it should be of interest to the ICLR community."""
346,"""End-to-End Hierarchical Text Classification with Label Assignment Policy""","['Hierarchical Classification', 'Text Classification']","""We present an end-to-end reinforcement learning approach to hierarchical text classification where documents are labeled by placing them at the right positions in a given hierarchy. While existing global methods construct hierarchical losses for model training, they either make local decisions at each hierarchy node or ignore the hierarchy structure during inference. To close the gap between training/inference and optimize holistic metrics in an end-to-end manner, we propose to learn a label assignment policy to determine where to place the documents and when to stop. The proposed method, HiLAP, optimizes holistic metrics over the hierarchy, makes inter-dependent decisions during inference, and can be combined with different text encoding models for end-to-end training. Experiments on three public datasets show that HiLAP yields an average improvement of 33.4% in Macro-F1 and 5.0% in Samples-F1, outperforming state-of-the-art methods by a large margin.""","""This paper presents a reinforcement learning approach to hierarchical text classification. Pros: A potentially interesting idea to drive the search process over a hierachical set of labels using reinforcement learning. Cons: The major concensus among all reviewers was that there were various concerns about experimental results, e.g., apple-to-apple comparisons against prior art (R1), proper tuning of hyper-parameters (R1, R2), the label space is too small (539) to have practical significance compared to tens of thousands of labels that have been used in other related work (R3), and other missing baselines (R3). In addition, even after the rebuttal, some of the technical clarity issues have not been fully resolved, e.g., what the proposed method is actually doing (optimizing F1 metric vs the ability to fix inconsistent labeling problem). Verdict: Reject. While authors came back with many detailed responses, they were not enough to address the major concerns reviewers had about the empirical significance of this work."""
347,"""Weakly-supervised Knowledge Graph Alignment with Adversarial Learning""","['Knowledge Graph Alignment', 'Generative Adversarial Network', 'Weakly Supervised']","""Aligning knowledge graphs from different sources or languages, which aims to align both the entity and relation, is critical to a variety of applications such as knowledge graph construction and question answering. Existing methods of knowledge graph alignment usually rely on a large number of aligned knowledge triplets to train effective models. However, these aligned triplets may not be available or are expensive to obtain for many domains. Therefore, in this paper we study how to design fully-unsupervised methods or weakly-supervised methods, i.e., to align knowledge graphs without or with only a few aligned triplets. We propose an unsupervised framework based on adversarial training, which is able to map the entities and relations in a source knowledge graph to those in a target knowledge graph. This framework can be further seamlessly integrated with existing supervised methods, where only a limited number of aligned triplets are utilized as guidance. Experiments on real-world datasets prove the effectiveness of our proposed approach in both the weakly-supervised and unsupervised settings.""","""This paper considers an important problem of aligning two knowledge graphs (the entities and relations therein). The reviewers found the use of adversarial training quite novel and appropriate for the the task, especially as it works in the unsupervised setting as well. The reviewers were also impressed that the proposed work outperforms existing approaches in terms of the accuracy of the alignment. The following potential weaknesses were raised by the reviewers and the AC: (1) Reviewer 3 brings up the fact that the hyperaparameters were set different from the original publications of the baselines, and thus are not convinced of the soundness of the results, (2) Reviewer 2 notes that the evaluation is limited, and more variations should be considered, such as varying the overlap, taking larger subsets of knowledge graphs, and going beyond TranE as the choice for embedding, and (3) Reviewer 3 notes that a simpler baseline based on alignment discrepancy should be considered, which would alleviate the need for RL based training. Although the reviewers raised very different concerns with the paper, none of them were addressed in a response or revision, and thus they agree that the paper should be rejected."""
348,"""Slimmable Neural Networks""","['Slimmable neural networks', 'mobile deep learning', 'accuracy-efficiency trade-offs']","""We present a simple and general method to train a single neural network executable at different widths (number of channels in a layer), permitting instant and adaptive accuracy-efficiency trade-offs at runtime. Instead of training individual networks with different width configurations, we train a shared network with switchable batch normalization. At runtime, the network can adjust its width on the fly according to on-device benchmarks and resource constraints, rather than downloading and offloading different models. Our trained networks, named slimmable neural networks, achieve similar (and in many cases better) ImageNet classification accuracy than individually trained models of MobileNet v1, MobileNet v2, ShuffleNet and ResNet-50 at different widths respectively. We also demonstrate better performance of slimmable models compared with individual ones across a wide range of applications including COCO bounding-box object detection, instance segmentation and person keypoint detection without tuning hyper-parameters. Lastly we visualize and discuss the learned features of slimmable networks. Code and models are available at: pseudo-url""","""This paper proposed a method that creates neural networks that can run under different resource constraints. The reviewers have consensus on accept. The pro is that the paper is novel and provides a practical approach to adjust model for different computation resource, and achieved performance improvement on object detection. One concern from reviewer2 and another public reviewer is the inconsistent performance impact on classification/detection (performance improvement on detection, but performance degradation on classification). Besides, the numbers reported in Table 1 should be confirmed: MobileNet v1 on Google Pixel 1 should have less than 120ms latency [1], not 296 ms. [1] Table 4 of pseudo-url"""
349,"""Learning Implicit Generative Models by Teaching Explicit Ones""",[],"""Implicit generative models are difficult to train as no explicit probability density functions are defined. Generative adversarial nets (GANs) propose a minimax framework to train such models, which suffer from mode collapse in practice due to the nature of the JS-divergence. In contrast, we propose a learning by teaching (LBT) framework to learn implicit models, which intrinsically avoid the mode collapse problem because of using the KL-divergence. In LBT, an auxiliary explicit model is introduced to learn the distribution defined by the implicit model while the later one's goal is to teach the explicit model to match the data distribution. LBT is formulated as a bilevel optimization problem, whose optimum implies that we obtain the maximum likelihood estimation of the implicit model. We adopt an unrolling approach to solve the challenging learning problem. Experimental results demonstrate the effectiveness of our method.""","""The paper proposes a learning by teaching (LBT) framework to train an implicit generative model via an explicit one. It is shown experimentally, that the framework can help to avoid mode collapse. The reviewers commonly raised the question why this is the case, which was answered in the rebuttal by pointing to the differences between the KL- and the JS-divergence and by showing a toy problem for which the JS-divergence has local minima while the KL-divergence has not. However, it still remains unclear why this should be generally and for explicit models with insufficient capacity the case, and if the model will be scalable to larger settings, therefore the paper can not be accepted in the current form."""
350,"""Fast adversarial training for semi-supervised learning""","['Deep learning', 'Semi-supervised learning', 'Adversarial training']","""In semi-supervised learning, Bad GAN approach is one of the most attractive method due to the intuitional simplicity and powerful performances. Bad GAN learns a classifier with bad samples distributed on complement of the support of the input data. But Bad GAN needs additional architectures, a generator and a density estimation model, which involves huge computation and memory consumption cost. VAT is another good semi-supervised learning algorithm, which utilizes unlabeled data to improve the invariance of the classifier with respect to perturbation of inputs. In this study, we propose a new method by combining the ideas of Bad GAN and VAT. The proposed method generates bad samples of high-quality by use of the adversarial training used in VAT. We give theoretical explanations why the adversarial training is good at both generating bad samples and semi-supervised learning. An advantage of the proposed method is to achieve the competitive performances with much fewer computations. We demonstrate advantages our method by various experiments with well known benchmark image datasets.""","""The paper combines the ideas of VAT and Bad GAN, replacing the fake samples in Bad GAN objective with VAT generated samples. The motivation behind using the K+1 SSL framework with VAT examples remains unclear, particularly in the light of Prop. 2 which shows smoothness of classifier around the unlabeled examples is enough (which VAT already encourages). R2 and R3 have raised the point of limited insight and lack of motivation behind combining VAT and Bad GAN objectives in this way. R2 and R3 are also concerned about the empirical results which show only marginal improvements over VAT/BadGAN in most settings. AC feels that the idea of the paper is interesting but agrees with R2/R3 that the proposed objective is not motivated well enough (what is the precise advantage of using K+1 SSL formulation with VAT examples?). The paper really falls on the borderline and could be improved if this point is addressed convincingly. """
351,"""Training generative latent models by variational f-divergence minimization""","['variational inference', 'generative model', 'f divergence']","""Probabilistic models are often trained by maximum likelihood, which corresponds to minimizing a specific form of f-divergence between the model and data distribution. We derive an upper bound that holds for all f-divergences, showing the intuitive result that the divergence between two joint distributions is at least as great as the divergence between their corresponding marginals. Additionally, the f-divergence is not formally defined when two distributions have different supports. We thus propose a noisy version of f-divergence which is well defined in such situations. We demonstrate how the bound and the new version of f-divergence can be readily used to train complex probabilistic generative models of data and that the fitted model can depend significantly on the particular divergence used.""","""The paper proposes a new method for training generative models by minimizing general f-divergences. The main technical idea is to optimize f-divergence between joint distributions which is rightly observed to be the upper bound of the f-divergence between the marginal distributions and address the disjoint support problem by convolving the data with a noise distribution. The basic ideas in this work are not completely novel but are put together in a new way. However, the key weakness of this work, as all the reviewer noticed, is that the empirical results are too week to support the usefulness of the proposed approach. The only quantitive results are in table 2, which is only a simple Gaussian example. It essential to have more substantial empirical results for supporting the new algorithm. """
352,"""Revealing interpretable object representations from human behavior""","['category representation', 'sparse coding', 'representation learning', 'interpretable representations']","""To study how mental object representations are related to behavior, we estimated sparse, non-negative representations of objects using human behavioral judgments on images representative of 1,854 object categories. These representations predicted a latent similarity structure between objects, which captured most of the explainable variance in human behavioral judgments. Individual dimensions in the low-dimensional embedding were found to be highly reproducible and interpretable as conveying degrees of taxonomic membership, functionality, and perceptual attributes. We further demonstrated the predictive power of the embeddings for explaining other forms of human behavior, including categorization, typicality judgments, and feature ratings, suggesting that the dimensions reflect human conceptual representations of objects beyond the specific task.""","""The reviewers viewed the work favorably, with only one reviewer providing a score slightly below acceptance. The authors thoroughly addressed the reviewer's original concerns, and they adjusted their score upwards afterwards. The low-rating reviewer remains skeptical of the significance of the work, but the other two reviewers make firm cases for the appeal of the work to the ICLR audience. In follow-up discussion after the author's responses were submitted and discussed, the low-rating reviewer did not make a clear case for rejecting the paper, and further, the higher-rating reviewers' arguments for the impact of the paper were convincing. Therefore, I recommend accepting this paper."""
353,"""Convergent Reinforcement Learning with Function Approximation: A Bilevel Optimization Perspective""","['reinforcement learning', 'Deep Q-networks', 'actor-critic algorithm', 'ODE approximation']",""" We study reinforcement learning algorithms with nonlinear function approximation in the online setting. By formulating both the problems of value function estimation and policy learning as bilevel optimization problems, we propose online Q-learning and actor-critic algorithms for these two problems respectively. Our algorithms are gradient-based methods and thus are computationally efficient. Moreover, by approximating the iterates using differential equations, we establish convergence guarantees for the proposed algorithms. Thorough numerical experiments are conducted to back up our theory.""","""The paper gives an bilevel optimization view for several standard RL algorithms, and proves their asymptotic convergence with function approximation under some assumptions. The analysis is a two-time scale one, and some empirical study is included. It's a difficult decision to make for this paper. It clearly has a few things to be liked: (1) the bilevel view seems new in the RL literature (although the view has been implicitly used throughout the literature); (2) the paper is solid and gives rigorous, nontrivial analyses. On the other hand, reviewers are not convinced it's ready for publication in its current stage: (1) Technical novelty, in the context of published works: extra challenges needed on top of Borkar; similarity to and differences from Dai et al.; ... (2) The practical significance is somewhat limited. Does the analysis provide additional insight into how to improve existing approaches? How restricted are the assumptions? Are the online-vs-batch distinction from Dai et al. really important in practice? (3) What does the paper want to show in the experiments, since no new algorithms are developed? Some claims are made based on very limited empirical evidence. It'd be much better to run algorithms on more controlled situations to show, say, the significance of two timescale updates. Also, as those algorithms are classic Q-learning and actor-critic (quote the authors in responses), how well do the algorithms solve the well-known divergent examples when function approximation is used? (4) Presentation needs to be improved. Reviewers pointed out some over claims and imprecise statements. While the author responses were helpful in clarifying some of the questions, reviewers felt that the remaining questions needed to be addressed and the changes would be large enough that another full review cycle is needed."""
354,"""Dirichlet Variational Autoencoder""","['Variational autoencoder', 'Unsupervised learning', '(Semi-)Supervised learning', 'Topic modeling']","""This paper proposes Dirichlet Variational Autoencoder (DirVAE) using a Dirichlet prior for a continuous latent variable that exhibits the characteristic of the categorical probabilities. To infer the parameters of DirVAE, we utilize the stochastic gradient method by approximating the Gamma distribution, which is a component of the Dirichlet distribution, with the inverse Gamma CDF approximation. Additionally, we reshape the component collapsing issue by investigating two problem sources, which are decoder weight collapsing and latent value collapsing, and we show that DirVAE has no component collapsing; while Gaussian VAE exhibits the decoder weight collapsing and Stick-Breaking VAE shows the latent value collapsing. The experimental results show that 1) DirVAE models the latent representation result with the best log-likelihood compared to the baselines; and 2) DirVAE produces more interpretable latent values with no collapsing issues which the baseline models suffer from. Also, we show that the learned latent representation from the DirVAE achieves the best classification accuracy in the semi-supervised and the supervised classification tasks on MNIST, OMNIGLOT, and SVHN compared to the baseline VAEs. Finally, we demonstrated that the DirVAE augmented topic models show better performances in most cases.""","""This paper applies Dirichlet distribution to the latent variables of a VAE in order to address the component collapsing issues for categorical probabilities. The method is clearly presented, and extensive experiments are carried out to prove the advantage against VAEs with other prior distributions. The main concern of the paper is the limited novelty. The main methodology contribution of this paper is to combine the decomposition a Dirichlet distribution as Gamma distributions, and approximating Gamma component with inverse Gamma CDF, but both components are common practices. R3 also points out that the paper is distracted by two different messages the authors try to convey. The presentation and experiments are not designed to provide a cohesive message. The concern is not solved in the authors' feedback. Based on the current reviews, this paper does not meet the standard for ICLR publication. Despite the limited novelty in the proposed model, if the paper could be revised to show that a simple modification is good for solve one problem with general applications, it would make a good publication in a future venue."""
355,"""Stochastic Gradient/Mirror Descent: Minimax Optimality and Implicit Regularization""","['optimization', 'stochastic gradient descent', 'mirror descent', 'implicit regularization', 'deep learning theory']","""Stochastic descent methods (of the gradient and mirror varieties) have become increasingly popular in optimization. In fact, it is now widely recognized that the success of deep learning is not only due to the special deep architecture of the models, but also due to the behavior of the stochastic descent methods used, which play a key role in reaching ""good"" solutions that generalize well to unseen data. In an attempt to shed some light on why this is the case, we revisit some minimax properties of stochastic gradient descent (SGD) for the square loss of linear models---originally developed in the 1990's---and extend them to \emph{general} stochastic mirror descent (SMD) algorithms for \emph{general} loss functions and \emph{nonlinear} models. In particular, we show that there is a fundamental identity which holds for SMD (and SGD) under very general conditions, and which implies the minimax optimality of SMD (and SGD) for sufficiently small step size, and for a general class of loss functions and general nonlinear models. We further show that this identity can be used to naturally establish other properties of SMD (and SGD), namely convergence and \emph{implicit regularization} for over-parameterized linear models (in what is now being called the ""interpolating regime""), some of which have been shown in certain cases in prior literature. We also argue how this identity can be used in the so-called ""highly over-parameterized"" nonlinear setting (where the number of parameters far exceeds the number of data points) to provide insights into why SMD (and SGD) may have similar convergence and implicit regularization properties for deep learning. ""","""The authors give a characterization of stochastic mirror descent (SMD) as a conservation law (17) in terms of the Bregman divergence of the loss. The identity allows the authors to show that SMD converges to the optimal solution of a particular minimax filtering problem. In the special overparametrized linear case, when SMD is simply SGD, the result recovers a recent theorem due to Gunasekar et al. (2018). The consequences for the overparametrized nonlinear case are more speculative. The main criticisms are around impact, however, I'm inclined to think that any new insight on this problem, especially one that imports results from other areas like control, are useful to incorporate into the literature. I will comment that the discussion of previous work is wholly inadequate. The authors essentially do not engage with previous work, and mostly make throwaway citations. This is a real pity. I would be nice to see better scholarship."""
356,"""Probabilistic Semantic Embedding""",[],"""We present an extension of a variational auto-encoder that creates semantically richcoupled probabilistic latent representations that capture the semantics of multiplemodalities of data. We demonstrate this model through experiments using imagesand textual descriptors as inputs and images as outputs. Our latent representationsare not only capable of driving a decoder to generate novel data, but can also be useddirectly for annotation or classification. Using the MNIST and Fashion-MNISTdatasets we show that the embedding not only provides better reconstruction andclassification performance than the current state-of-the-art, but it also allows us toexploit the semantic content of the pretrained word embedding spaces to do taskssuch as image generation from labels outside of those seen during training.""","""mnist and small picture variants are not that impressive. it is a minor extension of VAEs which also are not common in sota systems."""
357,"""Pyramid Recurrent Neural Networks for Multi-Scale Change-Point Detection""","['changepoint detection', 'multivariate time series data', 'multiscale RNN']","""Many real-world time series, such as in activity recognition, finance, or climate science, have changepoints where the system's structure or parameters change. Detecting changes is important as they may indicate critical events. However, existing methods for changepoint detection face challenges when (1) the patterns of change cannot be modeled using simple and predefined metrics, and (2) changes can occur gradually, at multiple time-scales. To address this, we show how changepoint detection can be treated as a supervised learning problem, and propose a new deep neural network architecture that can efficiently identify both abrupt and gradual changes at multiple scales. Our proposed method, pyramid recurrent neural network (PRNN), is designed to be scale-invariant, by incorporating wavelets and pyramid analysis techniques from multi-scale signal processing. Through experiments on synthetic and real-world datasets, we show that PRNN can detect abrupt and gradual changes with higher accuracy than the state of the art and can extrapolate to detect changepoints at novel timescales that have not been seen in training.""","""This paper studies change-point detection in time series using a multiscale neural network architecture which contains recurrent connections across different time scales. Reviewers were mixed in this submission. They found the paper generally clear and well-written, and the idea of adding a multiscale component to the model interesting. However, they also pointed out weaknesses in the related work section and found the experimental setup somewhat limited. In particular, the paper provides little to no analysis of the learnt features. Taking these assessments into consideration, the AC concludes this submission cannot be accepted at this time. """
358,"""DL2: Training and Querying Neural Networks with Logic""","['neural networks', 'training with constraints', 'querying networks', 'semantic training']","""We present DL2, a system for training and querying neural networks with logical constraints. The key idea is to translate these constraints into a differentiable loss with desirable mathematical properties and to then either train with this loss in an iterative manner or to use the loss for querying the network for inputs subject to the constraints. We empirically demonstrate that DL2 is effective in both training and querying scenarios, across a range of constraints and data sets.""","""Unfortunately, this paper fell just below the bar for acceptance. The reviewers all saw significant promise in this work, stating that it is intriguing, ""novel and provides an interesting solution to a challenging problem"" and that ""many interesting use cases are clear"". AnonReviewer2 particularly argued for acceptance, arguing that the proposed approach provides a very flexible method for incorporating constraints in neural network training. A concern of AnonReviewer2 was that there was no guarantee that this loss would be convex or converge to an optimum while statisfying the constraints. The other two reviewers unfortunately felt that while the proposed approach was ""interesting"", ""promising"" and ""intriguing"", the quality of the paper, in terms of exposition, was too low to justify acceptance. Arguably, it seems the writing doesn't do the idea justice in this case and the paper would ultimately be significantly more impactful if it was carefully rewritten. """
359,"""Quality Evaluation of GANs Using Cross Local Intrinsic Dimensionality""","['Generative Adversarial Networks', 'Evaluation Metric', 'Local Intrinsic Dimensionality']","""Generative Adversarial Networks (GANs) are an elegant mechanism for data generation. However, a key challenge when using GANs is how to best measure their ability to generate realistic data. In this paper, we demonstrate that an intrinsic dimensional characterization of the data space learned by a GAN model leads to an effective evaluation metric for GAN quality. In particular, we propose a new evaluation measure, CrossLID, that assesses the local intrinsic dimensionality (LID) of input data with respect to neighborhoods within GAN-generated samples. In experiments on 3 benchmark image datasets, we compare our proposed measure to several state-of-the-art evaluation metrics. Our experiments show that CrossLID is strongly correlated with sample quality, is sensitive to mode collapse, is robust to small-scale noise and image transformations, and can be applied in a model-free manner. Furthermore, we show how CrossLID can be used within the GAN training process to improve generation quality.""","""The paper propose a new metric for the evaluation of generative models, which they call CrossLID and which assesses the local intrinsic dimensionality (LID) of input data with respect to neighborhoods within generated samples, i.e. which is based on nearest neighbor distances between samples from the real data distribution and the generator. The paper is clearly written and provides an extensive experimental analysis, that shows that LID is an interesting metric to use in addition to exciting metrics as FID, at least for the case of not to complex image distributions The paper would be streghten by showing that the metric can also be applied in those more complex settings. """
360,"""code2seq: Generating Sequences from Structured Representations of Code""","['source code', 'programs', 'code2seq']","""The ability to generate natural language sequences from source code snippets has a variety of applications such as code summarization, documentation, and retrieval. Sequence-to-sequence (seq2seq) models, adopted from neural machine translation (NMT), have achieved state-of-the-art performance on these tasks by treating source code as a sequence of tokens. We present code2seq: an alternative approach that leverages the syntactic structure of programming languages to better encode source code. Our model represents a code snippet as the set of compositional paths in its abstract syntax tree (AST) and uses attention to select the relevant paths while decoding. We demonstrate the effectiveness of our approach for two tasks, two programming languages, and four datasets of up to 16M examples. Our model significantly outperforms previous models that were specifically designed for programming languages, as well as general state-of-the-art NMT models. An interactive online demo of our model is available at pseudo-url. Our code, data and trained models are available at pseudo-url.""","""Overall this paper presents a few improvements over the code2vec model of Alon et al., applying it to seq2seq tasks. The empirical results are very good, and there is fairly extensive experimentation. This is a relatively crowded space, so there are a few natural baselines that were not compared to, but I don't think that comparison to every single baseline is warranted or necessary, and the authors have done an admirable job. One thing that still is quite puzzling is the strength of the ""AST nodes only baseline"", which the authors have given a few explanations for (using nodes helps focus on variables, and also there is an effect of combining together things that are close together in the AST tree). Still, this result doesn't seem to mesh with the overall story of the paper all that well, and again opens up some obvious questions such as whether a Transformer model trained on only AST nodes would have done similarly, and if not why not. This paper is very much on the borderline, so if there is space in the conference I think it would be a reasonable addition, but there could also be an argument made that the paper would be stronger in a re-submission where the above questions are answered."""
361,"""Learning protein sequence embeddings using information from structure""","['sequence embedding', 'sequence alignment', 'RNN', 'LSTM', 'protein structure', 'amino acid sequence', 'contextual embeddings', 'transmembrane prediction']","""Inferring the structural properties of a protein from its amino acid sequence is a challenging yet important problem in biology. Structures are not known for the vast majority of protein sequences, but structure is critical for understanding function. Existing approaches for detecting structural similarity between proteins from sequence are unable to recognize and exploit structural patterns when sequences have diverged too far, limiting our ability to transfer knowledge between structurally related proteins. We newly approach this problem through the lens of representation learning. We introduce a framework that maps any protein sequence to a sequence of vector embeddings --- one per amino acid position --- that encode structural information. We train bidirectional long short-term memory (LSTM) models on protein sequences with a two-part feedback mechanism that incorporates information from (i) global structural similarity between proteins and (ii) pairwise residue contact maps for individual proteins. To enable learning from structural similarity information, we define a novel similarity measure between arbitrary-length sequences of vector embeddings based on a soft symmetric alignment (SSA) between them. Our method is able to learn useful position-specific embeddings despite lacking direct observations of position-level correspondence between sequences. We show empirically that our multi-task framework outperforms other sequence-based methods and even a top-performing structure-based alignment method when predicting structural similarity, our goal. Finally, we demonstrate that our learned embeddings can be transferred to other protein sequence problems, improving the state-of-the-art in transmembrane domain prediction.""","""The reviewers and authors had a productive conversation, leading to an improvement in the paper quality. The strengths of the paper highlighted by reviewers are a novel learning set-up and new loss functions that seem to help in the task of protein contact prediction and protein structural similarity prediction. The reviewers characterize the work as constituting an advance in an exciting application space, as well as containing a new configuration of methods to address the problem. Overall, it is clear the paper should be accepted, based on reviewer comments, which unanimously agreed on the quality of the work."""
362,"""DecayNet: A Study on the Cell States of Long Short Term Memories""","['Long short term memory', 'Recurrent neural network', 'Dynamical systems', 'Difference equation']","""It is unclear whether the extensively applied long-short term memory (LSTM) is an optimised architecture for recurrent neural networks. Its complicated design makes the network hard to analyse and non-immediately clear for its utilities in real-world data. This paper studies LSTMs as systems of difference equations, and takes a theoretical mathematical approach to study consecutive transitions in network variables. Our study shows that the cell state propagation is predominantly controlled by the forget gate. Hence, we introduce DecayNets, LSTMs with monotonically decreasing forget gates, to calibrate cell state dynamics. With recurrent batch normalisation, DecayNet outperforms the previous state of the art for permuted sequential MNIST. The Decay mechanism is also beneficial for LSTM-based optimisers, and decrease optimisee neural network losses more rapidly. Edit status: Revised paper.""","""there is a disagreement among the reviewers, and i am siding with the two reviewers (r1 and r3) and agree with r3 that it is rather unconventional to pick learning-to-learn to experiment with modelling variable-length sequences (it's not like there's no other task that has this characteristics, e.g., language modelling, translation, ...) """
363,"""Exploration by random network distillation""","['reinforcement learning', 'exploration', 'curiosity']","""We introduce an exploration bonus for deep reinforcement learning methods that is easy to implement and adds minimal overhead to the computation performed. The bonus is the error of a neural network predicting features of the observations given by a fixed randomly initialized neural network. We also introduce a method to flexibly combine intrinsic and extrinsic rewards. We find that the random network distillation (RND) bonus combined with this increased flexibility enables significant progress on several hard exploration Atari games. In particular we establish state of the art performance on Montezuma's Revenge, a game famously difficult for deep reinforcement learning methods. To the best of our knowledge, this is the first method that achieves better than average human performance on this game without using demonstrations or having access the underlying state of the game, and occasionally completes the first level. This suggests that relatively simple methods that scale well can be sufficient to tackle challenging exploration problems.""","""Pros: - novel, general idea for hard exploration domains - multiple additional tricks - ablations, control experiments - well-written paper - excellent results on Montezuma Cons: - low sample efficiency (2B+ frames) - unresolved questions (non-episodic intrinsic rewards) - could have done better apples-to-apples comparisons to baselines The reviewers did not reach consensus on whether to accept or reject the paper. In particular, after multiple rounds of discussion, reviewer 1 remains adamant that the downsides of the paper outweigh its good points. However, given that the other three reviewers argue strongly and credibly for acceptance, I think the paper should be accepted."""
364,"""Reducing Overconfident Errors outside the Known Distribution""","['Machine learning safety', 'confidence', 'overconfidence', 'unknown domain', 'novel distribution', 'generalization', 'distillation', 'ensemble', 'underrepresentation']","""Intuitively, unfamiliarity should lead to lack of confidence. In reality, current algorithms often make highly confident yet wrong predictions when faced with unexpected test samples from an unknown distribution different from training. Unlike domain adaptation methods, we cannot gather an ""unexpected dataset"" prior to test, and unlike novelty detection methods, a best-effort original task prediction is still expected. We compare a number of methods from related fields such as calibration and epistemic uncertainty modeling, as well as two proposed methods that reduce overconfident errors of samples from an unknown novel distribution without drastically increasing evaluation time: (1) G-distillation, training an ensemble of classifiers and then distill into a single model using both labeled and unlabeled examples, or (2) NCR, reducing prediction confidence based on its novelty detection score. Experimentally, we investigate the overconfidence problem and evaluate our solution by creating ""familiar"" and ""novel"" test splits, where ""familiar"" are identically distributed with training and ""novel"" are not. We discover that calibrating using temperature scaling on familiar data is the best single-model method for improving novel confidence, followed by our proposed methods. In addition, some methods' NLL performance are roughly equivalent to a regularly trained model with certain degree of smoothing. Calibrating can also reduce confident errors, for example, in gender recognition by 95% on demographic groups different from the training data.""","""The paper proposes methods to deal with estimating classification confidence on unseen data distributions. The reviewers and AC note the following potential weaknesses: (1) limited novelty and (2) the authors' new comparison with Guo et al. (2017) asked by Reviewer 2 is not convincing enough. AC thinks the proposed method has potential and is interesting, but decided that the authors need new ideas to meet the high standard of ICLR."""
365,"""Discovering General-Purpose Active Learning Strategies""","['active learning', 'meta learning', 'reinforcement learning']","""We propose a general-purpose approach to discovering active learning (AL) strategies from data. These strategies are transferable from one domain to another and can be used in conjunction with many machine learning models. To this end, we formalize the annotation process as a Markov decision process, design universal state and action spaces and introduce a new reward function that precisely reflects the AL objective of minimizing the annotation cost We seek to find an optimal (non-myopic) AL strategy using reinforcement learning. We evaluate the learned strategies on multiple unrelated domains and show that they consistently outperform state-of-the-art baselines.""","""This paper provides further insight into using RL for active learning, particularly by formulating AL as an MDP and then using RL methods for that MDP. Though the paper has a few insights, it does not sufficiently place itself amongst the many other similar strategies using an MDP formulation. I recommend better highlighting what is novel in this work (e.g., more focus on the reward function, if that is key). Additionally, avoid general statements like To this end, we formalize the annotation process as a Markov decision process, which suggests that this is part of the contribution, but as highlighted by reviewers, has been a standard approach. """
366,"""SPIGAN: Privileged Adversarial Learning from Simulation""","['domain adaptation', 'GAN', 'semantic segmentation', 'simulation', 'privileged information']","""Deep Learning for Computer Vision depends mainly on the source of supervision. Photo-realistic simulators can generate large-scale automatically labeled synthetic data, but introduce a domain gap negatively impacting performance. We propose a new unsupervised domain adaptation algorithm, called SPIGAN, relying on Simulator Privileged Information (PI) and Generative Adversarial Networks (GAN). We use internal data from the simulator as PI during the training of a target task network. We experimentally evaluate our approach on semantic segmentation. We train the networks on real-world Cityscapes and Vistas datasets, using only unlabeled real-world images and synthetic labeled data with z-buffer (depth) PI from the SYNTHIA dataset. Our method improves over no adaptation and state-of-the-art unsupervised domain adaptation techniques.""","""The paper proposes an unsupervised domain adaptation solution applied for semantic segmentation from simulated to real world driving scenes. The main contribution consists of introducing an auxiliary loss based on depth information from the simulator. All reviewers agree that the solution offers a new idea and contribution to the adaptation literature. The ablations provided effectively address the concern that the privileged information does in fact aid in transfer. The additional ablation on the perceptual loss done during rebuttal is also valuable and should be included in the final version. The work would benefit from application of the method across other sim2real dataset tasks so as to be compared to the recent approaches mentioned by the reviewers, but the current evaluation is sufficient to demonstrate the effectiveness of the approach over baseline solutions. """
367,"""Interpretable Continual Learning""","['Interpretability', 'Continual Learning']","""We present a framework for interpretable continual learning (ICL). We show that explanations of previously performed tasks can be used to improve performance on future tasks. ICL generates a good explanation of a finished task, then uses this to focus attention on what is important when facing a new task. The ICL idea is general and may be applied to many continual learning approaches. Here we focus on the variational continual learning framework to take advantage of its flexibility and efficacy in overcoming catastrophic forgetting. We use saliency maps to provide explanations of performed tasks and propose a new metric to assess their quality. Experiments show that ICL achieves state-of-the-art results in terms of overall continual learning performance as measured by average classification accuracy, and also in terms of its explanations, which are assessed qualitatively and quantitatively using the proposed metric.""","""The presented method proposes to use saliency maps as a component for an additional metric of forgetting in continual learning, and as a tool as additional information to improve learning on new tasks. Pros: + R2 & R3: Clearly written and easy to follow. + R3: New metric to compare saliency masks + R3: Interesting idea to utilize previously learned saliency masks to augment learning new tasks. + R1: Performance improvements observed. Cons: - R1 & R2: Novelty is limited in the context of prior works in this field. Unanswered by authors. - R2: Concerns around method's ability to use salient but disconnected components. Unanswered by authors. - R2: Experiments needed on more realistic datasets, such as ImageNet. Unanswered by authors. - R3: Performance gains are small. - R1 & R2: Literature review is insufficient. Reviewers are leaning reject, and R2's concerns have not been answered by the authors at all. Idea seems interesting, authors are encouraged to take into careful consideration the feedback from authors and continue their research."""
368,"""LARGE BATCH SIZE TRAINING OF NEURAL NETWORKS WITH ADVERSARIAL TRAINING AND SECOND-ORDER INFORMATION""","['adversarial training', 'large batch size', 'neural network']","""Stochastic Gradient Descent (SGD) methods using randomly selected batches are widely-used to train neural network (NN) models. Performing design exploration to find the best NN for a particular task often requires extensive training with different models on a large dataset, which is very computationally expensive. The most straightforward method to accelerate this computation is to distribute the batch of SGD over multiple processors. However, large batch training often times leads to degradation in accuracy, poor generalization, and even poor robustness to adversarial attacks. Existing solutions for large batch training either do not work or require massive hyper-parameter tuning. To address this issue, we propose a novel large batch training method which combines recent results in adversarial training (to regularize against ``sharp minima'') and second order optimization (to use curvature information to change batch size adaptively during training). We extensively evaluate our method on Cifar-10/100, SVHN, TinyImageNet, and ImageNet datasets, using multiple NNs, including residual networks as well as compressed networks such as SqueezeNext. Our new approach exceeds the performance of the existing solutions in terms of both accuracy and the number of SGD iterations (up to 1\% and pseudo-formula , respectively). We emphasize that this is achieved without any additional hyper-parameter tuning to tailor our method to any of these experiments. ""","""I would like to commend the authors on their work engaging with the reviewers and for working to improve training time. However, there is not enough support among the reviewers to accept this submission. The reviewers raised several important points about the paper, but I believe there are a few other issues not adequately highlighted in the reviews that prevent this work from being accepted: 1. [premises] It has not been adequately established that ""large batch training often times leads to degradation in accuracy"" inherently which is an important premise of this work. Reports from the literature can largely be explained by other things in the experimental protocol. Even the framing of this issue has become confused since, although it may be possible to achieve the same accuracy at any batch size with careful tuning, this might require using (at worst) the same number of steps as the smaller batch size in some cases and thus result in little to no speedup. For example see pseudo-url and recent work in pseudo-url for more information. Even Keskar et al. reported that data augmentation eliminated the solution quality difference between their larger batch size and their smaller batch size experiments which indicates that even if noisiness from small batches serving to regularize training other regularization techniques can serve just as well. 2. [baseline strength] The appropriate baseline is standard minibatch SGD w/momentum (or ADAM or whatever) algorithm with extremely careful tuning of *all* of the hyperparameters. None of the popular learning rate heuristics will always work and other optimization parameters need to be tuned as well. If learning rate decay is used, it should also be tuned especially if one is trying to measure a speedup. The submission does not provide a sufficiently convincing baseline. 3. [measurement protocol] The protocol for measuring a speedup is not convincing without more information on how the baselines were tuned to achieve the same accuracy in the fewest steps. Approximating the protocols in pseudo-url would be one alternative. Additionally there are a variety of framing of issues around hyperparameter tuning, but, because they are easier to fix, they are not as salient for the decision. """
369,"""Overfitting Detection of Deep Neural Networks without a Hold Out Set""","['deep learning', 'overfitting', 'generalization', 'memorization']","""Overfitting is an ubiquitous problem in neural network training and usually mitigated using a holdout data set. Here we challenge this rationale and investigate criteria for overfitting without using a holdout data set. Specifically, we train a model for a fixed number of epochs multiple times with varying fractions of randomized labels and for a range of regularization strengths. A properly trained model should not be able to attain an accuracy greater than the fraction of properly labeled data points. Otherwise the model overfits. We introduce two criteria for detecting overfitting and one to detect underfitting. We analyze early stopping, the regularization factor, and network depth. In safety critical applications we are interested in models and parameter settings which perform well and are not likely to overfit. The methods of this paper allow characterizing and identifying such models.""","""The reviewers reached a consensus that the paper is not fit for publication for the moment because a) the paper lacks thorough experiments and b) the criteria provided by the paper are relatively evague (see more details in reviewer 3's comments."""
370,"""Search-Guided, Lightly-supervised Training of Structured Prediction Energy Networks""","['structured prediction energy networks', 'indirect supervision', 'search-guided training', 'reward functions']",""" In structured output prediction tasks, labeling ground-truth training output is often expensive. However, for many tasks, even when the true output is unknown, we can evaluate predictions using a scalar reward function, which may be easily assembled from human knowledge or non-differentiable pipelines. But searching through the entire output space to find the best output with respect to this reward function is typically intractable. In this paper, we instead use efficient truncated randomized search in this reward function to train structured prediction energy networks (SPENs), which provide efficient test-time inference using gradient-based search on a smooth, learned representation of the score landscape, and have previously yielded state-of-the-art results in structured prediction. In particular, this truncated randomized search in the reward function yields previously unknown local improvements, providing effective supervision to SPENs, avoiding their traditional need for labeled training data. ""","""This paper proposes search-guided training for structured prediction energy networks (SPENs). The reviewers found some interest in this approach, though were somewhat underwhelmed by the experimental comparison and the details provided about the method. R1 was positive and recommends acceptance; R2 and R3 thought the paper was on the incremental side and recommend rejection. Given the space restriction to this year's conference, we have to reject some borderline papers. The AC thus recommends the authors to take the reviewers comments in consideration for a ""revise and resubmit""."""
371,"""Correction Networks: Meta-Learning for Zero-Shot Learning""","['zero-shot learning', 'image classification', 'fine-grained classification', 'meta-learning']","""We propose a model that learns to perform zero-shot classification using a meta-learner that is trained to produce a correction to the output of a previously trained learner. The model consists of two modules: a task module that supplies an initial prediction, and a correction module that updates the initial prediction. The task module is the learner and the correction module is the meta-learner. The correction module is trained in an episodic approach whereby many different task modules are trained on various subsets of the total training data, with the rest being used as unseen data for the correction module. The correction module takes as input a representation of the task module's training data so that the predicted correction is a function of the task module's training data. The correction module is trained to update the task module's prediction to be closer to the target value. This approach leads to state-of-the-art performance for zero-shot classification on natural language class descriptions on the CUB and NAB datasets. ""","""This is a difficult decision, as the reviewers are quite polarized on this paper, and did not come to a consensus through discussion. The positive elements of the paper are that the method itself is a novel and interesting approach, and that the performance is clearly state of the art. While impressive, the fact that a relatively simple task module trained on the features from Zhu et al. can match the performance of GAZSL suggests that it is difficult to compare these methods in an apples-to-apples way without using consistent features. There are two ways to deal with this: train the baseline methods using the features of Zhu, or train correction networks using less powerful features from other baselines. Reviewer 3 pointed this out, and asked for such a comparison. The defense given by the authors is that they use the same features as the current SOTA baselines, and therefore their comparison is sound. I agree to an extent, however it should be relatively simple to either elevate other baselines, or compare correction networks with different features. Otherwise, most of the rows in Table 1 should be ignored. Running correction networks in different features in an ablation study would also demonstrate that the gains are consistent. I think the authors should run these experiments, and if the results hold then there will be no doubt in my mind that this will be a worthy contribution. However, in their absence, I cant say with certainty how effective the proposed method really is. """
372,"""DADAM: A consensus-based distributed adaptive gradient method for online optimization""",[],"""Online and stochastic optimization methods such as SGD, ADAGRAD and ADAM are key algorithms in solving large-scale machine learning problems including deep learning. A number of schemes that are based on communications of nodes with a central server have been recently proposed in the literature to parallelize them. A bottleneck of such centralized algorithms lies on the high communication cost incurred by the central node. In this paper, we present a new consensus-based distributed adaptive moment estimation method (DADAM) for online optimization over a decentralized network that enables data parallelization, as well as decentralized computation. Such a framework note only can be extremely useful for learning agents with access to only local data in a communication constrained environment, but as shown in this work also outperform centralized adaptive algorithms such as ADAM for certain realistic classes of loss functions. We analyze the convergence properties of the proposed algorithm and provide a \textit{dynamic regret} bound on the convergence rate of adaptive moment estimation methods in both stochastic and deterministic settings. Empirical results demonstrate that DADAM works well in practice and compares favorably to competing online optimization methods.""","""The paper provides a distributed optimization method, applicable to decentralized computation while retaining provable guarantees. This was a borderline paper and a difficult decision. The proposed algorithm is straightforward (a compliment), showing how adaptive optimization algorithms can still be coordinated in a distributed fashion. The theoretical analysis is interesting, but additional assumptions about the mixing are needed to reach clear conclusions: for example, additional assumptions are required to demonstrate potential advantages over non-distributed adaptive optimization algorithms. The initial version of the paper was unfortunately sloppy, with numerous typographical errors. More importantly, some key relevant literature was not cited: - Duchi, John C., Alekh Agarwal, and Martin J. Wainwright. ""Dual averaging for distributed optimization: Convergence analysis and network scaling."" IEEE Transactions on Automatic control 57.3 (2012): 592-606. In addition to citing this work, this and the other related works need to be discussed in relation to the proposed approach earlier in the paper, as suggested by Reviewer 3. There was disagreement between the reviewers in the assessment of this paper. Generally the dissenting reviewer produced the highest quality assessment. This paper is on the borderline, however given the criticisms raised it would benefit from additional theoretical strengthening, improved experimental reporting, and better framing with respect to the existing literature."""
373,"""Ada-Boundary: Accelerating the DNN Training via Adaptive Boundary Batch Selection""","['acceleration', 'batch selection', 'convergence', 'decision boundary']","""Neural networks can converge faster with help from a smarter batch selection strategy. In this regard, we propose Ada-Boundary, a novel adaptive-batch selection algorithm that constructs an effective mini-batch according to the learning progress of the model.Our key idea is to present confusing samples what the true label is. Thus, the samples near the current decision boundary are considered as the most effective to expedite convergence. Taking advantage of our design, Ada-Boundary maintains its dominance in various degrees of training difficulty. We demonstrate the advantage of Ada-Boundary by extensive experiments using two convolutional neural networks for three benchmark data sets. The experiment results show that Ada-Boundary improves the training time by up to 31.7% compared with the state-of-the-art strategy and by up to 33.5% compared with the baseline strategy.""","""This paper introduced an adaptive importance sampling strategy to select mini-batches to speed up the convergence of network training. The method is well motivated and easy to follow. The main concerns raised by the reviewers are limited novelty of the proposed simple idea compared to related recent work, and moderate empirical performance. The authors argue that the particular choice of the adaptive sampling method comes after trying various methods. I believe providing more detailed discussion and comparison with different methods together with the ""active bias"" paper would help the readers appreciate the insights conveyed in this paper. The authors provide some additional experiments in the revision. It would make the whole experiment section a lot stronger and convincing if the authors could run more thorough experiments on extra challenging datasets and include all the results int the main text. Additional experiment to clarify the merit of the proposed method on either faster convergence or lower asymptotic error would also improve the contribution of this paper."""
374,"""Targeted Adversarial Examples for Black Box Audio Systems""","['adversarial attack', 'adversarial examples', 'audio processing', 'speech to text', 'deep learning', 'adversarial audio', 'black box', 'machine learning']","""The application of deep recurrent networks to audio transcription has led to impressive gains in automatic speech recognition (ASR) systems. Many have demonstrated that small adversarial perturbations can fool deep neural networks into incorrectly predicting a specified target with high confidence. Current work on fooling ASR systems have focused on white-box attacks, in which the model architecture and parameters are known. In this paper, we adopt a black-box approach to adversarial generation, combining the approaches of both genetic algorithms and gradient estimation to solve the task. We achieve a 89.25% targeted attack similarity after 3000 generations while maintaining 94.6% audio file similarity.""","""The authors propose an algorithm for generating adversarial examples for ASR systems treating them as black boxes. Strengths - One of the early works to demonstrate black box attacks on ASR system that recognize phrases instead of isolated words. Weaknesses - The approach assumes that the logits are available, which may not be realistic for most ASR systems when they are used in practice -- typically only the final transcription is available. - Although the technique is applied to continuous speech, algorithmic improvements over prior work of Alzanot et al. is minimal. - Evaluation is weak. For example, cross correlation cannot completely capture the adversarial nature of a generated audio sample. - The authors use a genetic algorithm for generating new set of examples which are pruned and mutated. Its not clear what guarantees exist that the algorithm will eventually succeed. The reviewers agree that the presented work puts forth an interesting research direction. But given the deficiencies of the current submission as pointed out by the reviewers, the recommendation is to reject the paper."""
375,"""Dynamic Channel Pruning: Feature Boosting and Suppression""","['dynamic network', 'faster CNNs', 'channel pruning']","""Making deep convolutional neural networks more accurate typically comes at the cost of increased computational and memory resources. In this paper, we reduce this cost by exploiting the fact that the importance of features computed by convolutional layers is highly input-dependent, and propose feature boosting and suppression (FBS), a new method to predictively amplify salient convolutional channels and skip unimportant ones at run-time. FBS introduces small auxiliary connections to existing convolutional layers. In contrast to channel pruning methods which permanently remove channels, it preserves the full network structures and accelerates convolution by dynamically skipping unimportant input and output channels. FBS-augmented networks are trained with conventional stochastic gradient descent, making it readily available for many state-of-the-art CNNs. We compare FBS to a range of existing channel pruning and dynamic execution schemes and demonstrate large improvements on ImageNet classification. Experiments show that FBS can respectively provide 5 and 2 savings in compute on VGG-16 and ResNet-18, both with less than 0.6% top-5 accuracy loss.""","""The authors propose a dynamic inference technique for accelerating neural network prediction with minimal accuracy loss. The method are simple and effective. The paper is clear and easy to follow. However, the real speedup on CPU/GPU is not demonstrated beyond the theoretical FLOPs reduction. Reviewers are also concerned that the idea of dynamic channel pruning is not novel. The evaluation is on fairly old networks."""
376,"""Janossy Pooling: Learning Deep Permutation-Invariant Functions for Variable-Size Inputs""","['representation learning', 'permutation invariance', 'set functions', 'feature pooling']","""We consider a simple and overarching representation for permutation-invariant functions of sequences (or set functions). Our approach, which we call Janossy pooling, expresses a permutation-invariant function as the average of a permutation-sensitive function applied to all reorderings of the input sequence. This allows us to leverage the rich and mature literature on permutation-sensitive functions to construct novel and flexible permutation-invariant functions. If carried out naively, Janossy pooling can be computationally prohibitive. To allow computational tractability, we consider three kinds of approximations: canonical orderings of sequences, functions with k-order interactions, and stochastic optimization algorithms with random permutations. Our framework unifies a variety of existing work in the literature, and suggests possible modeling and algorithmic extensions. We explore a few in our experiments, which demonstrate improved performance over current state-of-the-art methods.""","""AR1 is concerned about whether higher-order interactions are modeled explicitly and if pi-SGD convergence conditions can be easily satisfied. AR2 is concerned that basic JP has been conceptually discussed in the literature and \pi-SGD is not novel because it was realized by Hamilton et al. (2017) and Moore & Neville (2017). However, the authors provide some theoretical analysis for this setting in contrast to prior works. AR1 is also concerned that the effect of higher-order information has not been 'disentangled' experimentally from order invariance. AR4 is concerned about poor performance of higher order Janossy pooling compared to k =1 case and asks about the number of hyper-parameters. The authors showed a harder task of computing the variance of a sequence of numbers in response. On balance, despite justified concerns of AR2 about novelty and AR1 about experimental verification, the work appears to tackle an interesting topic. Reviewers find the problem interesting and see some hope in the proposed solutions. On balance, AC recommends this paper to be accepted at ICLR. The authors are asked to update manuscript to reflect honestly weaknesses as expressed by reviewers, e.g. issue with effects of 'higher-order information' and 'disentangled' from order invariance."""
377,"""Improving Sample-based Evaluation for Generative Adversarial Networks""",[],"""In this paper, we propose an improved quantitative evaluation framework for Generative Adversarial Networks (GANs) on generating domain-specific images, where we improve conventional evaluation methods on two levels: the feature representation and the evaluation metric. Unlike most existing evaluation frameworks which transfer the representation of ImageNet inception model to map images onto the feature space, our framework uses a specialized encoder to acquire fine-grained domain-specific representation. Moreover, for datasets with multiple classes, we propose Class-Aware Frechet Distance (CAFD), which employs a Gaussian mixture model on the feature space to better fit the multi-manifold feature distribution. Experiments and analysis on both the feature level and the image level were conducted to demonstrate improvements of our proposed framework over the recently proposed state-of-the-art FID method. To our best knowledge, we are the first to provide counter examples where FID gives inconsistent results with human judgments. It is shown in the experiments that our framework is able to overcome the shortness of FID and improves robustness. Code will be made available.""","""The paper proposes a novel sample based evaluation metric which extends the idea of FID by replacing the latent features of the inception network by those of a data-set specific (V)AE and the FID by the mean FID of the class-conditional distributions. Furthermore, the paper presents interesting examples for which FID fails to match the human judgment while the new metric does not. All reviewers agree, that while these ideas are interesting, they are not convinced about the originality and significance of the contribution and believe that the work could be improved by a deeper analysis and experimental investigation. """
378,"""Transfer Learning for Related Reinforcement Learning Tasks via Image-to-Image Translation""","['Transfer Learning', 'Reinforcement Learning', 'Generative Adversarial Networks', 'Video Games']","""Deep Reinforcement Learning has managed to achieve state-of-the-art results in learning control policies directly from raw pixels. However, despite its remarkable success, it fails to generalize, a fundamental component required in a stable Artificial Intelligence system. Using the Atari game Breakout, we demonstrate the difficulty of a trained agent in adjusting to simple modifications in the raw image, ones that a human could adapt to trivially. In transfer learning, the goal is to use the knowledge gained from the source task to make the training of the target task faster and better. We show that using various forms of fine-tuning, a common method for transfer learning, is not effective for adapting to such small visual changes. In fact, it is often easier to re-train the agent from scratch than to fine-tune a trained agent. We suggest that in some cases transfer learning can be improved by adding a dedicated component whose goal is to learn to visually map between the known domain and the new one. Concretely, we use Unaligned Generative Adversarial Networks (GANs) to create a mapping function to translate images in the target task to corresponding images in the source task. These mapping functions allow us to transform between various variations of the Breakout game, as well as between different levels of a Nintendo game, Road Fighter. We show that learning this mapping is substantially more efficient than re-training. A visualization of a trained agent playing Breakout and Road Fighter, with and without the GAN transfer, can be seen in \url{pseudo-url} and \url{pseudo-url}.""","""The paper proposes an transfer learning approach to reinforcement learning, where observations from a target domain are mapped to a source domain in which the algorithm was originally trained. Using unsupervised GAN models to learn this mapping from unaligned samples, the authors show that such a mapping allows the RL agent to successfully interact with the target domain without further training (apart from training the GAN models). The approach is empirically validated on modified versions of the Atari game breakout, as well as subsequent levels of Road Fighter, showing good performance on the transfer domain with a fraction of the samples that would be required for retraining the RL algorithm from scratch. The reviewers and AC note the strong motivation for this work and emphasize that they find the idea interesting and novel. Reviewer 3 emphasizes the detailed analysis and results. Reviewer 2 notes the innovative idea to evaluate GANs in this application domain. Reviewer 1 identifies a key contribution in the thorough empirical analysis of the generalization issues that plaque current RL algorithms, as well as the comparison between different GAN models and finding their performance to be task-specific. The reviewers and AC noted several potential weaknesses: The proposed training based on images collected by an untrained agent focus the data on experience that agents would see very early on in the game, and may lead to generalization issues in more advanced parts of the game. Indeed these generalization issues are one possible explanation for the discrepancies between qualitative and quantitative results noted by reviewer 1. While the quantitative results indicate good performance on the target task, the image to image translation makes substantial errors, e.g., hallucinating blocks in breakout and erasing cars in Road Fighter. To the AC, the current paper does not provide enough insight into why the translation approach works even in cases where key elements are added or removed from the scene. The paper would benefit from a revision that thoroughly analyses such cases as well as the reason why the trained RL policy is able to generalize to them. R1 further notes that the paper does not address the RL generalization issue, but rather presents an empirical study that shows that in specific cases it is easier to translate from a target to a source domain, than to learn a policy for the target domain. The AC shares this concern, especially given the limited error analysis and conceptual insights derived from the empirical study. There are further concerns about the experimental protocol and hyper-parameter selection on the target tasks. Finally reviewer 1 questions the claim of whether data efficiency matters more than training efficiency in the proposed setting. There is disagreement about this paper. Reviewers 2 and 3 gave high scores and positive reviews, but did not provide sufficient feedback to the concerns raised by reviewer 1, who put forward significant concerns. The AC is particularly concerned about the experimental protocol and hyper-parameter tuning directly on the test tasks. The authors counter this point by noting that ""We agree that selecting configurations based on the test set is far from ideal, but we also note that this is the de-facto standard in video game-playing RL works, so we do not believe our work is any worse than others in the literature in this regard."" The AC worries about the lack of motivation to identify a strong empirical setup to arrive at the strongest possible contribution. A key concern here is that the results seem to vary substantially by task, GAN model used, etc. and substantial tuning on the target domain seems to be required. This makes it hard to draw any generalizable conclusions. This concern can be alleviated by including additional analysis, e.g., error analysis of where a proposed approach fails, or additional experiments designed to isolate the factors that contribute to a particular performance level. However, the current paper does not go to this detail of empirical exploration. Given these concerns, I recommend not accepting the paper at the current stage."""
379,"""Generating Multiple Objects at Spatially Distinct Locations""","['controllable image generation', 'text-to-image synthesis', 'generative model', 'generative adversarial network', 'gan']","""Recent improvements to Generative Adversarial Networks (GANs) have made it possible to generate realistic images in high resolution based on natural language descriptions such as image captions. Furthermore, conditional GANs allow us to control the image generation process through labels or even natural language descriptions. However, fine-grained control of the image layout, i.e. where in the image specific objects should be located, is still difficult to achieve. This is especially true for images that should contain multiple distinct objects at different spatial locations. We introduce a new approach which allows us to control the location of arbitrarily many objects within an image by adding an object pathway to both the generator and the discriminator. Our approach does not need a detailed semantic layout but only bounding boxes and the respective labels of the desired objects are needed. The object pathway focuses solely on the individual objects and is iteratively applied at the locations specified by the bounding boxes. The global pathway focuses on the image background and the general image layout. We perform experiments on the Multi-MNIST, CLEVR, and the more complex MS-COCO data set. Our experiments show that through the use of the object pathway we can control object locations within images and can model complex scenes with multiple objects at various locations. We further show that the object pathway focuses on the individual objects and learns features relevant for these, while the global pathway focuses on global image characteristics and the image background.""","""The submission proposes a model to generate images where one can control the fine-grained locations of objects. This is achieved by adding an ""object pathway"" to the GAN architecture. Experiments on a number of baselines are performed, including a number of reviewer-suggested metrics that were added post-rebuttal. The method needs bounding boxes of the objects to be placed (and labels). The proposed method is simple and likely novel and I like the evaluating done with Yolov3 to get a sense of the object detection performance on the generated images. I find the results (qual & quant) and write-up compelling and I think that the method will be of practical relevance, especially in creative applications. Because of this, I recommend acceptance."""
380,"""ChoiceNet: Robust Learning by Revealing Output Correlations""","['Robust Deep Learning', 'weakly supervised learning']","""In this paper, we focus on the supervised learning problem with corrupt training data. We assume that the training dataset is generated from a mixture of a target distribution and other unknown distributions. We estimate the quality of each data by revealing the correlation between the generated distribution and the target distribution. To this end, we present a novel framework referred to here as ChoiceNet that can robustly infer the target distribution in the presence of inconsistent data. We demonstrate that the proposed framework is applicable to both classification and regression tasks. Particularly, ChoiceNet is evaluated in comprehensive experiments, where we show that it constantly outperforms existing baseline methods in the handling of noisy data in synthetic regression tasks as well as behavior cloning problems. In the classification tasks, we apply the proposed method to the MNIST and CIFAR-10 datasets and it shows superior performances in terms of robustness to different types of noisy labels.""","""The paper addresses an interesting problem (learning in the presence of noisy labels) and provides extensive experiments. However, while the experiments in some sense cover a good deal of ground, reviewers raised issues with their quality, especially concerning baselines and depth (in terms of realism of the data). The authors provided many additional experiments during the rebuttal, but the reviewers did not find them sufficiently convincing."""
381,"""Visual Explanation by Interpretation: Improving Visual Feedback Capabilities of Deep Neural Networks""","['model explanation', 'model interpretation', 'explainable ai', 'evaluation']","""Visual Interpretation and explanation of deep models is critical towards wide adoption of systems that rely on them. In this paper, we propose a novel scheme for both interpretation as well as explanation in which, given a pretrained model, we automatically identify internal features relevant for the set of classes considered by the model, without relying on additional annotations. We interpret the model through average visualizations of this reduced set of features. Then, at test time, we explain the network prediction by accompanying the predicted class label with supporting visualizations derived from the identified features. In addition, we propose a method to address the artifacts introduced by strided operations in deconvNet-based visualizations. Moreover, we introduce an8Flower , a dataset specifically designed for objective quantitative evaluation of methods for visual explanation. Experiments on the MNIST , ILSVRC 12, Fashion 144k and an8Flower datasets show that our method produces detailed explanations with good coverage of relevant features of the classes of interest.""","""This was a difficult decision to converge to. R2 strongly champions this work, R1 is strongly critical, and R3 did not participate in the discussions (or take a stand). On the one hand, the AC can sympathize with R1's concerns -- insights developed on synthetic datasets may fail to generalize and fundamentally, the burden is not on a reviewer to be able to provide to authors a realistic dataset for the paper to experiment on. Having said that, a carefully constructed synthetic dataset is often *exactly* what the community needs as the first step to studying a difficult problem. Moreover, it is better for a proceeding to include works that generate vigorous discussions than the routine bland incremental works that typically dominate. Welcome to ICLR19. """
382,"""DOM-Q-NET: Grounded RL on Structured Language""","['Reinforcement Learning', 'Web Navigation', 'Graph Neural Networks']","""Building agents to interact with the web would allow for significant improvements in knowledge understanding and representation learning. However, web navigation tasks are difficult for current deep reinforcement learning (RL) models due to the large discrete action space and the varying number of actions between the states. In this work, we introduce DOM-Q-NET, a novel architecture for RL-based web navigation to address both of these problems. It parametrizes Q functions with separate networks for different action categories: clicking a DOM element and typing a string input. Our model utilizes a graph neural network to represent the tree-structured HTML of a standard web page. We demonstrate the capabilities of our model on the MiniWoB environment where we can match or outperform existing work without the use of expert demonstrations. Furthermore, we show 2x improvements in sample efficiency when training in the multi-task setting, allowing our model to transfer learned behaviours across tasks. ""","""This paper considers the task of web navigation, i.e. given a goal expressed in natural language, the task is to navigate webs by filling up fields and clicking links. The proposed model uses reinforcement learning, introducing a novel extension where the graph embedding of the pages is incorporated into the Q-function. The results are sound, and the paper is overall well-written. The reviewers and AC note the following potential weaknesses. The primary concern that was raised was the novelty. Since the task could potentially be framed as semantic parsing, reviewer 4 mentioned there may be readily available approaches for baselines that the authors did not consider. The comparison to semantic parsing required a more detailed discussion, pointing not only the differences but also the similarities, that would encourage the two communities to explore novel approaches to their tasks. Further, reviewer 2 was concerned about the limited novelty, given the extensive work that combines GNN and RL, such as NerveNet. The authors provided comments and a revision to address these issues. They described why it is not trivial to formulate their setup as a semantic parsing problem, partly due to the fact that the environment is partially observable. Similarly, the authors described the differences between the proposed approach and methods like NerveNet, such as the use of a dynamic graph and off-policy RL, making the latter not a viable baseline for the task. These changes addressed most of the concerns raised by the reviewers. The reviewers agreed that this paper should be accepted."""
383,"""Step-wise Sensitivity Analysis: Identifying Partially Distributed Representations for Interpretable Deep Learning""","['Interpretability', 'Interpretable Deep Learning', 'XAI', 'dependency graph', 'sensitivity analysis', 'outlier detection', 'instance-specific', 'model-centric']",""" In this paper, we introduce a novel method, called step-wise sensitivity analysis, which makes three contributions towards increasing the interpretability of Deep Neural Networks (DNNs). First, we are the first to suggest a methodology that aggregates results across input stimuli to gain model-centric results. Second, we linearly approximate the neuron activation and propose to use the outlier weights to identify distributed code. Third, our method constructs a dependency graph of the relevant neurons across the network to gain fine-grained understanding of the nature and interactions of DNN's internal features. The dependency graph illustrates shared subgraphs that generalise across 10 classes and can be clustered into semantically related groups. This is the first step towards building decision trees as an interpretation of learned representations.""","""This work proposes a modification of gradient based saliency map methods that measure the importance of all nodes at each layer. The reviewers found the novelty is rather marginal and that the evaluation is not up to par (since it's mostly qualitative). The reviewers are in strong agreement that this work does not pass the bar for acceptance."""
384,"""Learning with Reflective Likelihoods""","['new learning criterion', 'penalized maximum likelihood', 'posterior inference in deep generative models', 'input forgetting issue', 'latent variable collapse issue']","""Models parameterized by deep neural networks have achieved state-of-the-art results in many domains. These models are usually trained using the maximum likelihood principle with a finite set of observations. However, training deep probabilistic models with maximum likelihood can lead to the issue we refer to as input forgetting. In deep generative latent-variable models, input forgetting corresponds to posterior collapse---a phenomenon in which the latent variables are driven independent from the observations. However input forgetting can happen even in the absence of latent variables. We attribute input forgetting in deep probabilistic models to the finite sample dilemma of maximum likelihood. We formalize this problem and propose a learning criterion---termed reflective likelihood---that explicitly prevents input forgetting. We empirically observe that the proposed criterion significantly outperforms the maximum likelihood objective when used in classification under a skewed class distribution. Furthermore, the reflective likelihood objective prevents posterior collapse when used to train stochastic auto-encoders with amortized inference. For example in a neural topic modeling experiment, the reflective likelihood objective leads to better quantitative and qualitative results than the variational auto-encoder and the importance-weighted auto-encoder.""","""The proposed input forgetting problem is interesting, and the reflective likelihood can come to be seen as a natural solution, however the reviewers overall are concerned about the rigor of the paper. Reviewer 2 pointed out a technical flaw and this was addressed, however the reviewers remain unconvinced about the theoretical justification for the approach. One suggestion made by reviewer 1 is to focus on simpler models that can be studied more rigorously. Alternatively, it could be useful to focus on stronger empirical results. The method works in the experiments given, but for example in the imbalanced data experiments, only MLE is compared to as a baseline. I think it would be more convincing to compare against stronger baselines from the literature. If they are orthogonal to the choice of estimator, then it would be even better to show that these baselines + RLL outperforms the baselines + MLE. Alternatively, you mention some challenging tasks like seq2seq, where a convincing demonstration would greatly strengthen the paper. While the paper is not yet ready in its current form, it seems like a promising approach that is worth further exploration."""
385,"""EXPLORATION OF EFFICIENT ON-DEVICE ACOUSTIC MODELING WITH NEURAL NETWORKS""","['Parallelization', 'Speech Recognition', 'Sequence Modeling', 'Recurrent Neural Network', 'Embedded Systems']","""Real-time speech recognition on mobile and embedded devices is an important application of neural networks. Acoustic modeling is the fundamental part of speech recognition and is usually implemented with long short-term memory (LSTM)-based recurrent neural networks (RNNs). However, the single thread execution of an LSTM RNN is extremely slow in most embedded devices because the algorithm needs to fetch a large number of parameters from the DRAM for computing each output sample. We explore a few acoustic modeling algorithms that can be executed very efficiently on embedded devices. These algorithms reduce the overhead of memory accesses using multi-timestep parallelization that computes multiple output samples at a time by reading the parameters only once from the DRAM. The algorithms considered are the quasi RNNs (QRNNs), Gated ConvNets, and diagonalized LSTMs. In addition, we explore neural networks that equip one-dimensional (1-D) convolution at each layer of these algorithms, and by which can obtain a very large performance increase in the QRNNs and Gated ConvNets. The experiments were conducted using two tasks, one is the connectionist temporal classification (CTC)-based end-to-end speech recognition on WSJ corpus and the other is the phoneme classification on TIMIT dataset. We not only significantly increase the execution speed but also obtain a much higher accuracy, compared to LSTM RNN-based modeling. Thus, this work can be applicable not only to embedded system-based implementations but also to server-based ones.""","""In this work, the authors conduct experiments using variants of RNNs and Gated CNNs on a speech recognition task, motivated by the goal of reducing the computational requirements when deploying these models on mobile devices. While this is an important concern for practical deployment of ASR systems, the main concerns expressed by the reviewers is that the work lacks novelty. Further, the authors choice to investigate CTC based systems which predict characters. These models are not state-of-the-art for ASR, and as such it is hard to judge the impact of this work on a state-of-the-art embedded ASR system. Finally, it would be beneficial to replicate results on a much larger corpus such as Librispeech or Switchboard. Based on the unanimous decision from the reviewers, the AC agrees that the work, in the present form, should be rejected. """
386,"""Near-Optimal Representation Learning for Hierarchical Reinforcement Learning""",['representation hierarchy reinforcement learning'],"""We study the problem of representation learning in goal-conditioned hierarchical reinforcement learning. In such hierarchical structures, a higher-level controller solves tasks by iteratively communicating goals which a lower-level policy is trained to reach. Accordingly, the choice of representation -- the mapping of observation space to goal space -- is crucial. To study this problem, we develop a notion of sub-optimality of a representation, defined in terms of expected reward of the optimal hierarchical policy using this representation. We derive expressions which bound the sub-optimality and show how these expressions can be translated to representation learning objectives which may be optimized in practice. Results on a number of difficult continuous-control tasks show that our approach to representation learning yields qualitatively better representations as well as quantitatively better hierarchical policies, compared to existing methods.""","""Strong paper on hierarchical RL with very strong reviews from people expert in this subarea that I know well. """
387,"""Dynamically Unfolding Recurrent Restorer: A Moving Endpoint Control Method for Image Restoration""","['image restoration', 'differential equation']","""In this paper, we propose a new control framework called the moving endpoint control to restore images corrupted by different degradation levels in one model. The proposed control problem contains a restoration dynamics which is modeled by an RNN. The moving endpoint, which is essentially the terminal time of the associated dynamics, is determined by a policy network. We call the proposed model the dynamically unfolding recurrent restorer (DURR). Numerical experiments show that DURR is able to achieve state-of-the-art performances on blind image denoising and JPEG image deblocking. Furthermore, DURR can well generalize to images with higher degradation levels that are not included in the training stage.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The approach is novel - The experimental results are convincing. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The authors didn't show results with non-Gaussian noise - Some details that could help the understanding of the method are missing. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. No major points of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be accepted. """
388,"""BEHAVIOR MODULE IN NEURAL NETWORKS""","['Modular Networks', 'Reinforcement Learning', 'Task Separation', 'Representation Learning', 'Transfer Learning', 'Adversarial Transfer']","""Prefrontal cortex (PFC) is a part of the brain which is responsible for behavior repertoire. Inspired by PFC functionality and connectivity, as well as human behavior formation process, we propose a novel modular architecture of neural networks with a Behavioral Module (BM) and corresponding end-to-end training strategy. This approach allows the efficient learning of behaviors and preferences representation. This property is particularly useful for user modeling (as for dialog agents) and recommendation tasks, as allows learning personalized representations of different user states. In the experiment with video games playing, the resultsshow that the proposed method allows separation of main tasks objectives andbehaviors between different BMs. The experiments also show network extendability through independent learning of new behavior patterns. Moreover, we demonstrate a strategy for an efficient transfer of newly learned BMs to unseen tasks.""","""This paper takes inspiration from the brain to add a behavioral module to a deep reinforcement learning architecture. Unfortunately, the paper's structure and execution lacks clarity and requires a lot more work: as noted by reviewers, the link link between motivation and experiments is too fuzzy and their execution is not convincing."""
389,"""On the Statistical and Information Theoretical Characteristics of DNN Representations""","['learned representation', 'statistical characteristics', 'information theoretical characteristics', 'deep network']","""It has been common to argue or imply that a regularizer can be used to alter a statistical property of a hidden layer's representation and thus improve generalization or performance of deep networks. For instance, dropout has been known to improve performance by reducing co-adaptation, and representational sparsity has been argued as a good characteristic because many data-generation processes have only a small number of factors that are independent. In this work, we analytically and empirically investigate the popular characteristics of learned representations, including correlation, sparsity, dead unit, rank, and mutual information, and disprove many of the \textit{conventional wisdom}. We first show that infinitely many Identical Output Networks (IONs) can be constructed for any deep network with a linear layer, where any invertible affine transformation can be applied to alter the layer's representation characteristics. The existence of ION proves that the correlation characteristics of representation can be either low or high for a well-performing network. Extensions to ReLU layers are provided, too. Then, we consider sparsity, dead unit, and rank to show that only loose relationships exist among the three characteristics. It is shown that a higher sparsity or additional dead units do not imply a better or worse performance when the rank of representation is fixed. We also develop a rank regularizer and show that neither representation sparsity nor lower rank is helpful for improving performance even when the data-generation process has only a small number of independent factors. Mutual information pseudo-formula and pseudo-formula are investigated as well, and we show that regularizers can affect pseudo-formula and thus indirectly influence the performance. Finally, we explain how a rich set of regularizers can be used as a powerful tool for performance tuning. ""","""The paper considers an important problem of investigating the effects different statistical characteristics of representations (hidden unit activations) , such as sparsity, low correlation, etc, have on the neural network performance; while all reviewers agree that this is clearly a very important topic, there is also a consensus that perhaps the authors must strengthen and emphasize their contribution more clearly. """
390,"""Implicit Maximum Likelihood Estimation""","['likelihood-free inference', 'implicit probabilistic models']","""Implicit probabilistic models are models defined naturally in terms of a sampling procedure and often induces a likelihood function that cannot be expressed explicitly. We develop a simple method for estimating parameters in implicit models that does not require knowledge of the form of the likelihood function or any derived quantities, but can be shown to be equivalent to maximizing likelihood under some conditions. Our result holds in the non-asymptotic parametric setting, where both the capacity of the model and the number of data examples are finite. We also demonstrate encouraging experimental results. ""","""The manuscript proposes a novel estimation technique for generative models based on fast nearest neighbors and inspired by maximum likelihood estimation. Overall, reviewers and AC agree that the general problem statement is timely and interesting, and the subject is of interest to the ICLR community The reviewers and ACs note weakness in the evaluation of the proposed method. In particular, reviewers note that the Parzen-based log-likelihood estimate is known to be unreliable in high-dimensions. This makes a quantitative evaluation of the results challenging, thus other metrics should be evaluated. Reviewers also expressed concerns about the strengths of the baselines compared. Additional concerns are raised with regards to scalability which the authors address in the rebuttal. """
391,"""Self-Supervised Generalisation with Meta Auxiliary Learning""","['meta learning', 'auxiliary learning', 'multi-task learning', 'self-supervised learning']","""Auxiliary learning has been shown to improve the generalisation performance of a principal task. But typically, this requires manually-defined auxiliary tasks based on domain knowledge. In this paper, we consider that it may be possible to automatically learn these auxiliary tasks to best suit the principal task, towards optimum auxiliary tasks without any human knowledge. We propose a novel method, Meta Auxiliary Learning (MAXL), which we design for the task of image classification, where the auxiliary task is hierarchical sub-class image classification. The role of the meta learner is to determine sub-class target labels to train a multi-task evaluator, such that these labels improve the generalisation performance on the principal task. Experiments on three different CIFAR datasets show that MAXL outperforms baseline auxiliary learning methods, and is competitive even with a method which uses human-defined sub-class hierarchies. MAXL is self-supervised and general, and therefore offers a promising new direction towards automated generalisation.""","""This paper proposes a framework for generating auxiliary tasks as a means to regularize learning. The idea is interesting, and the method is simple. Two of the three reviewers found the paper to be well-written. The experiment include a promising result on the CIFAR dataset. The reviewer's brought up several concerns regarding the description of the method, the generality of the method (e.g. the requirement for class hierarchy), the validity and description of the comparisons, and the lack of experiments on domains with much more complex hierarchies. None of these concerns were not addressed in revisions to the paper. Hence, the paper in it's current state does not meet the bar for publication."""
392,"""A Better Baseline for Second Order Gradient Estimation in Stochastic Computation Graphs""","['Reinforcement learning', 'meta-learning', 'higher order derivatives', 'gradient estimation', 'stochastic computation graphs']","""Motivated by the need for higher order gradients in multi-agent reinforcement learning and meta-learning, this paper studies the construction of baselines for second order Monte Carlo gradient estimators in order to reduce the sample variance. Following the construction of a stochastic computation graph (SCG), the Infinitely Differentiable Monte-Carlo Estimator (DiCE) can generate correct estimates of arbitrary order gradients through differentiation. However, a baseline term that serves as a control variate for reducing variance is currently provided only for first order gradient estimation, limiting the utility of higher-order gradient estimates. To improve the sample efficiency of DiCE, we propose a new baseline term for higher order gradient estimation. This term may be easily included in the objective, and produces unbiased variance-reduced estimators under (automatic) differentiation, without affecting the estimate of the objective itself or of the first order gradient. We provide theoretical analysis and numerical evaluations of our baseline term, which demonstrate that it can dramatically reduce the variance of second order gradient estimators produced by DiCE. This computational tool can be easily used to estimate second order gradients with unprecedented efficiency wherever automatic differentiation is utilised, and has the potential to unlock applications of higher order gradients in reinforcement learning and meta-learning.""","""This paper extends the DiCE estimator with a better control variate baseline for variance reduction. The reviewers all think the paper is fairly clear and well written. However, as the reviews and discussion indicates, there are several critical issues, including lack of explanation of the choice of baseline, the lack more realistic experiments and a few misleading assertions. We encourage the authors to rewrite the paper to address these criticism. We believe this work will make a successful submission with proper modification in the future. """
393,"""Language Model Pre-training for Hierarchical Document Representations""",[],"""Hierarchical neural architectures can efficiently capture long-distance dependencies and have been used for many document-level tasks such as summarization, document segmentation, and fine-grained sentiment analysis. However, effective usage of such a large context can difficult to learn, especially in the case where there is limited labeled data available. Building on the recent success of language model pretraining methods for learning flat representations of text, we propose algorithms for pre-training hierarchical document representations from unlabeled data. Unlike prior work, which has focused on pre-training contextual token representations or context-independent sentence/paragraph representations, our hierarchical document representations include fixed-length sentence/paragraph representations which integrate contextual information from the entire documents. Experiments on document segmentation, document-level question answering, and extractive document summarization demonstrate the effectiveness of the proposed pre-training algorithms.""","""This paper proposes to pre-train hierarchical document representations for use in downstream tasks. All reviewers agreed that the results were reasonable. However, the methodological novelty is limited. While I believe there is a place for solid empirical results, even if not incredibly novel, there is also little qualitative or quantitative analysis to shed additional insights. Given the high quality bar for ICLR, I can't recommend the paper for acceptance at this time."""
394,"""EFFICIENT SEQUENCE LABELING WITH ACTOR-CRITIC TRAINING""","['Structured Prediction', 'Reinforcement Learning', 'NLP']","""Neural approaches to sequence labeling often use a Conditional Random Field (CRF) to model their output dependencies, while Recurrent Neural Networks (RNN) are used for the same purpose in other tasks. We set out to establish RNNs as an attractive alternative to CRFs for sequence labeling. To do so, we address one of the RNNs most prominent shortcomings, the fact that it is not exposed to its own errors with the maximum-likelihood training. We frame the prediction of the output sequence as a sequential decision-making process, where we train the network with an adjusted actor-critic algorithm (AC-RNN). We comprehensively compare this strategy with maximum-likelihood training for both RNNs and CRFs on three structured-output tasks. The proposed AC-RNN efficiently matches the performance of the CRF on NER and CCG tagging, and outperforms it on Machine Transliteration. We also show that our training strategy is significantly better than other techniques for addressing RNNs exposure bias, such as Scheduled Sampling, and Self-Critical policy training. ""","""this is an interesting approach to use reinforcement learning to replace CRF for sequence tagging, which would potentially be beneficial when the tag set is gigantic. unfortunately the conducted experiments do not really show this, which makes it difficult to see whether the proposed approach is indeed a viable alternative to CRF for sequence tagging with a large tag set. this sentiment was shared by all the reviewers, and R1 especially pointed out major and minor issues with the submission and was not convinced by the authors' response."""
395,"""On the effect of the activation function on the distribution of hidden nodes in a deep network""","['theory', 'length map', 'initialization']","""We analyze the joint probability distribution on the lengths of the vectors of hidden variables in different layers of a fully connected deep network, when the weights and biases are chosen randomly according to Gaussian distributions, and the input is binary-valued. We show that, if the activation function satisfies a minimal set of assumptions, satisfied by all activation functions that we know that are used in practice, then, as the width of the network gets large, the ``length process'' converges in probability to a length map that is determined as a simple function of the variances of the random weights and biases, and the activation function. We also show that this convergence may fail for activation functions that violate our assumptions.""","""I appreciate that the authors are refuting a technical claim in Poole et al., however the paper has garnered zero enthusiasm the way it is written. I suggest to the authors that they rewrite the paper as a refutation of Poole et al., and name it as such."""
396,"""Intriguing Properties of Learned Representations""","['deep learning', 'low rank representations', 'adversarial robustness']","""A key feature of neural networks, particularly deep convolutional neural networks, is their ability to learn useful representations from data. The very last layer of a neural network is then simply a linear model trained on these learned representations. Despite their numerous applications in other tasks such as classification, retrieval, clustering etc., a.k.a. transfer learning, not much work has been published that investigates the structure of these representations or indeed whether structure can be imposed on them during the training process. In this paper, we study the effective dimensionality of the learned representations by models that have proved highly successful for image classification. We focus on ResNet-18, ResNet-50 and VGG-19 and observe that when trained on CIFAR10 or CIFAR100, the learned representations exhibit a fairly low rank structure. We propose a modification to the training procedure, which further encourages low rank structure on learned activations. Empirically, we show that this has implications for robustness to adversarial examples and compression.""","""Dear authors, The reviewers all appreciated the interest of studying properties of the latent representations rather than of the weights. The impact of the rank on the robustness to adversarial attacks is also of interest. There were, however, two main issues raised. Due to the lack of confidence of some reviewers, I reviewed the paper myself and found the same issues: - Clarity could be improved. Some models are mentioned before being described (N-LR) and some important details are missing. In particular, we sometimes lose track of the goal of the experiments. For instance, there are quite a few experiments on the further reduction of the rank of the representation but it is not clear what to extract from them. - More importantly, there are several important gaps in the analysis. In particular: a/ As many reviewers have pointed out, low-rank constraints on the weight matrices induce low-rank representations if the activation function is linear. As it is not, this might not be true but deserves a discussion. b/ You state that the rank constraint has little effect given that the actual rank is much less than the constraint. However, one would expect the resulting rank to be a smooth function of the rank of the constraint. Since there is a discrepancy between ResNet N-LR and ResNet 1-LR, this should be investigated. c/ For the robustness to black-box adversarial attacks, these attacks are constructed using the N-LR models. Is is thus not too surprising that those models do not perform as well. Thus, despite the lack of confidence of one reviewer (the question about the N-LR models might stem from the fact that it is used before being introduced), I strongly encourage you to take their comments into account for a future submission."""
397,"""Dopamine: A Research Framework for Deep Reinforcement Learning""","['reinforcement learning', 'software', 'framework', 'reproducibility']","""Deep reinforcement learning (deep RL) research has grown significantly in recent years. A number of software offerings now exist that provide stable, comprehensive implementations for benchmarking. At the same time, recent deep RL research has become more diverse in its goals. In this paper we introduce Dopamine, a new research framework for deep RL that aims to support some of that diversity. Dopamine is open-source, TensorFlow-based, and provides compact yet reliable implementations of some state-of-the-art deep RL agents. We complement this offering with a taxonomy of the different research objectives in deep RL research. While by no means exhaustive, our analysis highlights the heterogeneity of research in the field, and the value of frameworks such as ours.""","""The paper presents Dopamine, an open-source implementation of plenty of DRL methods. It presents a case study of DQN and experiments on Atari. The paper is clear and easy to follow. While I believe Dopamine is a very welcomed contribution to the DRL software landscape, it seems there is not enough scientific content in this paper to warrant publication at ICLR. Regarding specifically the ELF and RLlib papers, I think that the ELF paper had a novelty component, and presented RL baselines to a new environment (miniRTS), while the RLlib paper had a stronger ""systems research"" contribution. This says nothing about the future impact of Dopamine, ELF, and RLlib the respective software."""
398,"""Understanding the Asymptotic Performance of Model-Based RL Methods""","['model-based reinforcement learning', 'mbrl', 'reinforcement learning', 'predictive models', 'predictive learning', 'forward models', 'deep learning']","""In complex simulated environments, model-based reinforcement learning methods typically lag the asymptotic performance of model-free approaches. This paper uses two MuJoCo environments to understand this gap through a series of ablation experiments designed to separate the contributions of the dynamics model and planner. These reveal the importance of long planning horizons, beyond those typically used. A dynamics model that directly predicts distant states, based on current state and a long sequence of actions, is introduced. This avoids the need for many recursions during long-range planning, and thus is able to yield more accurate state estimates. These accurate predictions allow us to uncover the relationship between model accuracy and performance, and translate to higher task reward that matches or exceeds current state-of-the-art model-free approaches.""","""The issue of when model based methods can be used successfully in RL is an interesting one. However, the reviewers had a number of concerns about the significance and framing of this work with respect to the related literature. In addition, the abstract and title suggest a very generic contribution will be made, whereas the actual contribution is to a much more specific subclass. Some relevant papers (and their related efforts) include the following. The Dependence of Effective Planning Horizon on Model Accuracy. (AAMAS-15, best paper award) Nan Jiang, Alex Kulesza, Satinder Singh, Richard Lewis. Self-Correcting Models for Model-Based Reinforcement Learning. Erik Talvitie. In 'Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI).' 2017. """
399,"""The Cakewalk Method""","['policy gradient', 'combinatorial optimization', 'blackbox optimization', 'stochastic optimization', 'reinforcement learning']","""Combinatorial optimization is a common theme in computer science. While in general such problems are NP-Hard, from a practical point of view, locally optimal solutions can be useful. In some combinatorial problems however, it can be hard to define meaningful solution neighborhoods that connect large portions of the search space, thus hindering methods that search this space directly. We suggest to circumvent such cases by utilizing a policy gradient algorithm that transforms the problem to the continuous domain, and to optimize a new surrogate objective that renders the former as generic stochastic optimizer. This is achieved by producing a surrogate objective whose distribution is fixed and predetermined, thus removing the need to fine-tune various hyper-parameters in a case by case manner. Since we are interested in methods which can successfully recover locally optimal solutions, we use the problem of finding locally maximal cliques as a challenging experimental benchmark, and we report results on a large dataset of graphs that is designed to test clique finding algorithms. Notably, we show in this benchmark that fixing the distribution of the surrogate is key to consistently recovering locally optimal solutions, and that our surrogate objective leads to an algorithm that outperforms other methods we have tested in a number of measures.""","""The paper investigates a variant of the ""cross-entropy method"" (CME) for heuristic combinatorial optimization, based on stochastically improving a search distribution via policy optimization in a surrogate objective. Unfortunately, the reviewers unanimously recommended rejection, noting that the significance of the contribution over CME remains far from clear and insufficiently supported by the given evidence. The experimental evaluation was unconvincing to all of the reviewers, particularly since only one artificial problem (clique finding) was considered in the paper (with an additional problem, k-medoid clustering, briefly and incompletely considered in the appendix). Several additional concerns were raised about the experimental evaluation, which triggered lengthy author responses but really need to be properly handled in the paper itself: - The sensitivity of performance to the optimization algorithm is a concern and requires more detailed understanding so that reasonable choices can be made in practice. - The independence assumption between search components is an extreme simplification that limits the appeal and applicability of the proposed approach. Even after author response, it remains unconvincing that an independent search distribution over subcomponents can be effective in challenging combinatorial spaces. Concrete evidence on challenging problems would be a more effective evidence than discussion. - The comparisons omitted any tailored algorithms for the specific problems. Even if the authors insist on only comparing to more ""general purpose"" methods, there is a large space of evolutionary and Bayesian optimization strategies that have been neglected from the comparison. A justification is needed for such an omission (if indeed it is even justifiable)."""
400,"""Multi-agent Deep Reinforcement Learning with Extremely Noisy Observations""","['Reinforcement learning', 'multi-agent', 'hierarchical', 'noisy observation', 'partial observability', 'deep learning']","""Multi-agent reinforcement learning systems aim to provide interacting agents with the ability to collaboratively learn and adapt to the behaviour of other agents. In many real-world applications, the agents can only acquire a partial view of the world. Here we consider a setting whereby most agents' observations are also extremely noisy, hence only weakly correlated to the true state of the environment. Under these circumstances, learning an optimal policy becomes particularly challenging, even in the unrealistic case that an agent's policy can be made conditional upon all other agents observations. To overcome these difficulties, we propose a multi-agent deep deterministic policy gradient algorithm enhanced by a communication medium (MADDPG-M), which implements a two-level, concurrent learning mechanism. An agent's policy depends on its own private observations as well as those explicitly shared by others through a communication medium. At any given point in time, an agent must decide whether its private observations are sufficiently informative to be shared with others. However, our environments provide no explicit feedback informing an agent whether a communication action is beneficial, rather the communication policies must also be learned through experience concurrently to the main policies. Our experimental results demonstrate that the algorithm performs well in six highly non-stationary environments of progressively higher complexity, and offers substantial performance gains compared to the baselines.""","""The paper presents an extension of MADDPG, adding communication between agents. The methods targets extremely noisy observations settings, so that agents need to decide if they communicate their private observations (or not). There is no intrinsic/explicit reward to guide the learning of the communication, only the extrinsic/implicit reward of the downstream task. The paper is clear and easy to follow, in particular after the updated writing. I believe some of the reviewers' points were addressed by the rebuttal. Nonetheless, some of the weaknesses of the paper still hold: namely the complexity of the approach compounded with a very specific experimental evaluation. The more complex an approach is (and it may be justified by the complexity of the setting!), the more varied its supporting evidence should be. In its current form, the paper would constitute a good workshop contribution (to discuss the approach), but I believe it needs more varied (and/or harder) experiments to be published at ICLR."""
401,"""Do Deep Generative Models Know What They Don't Know? ""","['deep generative models', 'out-of-distribution inputs', 'flow-based models', 'uncertainty', 'density']","""A neural network deployed in the wild may be asked to make predictions for inputs that were drawn from a different distribution than that of the training data. A plethora of work has demonstrated that it is easy to find or synthesize inputs for which a neural network is highly confident yet wrong. Generative models are widely viewed to be robust to such mistaken confidence as modeling the density of the input features can be used to detect novel, out-of-distribution inputs. In this paper we challenge this assumption. We find that the density learned by flow-based models, VAEs, and PixelCNNs cannot distinguish images of common objects such as dogs, trucks, and horses (i.e. CIFAR-10) from those of house numbers (i.e. SVHN), assigning a higher likelihood to the latter when the model is trained on the former. Moreover, we find evidence of this phenomenon when pairing several popular image data sets: FashionMNIST vs MNIST, CelebA vs SVHN, ImageNet vs CIFAR-10 / CIFAR-100 / SVHN. To investigate this curious behavior, we focus analysis on flow-based generative models in particular since they are trained and evaluated via the exact marginal likelihood. We find such behavior persists even when we restrict the flows to constant-volume transformations. These transformations admit some theoretical analysis, and we show that the difference in likelihoods can be explained by the location and variances of the data and the model curvature. Our results caution against using the density estimates from deep generative models to identify inputs similar to the training distribution until their behavior for out-of-distribution inputs is better understood.""","""This paper makes the intriguing observation that a density model trained on CIFAR10 has higher likelihood on SVHN than CIFAR10, i.e., it assigns higher probability to inputs that are out of the training distribution. This phenomenon is also shown to occur for several other dataset pairs. This finding is surprising and interesting, and the exposition is generally clear. The authors provide empirical and theoretical analysis, although based on rather strong assumptions. Overall, there's consensus among the reviewers that the paper would make a valuable contribution to the proceedings, and should therefore be accepted for publication."""
402,"""Diagnosing and Enhancing VAE Models""","['variational autoencoder', 'generative models']","""Although variational autoencoders (VAEs) represent a widely influential deep generative model, many aspects of the underlying energy function remain poorly understood. In particular, it is commonly believed that Gaussian encoder/decoder assumptions reduce the effectiveness of VAEs in generating realistic samples. In this regard, we rigorously analyze the VAE objective, differentiating situations where this belief is and is not actually true. We then leverage the corresponding insights to develop a simple VAE enhancement that requires no additional hyperparameters or sensitive tuning. Quantitatively, this proposal produces crisp samples and stable FID scores that are actually competitive with a variety of GAN models, all while retaining desirable attributes of the original VAE architecture. The code for our model is available at \url{pseudo-url}.""","""The reviewers acknowledge the value of the careful analysis of Gaussian encoder/decoder VAE presented in the paper. The proposed algorithm shows impressive FID scores that are comparable to those obtained by state of the art GANs. The paper will be a valuable addition to the ICLR program. """
403,"""GO Gradient for Expectation-Based Objectives""","['generalized reparameterization gradient', 'variance reduction', 'non-reparameterizable', 'discrete random variable', 'GO gradient', 'general and one-sample gradient', 'expectation-based objective', 'variable nabla', 'statistical back-propagation', 'hierarchical', 'graphical model']","""Within many machine learning algorithms, a fundamental problem concerns efficient calculation of an unbiased gradient wrt parameters pseudo-formula for expectation-based objectives (\boldsymbol{y})} [f (\boldsymbol{y}) ] Most existing methods either ( pseudo-formula ) suffer from high variance, seeking help from (often) complicated variance-reduction techniques; or ( pseudo-formula ) they only apply to reparameterizable continuous random variables and employ a reparameterization trick. To address these limitations, we propose a General and One-sample (GO) gradient that ( pseudo-formula ) applies to many distributions associated with non-reparameterizable continuous {\em or} discrete random variables, and ( pseudo-formula ) has the same low-variance as the reparameterization trick. We find that the GO gradient often works well in practice based on only one Monte Carlo sample (although one can of course use more samples if desired). Alongside the GO gradient, we develop a means of propagating the chain rule through distributions, yielding statistical back-propagation, coupling neural networks to common random variables.""","""This clearly written paper develops a novel, sound and comprehensive mathematical framework for computing low variance gradients of expectation-based objectives. The approach generalizes and encompasses several previous approaches for continuous random variables (reparametrization trick, Implicit Rep, pathwise gradients), and conveys novel insights. Importantly, and originally, it extends to discrete random variables, and to chains of continuous random variables with optionally discrete terminal variables. These contributions are well exposed, and supported by convincing experiments. Questions from reviewers were well addressed in the rebuttal and helped significantly clarify and improve the paper, in particular for delineating the novel contribution against prior related work. """
404,"""Robustness and Equivariance of Neural Networks""","['robust', 'adversarial', 'equivariance', 'rotations', 'GCNNs', 'CNNs', 'steerable', 'neural networks']","""Neural networks models are known to be vulnerable to geometric transformations as well as small pixel-wise perturbations of input. Convolutional Neural Networks (CNNs) are translation-equivariant but can be easily fooled using rotations and small pixel-wise perturbations. Moreover, CNNs require sufficient translations in their training data to achieve translation-invariance. Recent work by Cohen & Welling (2016), Worrall et al. (2016), Kondor & Trivedi (2018), Cohen & Welling (2017), Marcos et al. (2017), and Esteves et al. (2018) has gone beyond translations, and constructed rotation-equivariant or more general group-equivariant neural network models. In this paper, we do an extensive empirical study of various rotation-equivariant neural network models to understand how effectively they learn rotations. This includes Group-equivariant Convolutional Networks (GCNNs) by Cohen & Welling (2016), Harmonic Networks (H-Nets) by Worrall et al. (2016), Polar Transformer Networks (PTN) by Esteves et al. (2018) and Rotation equivariant vector field networks by Marcos et al. (2017). We empirically compare the ability of these networks to learn rotations efficiently in terms of their number of parameters, sample complexity, rotation augmentation used in training. We compare them against each other as well as Standard CNNs. We observe that as these rotation-equivariant neural networks learn rotations, they instead become more vulnerable to small pixel-wise adversarial attacks, e.g., Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD), in comparison with Standard CNNs. In other words, robustness to geometric transformations in these models comes at the cost of robustness to small pixel-wise perturbations.""","""Positives: The paper proposes an interesting idea: to study the effect on vulnerability to adversarial attacks of training for invariance with respect to rotations. Experiments on MNIST, FashionMNIST, and CIFAR10. An interesting hypothesis partially borne out in experiments. Negatives: no accept recommendation from any reviewer insufficient empirical results not a clear enough message very limited theoretical contribution Although additional experimental results on FashionMNIST and CIFAR10 were added to the initial very limited results on MNIST, the main claim of the paper seems to be somewhat weakened. The effect of increased vulnerability to adversarial attacks as invariance is increased is less pronounced on the additional datasets. This calls into question how relevant this effect is on more realistic data than the toy problems considered here. The size of the network is not varied in the experiments. If increased invariance results in poorer performance with respect to attacks, one possible explanation is that the invariance taxes the capacity of the network architecture. Varying architecture depth could partially answer whether this is relevant. Given the lack of theoretical contribution, more insights along these lines would potentially strengthen the work. The title uses the term ""equivariance,"" which strictly speaking is when the inputs and outputs of a function vary equally, e.g. an image and its segmentation are equivariant under rotations, but classification tasks should probably be called ""invariant."" The reviewers were unanimous in not recommending the paper for acceptance. The key concerns remain after the author response. """
405,"""Towards the Latent Transcriptome""","['representation learning', 'RNA-Seq', 'gene expression', 'bioinformatics', 'computational biology', 'transcriptomics', 'deep learning', 'genomics']","""In this work we propose a method to compute continuous embeddings for kmers from raw RNA-seq data, in a reference-free fashion. We report that our model captures information of both DNA sequence similarity as well as DNA sequence abundance in the embedding latent space. We confirm the quality of these vectors by comparing them to known gene sub-structures and report that the latent space recovers exon information from raw RNA-Seq data from acute myeloid leukemia patients. Furthermore we show that this latent space allows the detection of genomic abnormalities such as translocations as well as patient-specific mutations, making this representation space both useful for visualization as well as analysis.""","""This paper proposes to learn continuous of k-mer embeddings for RNA-seq analysis. Major concerns of the paper include: 1. novelty seems limited; 2. questions about the scalability of the approach; 3. evaluation experiments were not suitable for supporting the aim. Overall, this paper cannot be accepted yet. """
406,"""A RECURRENT NEURAL CASCADE-BASED MODEL FOR CONTINUOUS-TIME DIFFUSION PROCESS""","['Information Diffusion', 'Recurrent Neural Network', 'Black Box Inference']","""Many works have been proposed in the literature to capture the dynamics of diffusion in networks. While some of them define graphical markovian models to extract temporal relationships between node infections in networks, others consider diffusion episodes as sequences of infections via recurrent neural models. In this paper we propose a model at the crossroads of these two extremes, which embeds the history of diffusion in infected nodes as hidden continuous states. Depending on the trajectory followed by the content before reaching a given node, the distribution of influence probabilities may vary. However, content trajectories are usually hidden in the data, which induces challenging learning problems. We propose a topological recurrent neural model which exhibits good experimental performances for diffusion modelling and prediction. ""","""This paper introduces a recurrent neural network approach for learning diffusion dynamics in networks. The main advantage is that it embeds the history of diffusion and incorporates the structure of independent cascades for diffusion modeling and prediction. This is an important problem, and the proposed approach is novel and provides some empirical improvements. However, there is a lack of theoretical analysis, and in particular modeling choices and consequences of these choices should be emphasized more clearly. While there wasn't a consensus, a majority of the reviewers believe the paper is not ready for publication. """
407,"""Towards Understanding Regularization in Batch Normalization""","['batch normalization', 'regularization', 'deep learning']","""Batch Normalization (BN) improves both convergence and generalization in training neural networks. This work understands these phenomena theoretically. We analyze BN by using a basic block of neural networks, consisting of a kernel layer, a BN layer, and a nonlinear activation function. This basic network helps us understand the impacts of BN in three aspects. First, by viewing BN as an implicit regularizer, BN can be decomposed into population normalization (PN) and gamma decay as an explicit regularization. Second, learning dynamics of BN and the regularization show that training converged with large maximum and effective learning rate. Third, generalization of BN is explored by using statistical mechanics. Experiments demonstrate that BN in convolutional neural networks share the same traits of regularization as the above analyses.""","""+ the ideas presented in the paper are quite intriguing and draw on a variety of different connections - the presentation has a lot of room for improvement. In particular, the statement of Theorem 1, in its current form, requires rephrasing and making it more rigorous. Still, the general consensus is that, once these presentation shortcomings are address, this will be an interesting paper. """
408,"""NETWORK COMPRESSION USING CORRELATION ANALYSIS OF LAYER RESPONSES""","['Artificial Intelligence', 'Deep learning', 'Machine learning', 'Compression']","""Principal Filter Analysis (PFA) is an easy to implement, yet effective method for neural network compression. PFA exploits the intrinsic correlation between filter responses within network layers to recommend a smaller network footprint. We propose two compression algorithms: the first allows a user to specify the proportion of the original spectral energy that should be preserved in each layer after compression, while the second is a heuristic that leads to a parameter-free approach that automatically selects the compression used at each layer. Both algorithms are evaluated against several architectures and datasets, and we show considerable compression rates without compromising accuracy, e.g., for VGG-16 on CIFAR-10, CIFAR-100 and ImageNet, PFA achieves a compression rate of 8x, 3x, and 1.4x with an accuracy gain of 0.4%, 1.4% points, and 2.4% respectively. In our tests we also demonstrate that networks compressed with PFA achieve an accuracy that is very close to the empirical upper bound for a given compression ratio. Finally, we show how PFA is an effective tool for simultaneous compression and domain adaptation.""","""The authors propose a technique for compressing neural networks by examining the correlations between filter responses, by removing filters which are highly correlated. This differentiates the authors work from many other works which compress the weights independent of the task/domain. Strengths: Clearly written paper PFA-KL does not require additional hyperparameter tuning (apart from those implicit in choosing \psi) Experiments demonstrate that the number of filters determined by the algorithm scale with complexity of the task Weaknesses: Results on large-scale tasks such as Imagenet (subsequently added by the authors during the rebuttal period) Compression after the fact may not be as good as training with a modified loss function that does compression jointly Insufficient comparisons on ResNet architectures which make comparisons against previous works harder Overall, the reviewers were in agreement that this work (particularly, the revised version) was close to the acceptance threshold. In the ACs view, the authors addressed many of the concerns raised by the reviewers in the revisions. However, after much deliberation, the AC decided that the weaknesses 2, and 3 above were significant, and that these should be addressed in a subsequent submission."""
409,"""On-Policy Trust Region Policy Optimisation with Replay Buffers""","['reinforcement learning', 'on-policy learning', 'trust region policy optimisation', 'replay buffer']","""Building upon the recent success of deep reinforcement learning methods, we investigate the possibility of on-policy reinforcement learning improvement by reusing the data from several consecutive policies. On-policy methods bring many benefits, such as ability to evaluate each resulting policy. However, they usually discard all the information about the policies which existed before. In this work, we propose adaptation of the replay buffer concept, borrowed from the off-policy learning setting, to the on-policy algorithms. To achieve this, the proposed algorithm generalises the Q-, value and advantage functions for data from multiple policies. The method uses trust region optimisation, while avoiding some of the common problems of the algorithms such as TRPO or ACKTR: it uses hyperparameters to replace the trust region selection heuristics, as well as the trainable covariance matrix instead of the fixed one. In many cases, the method not only improves the results comparing to the state-of-the-art trust region on-policy learning algorithms such as ACKTR and TRPO, but also with respect to their off-policy counterpart DDPG. ""","""The reviewers raise an important issue about the parameters in the proposed gradient in Theorem 1. There could be different parameters for each policy in the gradient (though some parameter sharing could be possible), and computing this gradient would be prohibitive. The solution is to just use the most recent parameters, but then the gradients become off-policy again without motivation for why this is acceptable. This approximation needs to be better justified. As an additional point, there are other off-policy policy gradient methods, than just DPG. The authors could consider comparing to these strategies (which can use replay buffers) and explain why the proposed strategy provides benefits beyond these. What is inadequate about these methods? Further motivation is needed for the proposed strategy. This is additionally true because the proposed strategy requires entire sampled trajectories for a fixed policy (to make the policy gradient sound, with weighting dpi_n(s)), whereas DPG and other off-policy AC methods do not need that."""
410,"""The Comparative Power of ReLU Networks and Polynomial Kernels in the Presence of Sparse Latent Structure""","['theory', 'representational power', 'universal approximators', 'polynomial kernels', 'latent sparsity', 'beyond worst case', 'separation result']","""There has been a large amount of interest, both in the past and particularly recently, into the relative advantage of different families of universal function approximators, for instance neural networks, polynomials, rational functions, etc. However, current research has focused almost exclusively on understanding this problem in a worst case setting: e.g. characterizing the best L1 or L_{infty} approximation in a box (or sometimes, even under an adversarially constructed data distribution.) In this setting many classical tools from approximation theory can be effectively used. However, in typical applications we expect data to be high dimensional, but structured -- so, it would only be important to approximate the desired function well on the relevant part of its domain, e.g. a small manifold on which real input data actually lies. Moreover, even within this domain the desired quality of approximation may not be uniform; for instance in classification problems, the approximation needs to be more accurate near the decision boundary. These issues, to the best of our knowledge, have remain unexplored until now. With this in mind, we analyze the performance of neural networks and polynomial kernels in a natural regression setting where the data enjoys sparse latent structure, and the labels depend in a simple way on the latent variables. We give an almost-tight theoretical analysis of the performance of both neural networks and polynomials for this problem, as well as verify our theory with simulations. Our results both involve new (complex-analytic) techniques, which may be of independent interest, and show substantial qualitative differences with what is known in the worst-case setting.""","""This paper makes a substantial contribution to the understanding of the approximation ability of deep networks in comparison to classical approximation classes, such as polynomials. Strong results are given that show fundamental advantages for neural network function approximators in the presence of a natural form of latent structure. The analysis techniques required to achieve these results are novel and worth reporting to the community. The reviewers are uniformly supportive."""
411,"""Label Propagation Networks""","['semi supervised learning', 'graph networks', 'deep learning architectures']","""Graph networks have recently attracted considerable interest, and in particular in the context of semi-supervised learning. These methods typically work by generating node representations that are propagated throughout a given weighted graph. Here we argue that for semi-supervised learning, it is more natural to consider propagating labels in the graph instead. Towards this end, we propose a differentiable neural version of the classic Label Propagation (LP) algorithm. This formulation can be used for learning edge weights, unlike other methods where weights are set heuristically. Starting from a layer implementing a single iteration of LP, we proceed by adding several important non-linear steps that significantly enhance the label-propagating mechanism. Experiments in two distinct settings demonstrate the utility of our approach. ""","""This paper is on graph based semi-supervised learning where the goal is to develop an approach to jointly the node labeling function together with the edge weights. A natural way to formulate this problem as a bi-level optimization problem. However, the authors claim that this approach introduces two main difficulties: (a) the ""upper"" objective function is itself the solution to the ""lower"" optimization problem (Eq. (2)), and (b) optimization is challenging (Eq. (3)). The AC disagrees. Firstly, there is a close connection between the constrained version and the regression version of the problem (e.g., Belkin, Matveeva and Niyogi) -- the former is infact a special case of the latter for a certain choice of regularization parameter. The latter reduces to an linear system. The outer problem can be optimized using standard gradient descent using the implicit function theorem trick common in bilevel optimization. Reviewers have also raised concerns about clarity, and experimental support in this paper and comparisons with related work. """
412,"""Multilingual Neural Machine Translation with Knowledge Distillation""","['NMT', 'Multilingual NMT', 'Knowledge Distillation']","""Multilingual machine translation, which translates multiple languages with a single model, has attracted much attention due to its efficiency of offline training and online serving. However, traditional multilingual translation usually yields inferior accuracy compared with the counterpart using individual models for each language pair, due to language diversity and model capacity limitations. In this paper, we propose a distillation-based approach to boost the accuracy of multilingual machine translation. Specifically, individual models are first trained and regarded as teachers, and then the multilingual model is trained to fit the training data and match the outputs of individual models simultaneously through knowledge distillation. Experiments on IWSLT, WMT and Ted talk translation datasets demonstrate the effectiveness of our method. Particularly, we show that one model is enough to handle multiple languages (up to 44 languages in our experiment), with comparable or even better accuracy than individual models.""","""This paper presents good empirical results on an important and interesting task (translation between several language pairs with a single model). There was solid communication between the authors and the reviewers leading to an improved updated version and consensus among the reviewers about the merits of the paper."""
413,"""Lipschitz regularized Deep Neural Networks generalize""","['Deep Neural Networks', 'Regularization', 'Generalization', 'Convergence', 'Lipschitz', 'Stability']","""We show that if the usual training loss is augmented by a Lipschitz regularization term, then the networks generalize. We prove generalization by first establishing a stronger convergence result, along with a rate of convergence. A second result resolves a question posed in Zhang et al. (2016): how can a model distinguish between the case of clean labels, and randomized labels? Our answer is that Lipschitz regularization using the Lipschitz constant of the clean data makes this distinction. In this case, the model learns a different function which we hypothesize correctly fails to learn the dirty labels. ""","""This paper is entitled Lipschitz regularized deep networks generalize. In fact, the paper has nothing in particular to do with neural networks. It is really the study of minimizers of a Lipschitz-regularized risk functional over certain nonparametric classes. The connection with neural networks is simply that one can usually achieve zero empirical risk for (overparametrized) neural networks and so, in deep learning practice, neural networks behave like a nonparametric class. Given the lack of connection with neural networks, one cannot logically learn anything specific about neural networks from this paper. It should be renamed... perhaps ""Lipschitz regularization with an application to deep learning"". One could raise issues of technical novelty, as it seems many of the key results are known. I also question the insight that the bounds provide: they end up depending exponentially on the dimension of the data manifold. In the noiseless case, this exponential dependence arises from a triangle inequality between an arbitrary data point x and the nearest training data point! In the noisy case, this exponential dependence appears in a nonasymptotic uniform law of large numbers over the class of L-Lipschitz functions. There's no insight into deep learning here. It's also hard to judge whether these rates are what is explaining deep learning practice: it's unclear what the manifold dimensionality is, but it seems unlikely that this bound explains empirical performance (even if it describes the asymptotic rate of convergence). One of the main results shows that, in the face of corrupted label (corrupted in a particular way), Lipschitz regularization can ```""undo"" the corruption. However, convergence is not measured with respect to the true labeling function, but rather to the solution to the population regularized risk functional. How this solution relates to the true labeling function is unclear. The paper also purports to resolve a mystery of generalization raised by Zhang et al (ICLR 2017). In that paper, the authors point to the diametrically opposed generalization performance on ""true"" and ""random"" labels. In fact, this paper does not resolve this problem because Zhang et al. were interested in how SGD solves this problem without explicit regularization. That Lipschitz regularization could solve this problem is borderline obvious. I wanted to make a few comments. In the rebuttal with reviewers, the question of parametric rates comes up. I think there's some confusion on both the part of the reviewer and authors. The parametric rates are often apparent but not real. The complexity terms often have an uncharacterized dependence on the number of data (through the learning algorithm) and on the size of the network (which is implicitly chosen based on the data complexity). In practice, these bounds are vacuous. At some point, the authors argue that ""In practice, u_n(x) is rounded to the nearest label, so once |u_n-u_0| < 1/2, all classification results will be correct after rounding."" I'm not entirely sure I understand the logic here. First, convergence to u_0 is not controlled, but rather convergence to u*. u* may spend most of its time near the decision boundary, rendering uniform convergence almost useless. One would need noise conditions (Tysbakov) to make some claim. Some other issues: 1. in (1), u ranges over X\to Y, but is then applied also to a weight vector. 2. Is""continuum variational problem"" jargon? If so, cite. Otherwise, taking limits of rho_n and J makes sense only if J is suitably continuous, which depends on the loss function. You later address this convergence and so you should foreshadow. 3. Notation L[u;\rho] in (5) should be L[u,\rho], no? 4. (Goodfellow et al., 2016, Section 5.2) is an inappropriate citation for the term ""Generalization"". 5. in Thm 2.7., there is reference to a sequence mu_n and i assume the sequence elements is indexed by n, but then n appears in the probability with which the bound holds, and so this bound is not about the sequence but about a solution for \rho_n for fixed n. 6. Id should not be italicized in the statement of Lemma 2.10. Use mathrm not text/textrm. it should also be defined. 7. ""convex convex"" typo."""
414,"""Experience replay for continual learning""","['continual learning', 'catastrophic forgetting', 'lifelong learning', 'behavioral cloning', 'reinforcement learning', 'interference', 'stability-plasticity']","""Continual learning is the problem of learning new tasks or knowledge while protecting old knowledge and ideally generalizing from old experience to learn new tasks faster. Neural networks trained by stochastic gradient descent often degrade on old tasks when trained successively on new tasks with different data distributions. This phenomenon, referred to as catastrophic forgetting, is considered a major hurdle to learning with non-stationary data or sequences of new tasks, and prevents networks from continually accumulating knowledge and skills. We examine this issue in the context of reinforcement learning, in a setting where an agent is exposed to tasks in a sequence. Unlike most other work, we do not provide an explicit indication to the model of task boundaries, which is the most general circumstance for a learning agent exposed to continuous experience. While various methods to counteract catastrophic forgetting have recently been proposed, we explore a straightforward, general, and seemingly overlooked solution - that of using experience replay buffers for all past events - with a mixture of on- and off-policy learning, leveraging behavioral cloning. We show that this strategy can still learn new tasks quickly yet can substantially reduce catastrophic forgetting in both Atari and DMLab domains, even matching the performance of methods that require task identities. When buffer storage is constrained, we confirm that a simple mechanism for randomly discarding data allows a limited size buffer to perform almost as well as an unbounded one.""","""This paper and revisions have some interesting insights into using ER for catastrophic forgetting, and comparisons to other methods for reducing catastrophic forgetting. However, the paper is currently pitched as the first to notice that ER can be used for this purpose, whereas it was well explored in the cited paper ""Selective Experience Replay for Lifelong Learning"", 2018. For example, the abstract says ""While various methods to counteract catastrophic forgetting have recently been proposed, we explore a straightforward, general, and seemingly overlooked solution that of using experience replay buffers for all past events"". It seems unnecessary to claim this as a main contribution in this work. Rather, the main contributions seem to be to include behavioural cloning, and do provide further empirical evidence that selective ER can be effective for catastrophic forgetting. Further, to make the paper even stronger, it would be interesting to better understand even smaller replay buffers. A buffer size of 5 million is still quite large. What is a realistic size for continual learning? Hypothesizing how ER can be part of a real continual learning solution, which will likely have more than 3 tasks, is important to understand how to properly restrict the buffer size. Finally, it is recommended to reconsider the strong stance on catastrophic interference and forgetting. Catastrophic interference has been considered for incremental training, where recent updates can interfere with estimates for older (or other values). This definition does not precisely match the provided definition in the paper. Further, it is true that forgetting has often been used explicitly for multiple tasks, trained in sequence; however, the issues are similar (new learning overriding older learning). These two definitions need not be so separate, and further it is not clear that the provided definitions are congruent with older literature on interference. Overall, there is most definitely useful ideas and experiments in this paper, but it is as yet a bit preliminary. Improvements on placement, motivation and experimental choices would make this work much stronger, and provide needed clarity on the use of ER for forgetting."""
415,"""DONT JUDGE A BOOK BY ITS COVER - ON THE DYNAMICS OF RECURRENT NEURAL NETWORKS""",[],"""To be effective in sequential data processing, Recurrent Neural Networks (RNNs) are required to keep track of past events by creating memories. Consequently RNNs are harder to train than their feedforward counterparts, prompting the developments of both dedicated units such as LSTM and GRU and of a handful of training tricks. In this paper, we investigate the effect of different training protocols on the representation of memories in RNN. While reaching similar performance for different protocols, RNNs are shown to exhibit substantial differences in their ability to generalize for unforeseen tasks or conditions. We analyze the dynamics of the networks hidden state, and uncover the reasons for this difference. Each memory is found to be associated with a nearly steady state of the dynamics whose speed predicts performance on unforeseen tasks and which we refer to as a slow point. By tracing the formation of the slow points we are able to understand the origin of differences between training protocols. Our results show that multiple solutions to the same task exist but may rely on different dynamical mechanisms, and that training protocols can bias the choice of such solutions in an interpretable way.""","""This paper analyses the dynamics of RNNs, cq GRU and LSTM. The paper is mostly experimental w.r.t. the difficulty of training RNNs; this is also caused by the fact that the theoretical foundations of the paper seem not to be solid enough. Experimentation with CIFAR10 is not completely stable. The review results make the paper balance at the middle. The merit of the paper for the greater community is doubted, in its current form."""
416,"""Inference of unobserved event streams with neural Hawkes particle smoothing""",[],"""Events that we observe in the world may be caused by other, unobserved events. We consider sequences of discrete events in continuous time. When only some of the events are observed, we propose particle smoothing to infer the missing events. Particle smoothing is an extension of particle filtering in which proposed events are conditioned on the future as well as the past. For our setting, we develop a novel proposal distribution that is a type of continuous-time bidirectional LSTM. We use the sampled particles in an approximate minimum Bayes risk decoder that outputs a single low-risk prediction of the missing events. We experiment in multiple synthetic and real domains, modeling the complete sequences in each domain with a neural Hawkes process (Mei & Eisner, 2017). On held-out incomplete sequences, our method is effective at inferring the ground-truth unobserved events. In particular, particle smoothing consistently improves upon particle filtering, showing the benefit of training a bidirectional proposal distribution.""","""All reviewers agree to reject. While there were many positive points to this work, reviewers believed that it was not yet ready for acceptance."""
417,"""Laplacian Smoothing Gradient Descent""","['Laplacian Smoothing', 'Nonconvex Optimization', 'Deep Learning']","""We propose a class of very simple modifications of gradient descent and stochastic gradient descent. We show that when applied to a large variety of machine learning problems, ranging from softmax regression to deep neural nets, the proposed surrogates can dramatically reduce the variance and improve the generalization accuracy. The methods only involve multiplying the usual (stochastic) gradient by the inverse of a positive definitive matrix coming from the discrete Laplacian or its high order generalizations. The theory of Hamilton-Jacobi partial differential equations demonstrates that the implicit version of new algorithm is almost the same as doing gradient descent on a new function which (i) has the same global minima as the original function and (ii) is ``more convex"". We show that optimization algorithms with these surrogates converge uniformly in the discrete Sobolev pseudo-formula sense and reduce the optimality gap for convex optimization problems. We implement our algorithm into both PyTorch and Tensorflow platforms which only involves changing of a few lines of code. The code will be available on Github.""","""Dear authors, The topic of variance reduction in optimization is timely and the reviewers appreciated your attempt at circumventing the issues faced with the current popular methods. They however had a concern about the significance of the results, which I echo: - First, there have been previous attempts at variance reduction which share some similarity with yours, for instance ""No more pesky learning rate"", ""Topmoumoute online natural gradient algorithm"" or even Adam (which does variance reduction without mentioning it). - The fact that previous similar methods exist is a non-issue should yours perform better. However, the absence of stepsize tuning in the experimental evaluation is a big issue as the performance of an iterative algorithm is highly sensitive to it. Finally, the link between flatness of the minimum and generalization is dubious, as mentioned for instance by Dinh et al. (2017). As a consequence, I cannot accept this work for publication to ICLR but I encourage you to address the points of the reviewers should you wish to resubmit it to a future conference. """
418,"""Measuring Density and Similarity of Task Relevant Information in Neural Representations""","['Neural Networks', 'Representation', 'Information density', 'Transfer Learning']","""Neural models achieve state-of-the-art performance due to their ability to extract salient features useful to downstream tasks. However, our understanding of how this task-relevant information is included in these networks is still incomplete. In this paper, we examine two questions (1) how densely is information included in extracted representations, and (2) how similar is the encoding of relevant information between related tasks. We propose metrics to measure information density and cross-task similarity, and perform an extensive analysis in the domain of natural language processing, using four varieties of sentence representation and 13 tasks. We also demonstrate how the proposed analysis tools can find immediate use in choosing tasks for transfer learning.""","""This paper addresses important general questions about how linear classifiers use features, and about the transferability of those features across tasks. The paper presents a specific new analysis method, and demonstrates it on a family of NLP tasks. All four reviewers (counting the emergency fourth review) found the general direction of research to be interesting and worthwhile, but all four shared several serious concerns about the impact and soundness of the proposed method. The impact concerns mostly dealt with the observation that the method is specific to linear classifiers, and that it's only applicable to tasks for which a substantial amount of training data is available. As the AC, I'm willing to accept that it should still be possible to conduct an informative analysis under these conditions, but I'm more concerned about the soundness issues: The reviewers were not convinced that a method based on the counting of specific features was appropriate for the proposed setting (due to rotation sensitivity, among other issues), and did not find that the experiments were sufficiently extensive to overcome these doubts."""
419,"""On the Sensitivity of Adversarial Robustness to Input Data Distributions""","['adversarial robustness', 'adversarial training', 'PGD training', 'adversarial perturbation', 'input data distribution']","""Neural networks are vulnerable to small adversarial perturbations. Existing literature largely focused on understanding and mitigating the vulnerability of learned models. In this paper, we demonstrate an intriguing phenomenon about the most popular robust training method in the literature, adversarial training: Adversarial robustness, unlike clean accuracy, is sensitive to the input data distribution. Even a semantics-preserving transformations on the input data distribution can cause a significantly different robustness for the adversarial trained model that is both trained and evaluated on the new distribution. Our discovery of such sensitivity on data distribution is based on a study which disentangles the behaviors of clean accuracy and robust accuracy of the Bayes classifier. Empirical investigations further confirm our finding. We construct semantically-identical variants for MNIST and CIFAR10 respectively, and show that standardly trained models achieve comparable clean accuracies on them, but adversarially trained models achieve significantly different robustness accuracies. This counter-intuitive phenomenon indicates that input data distribution alone can affect the adversarial robustness of trained neural networks, not necessarily the tasks themselves. Lastly, we discuss the practical implications on evaluating adversarial robustness, and make initial attempts to understand this complex phenomenon.""","""This paper studies an interesting phenomenon related to adversarial training -- that adversarial robustness is quite sensitive to semantically lossless shifts in input data distribution. Strengths - Characterizes a previously unobserved phenomenon in adversarial training, which is quite relevant to ongoing research in the area. - Interesting and novel theoretical analysis that motivates the relationship between adversarial robustness and the shape of input distribution. Weaknesses - Reviewers pointed out some shortcomings in experiments, and analysis of causes and remedies to adversarial robustness. The authors agree that given the current state of understanding, these are hard questions to pose good answers for. The result and observations by themselves are interesting and useful for the community. The weakness that the paper does not propose a solution for the observed phenomenon remains, but all reviewers agree that the observation in itself is interesting. Therefore, I recommend that the paper be accepted. """
420,"""Trellis Networks for Sequence Modeling""","['sequence modeling', 'language modeling', 'recurrent networks', 'convolutional networks', 'trellis networks']","""We present trellis networks, a new architecture for sequence modeling. On the one hand, a trellis network is a temporal convolutional network with special structure, characterized by weight tying across depth and direct injection of the input into deep layers. On the other hand, we show that truncated recurrent networks are equivalent to trellis networks with special sparsity structure in their weight matrices. Thus trellis networks with general weight matrices generalize truncated recurrent networks. We leverage these connections to design high-performing trellis networks that absorb structural and algorithmic elements from both recurrent and convolutional models. Experiments demonstrate that trellis networks outperform the current state of the art methods on a variety of challenging benchmarks, including word-level language modeling and character-level language modeling tasks, and stress tests designed to evaluate long-term memory retention. The code is available at pseudo-url .""","""The paper proposes a novel network architecture for sequential learning, called trellis networks, which generalizes truncated RNNs and also links them to temporal convnets. The advantages of both types of nets are used to design trellis networks which appear to outperform state of art on several datasets. The paper is well-written and the results are convincing."""
421,"""Security Analysis of Deep Neural Networks Operating in the Presence of Cache Side-Channel Attacks""","['DNN Security Analysis', 'Fingerprinting Attacks', 'Cache Side-Channel']","""Recent work has introduced attacks that extract the architecture information of deep neural networks (DNN), as this knowledge enhances an adversarys capability to conduct attacks on black-box networks. This paper presents the first in-depth security analysis of DNN fingerprinting attacks that exploit cache side-channels. First, we define the threat model for these attacks: our adversary does not need the ability to query the victim model; instead, she runs a co-located process on the host machine victim s deep learning (DL) system is running and passively monitors the accesses of the target functions in the shared framework. Second, we introduce DeepRecon, an attack that reconstructs the architecture of the victim network by using the internal information extracted via Flush+Reload, a cache side-channel technique. Once the attacker observes function invocations that map directly to architecture attributes of the victim network, the attacker can reconstruct the victims entire network architecture. In our evaluation, we demonstrate that an attacker can accurately reconstruct two complex networks (VGG19 and ResNet50) having only observed one forward propagation. Based on the extracted architecture attributes, we also demonstrate that an attacker can build a meta-model that accurately fingerprints the architecture and family of the pre-trained model in a transfer learning setting. From this meta-model, we evaluate the importance of the observed attributes in the fingerprinting process. Third, we propose and evaluate new framework-level defense techniques that obfuscate our attackers observations. Our empirical security analysis represents a step toward understanding the DNNs vulnerability to cache side-channel attacks.""","""The reviewers generally had concerns that the goal of recovering only the model architecture was unmotivated (given that knowing the architecture is not a large threat on its own, and there are existing attacks that work without knowledge of the model architecture). Moreover, given the strength of the assumed attack model, recovering model architecture is a fairly unambitious goal (again, more serious attacks have already been demonstrated under weaker attack models). Finally, though less seriously, the analysis is fairly preliminary, e.g. it is unclear if the attack can generalize to nearby architectures that were outside the training set."""
422,"""Meta-Learning For Stochastic Gradient MCMC""","['Meta Learning', 'MCMC']","""Stochastic gradient Markov chain Monte Carlo (SG-MCMC) has become increasingly popular for simulating posterior samples in large-scale Bayesian modeling. However, existing SG-MCMC schemes are not tailored to any specific probabilistic model, even a simple modification of the underlying dynamical system requires significant physical intuition. This paper presents the first meta-learning algorithm that allows automated design for the underlying continuous dynamics of an SG-MCMC sampler. The learned sampler generalizes Hamiltonian dynamics with state-dependent drift and diffusion, enabling fast traversal and efficient exploration of energy landscapes. Experiments validate the proposed approach on Bayesian fully connected neural network, Bayesian convolutional neural network and Bayesian recurrent neural network tasks, showing that the learned sampler outperforms generic, hand-designed SG-MCMC algorithms, and generalizes to different datasets and larger architectures.""","""This paper proposes to use meta-learning to design MCMC sampling distributions based on Hamiltonian dynamics, aiming to mix faster on set of problems that are related to the training problems. The reviewers agree that the paper is well-written and the ideas are interesting and novel. The main weaknesses of the paper are that (1) there is not a clear case for using this method over SG-HMC, and (2) there are many design choices that are not validated. The authors revised the paper to address some aspects of the latter concern, but are encouraged to add additional revisions to clarify the points brought up by the reviewers. Despite the weaknesses, the reviewers all agree that the paper exceeds the bar for acceptance. I also recommend accept."""
423,"""Detecting Egregious Responses in Neural Sequence-to-sequence Models""","['Deep Learning', 'Natural Language Processing', 'Adversarial Attacks', 'Dialogue Response Generation']","""In this work, we attempt to answer a critical question: whether there exists some input sequence that will cause a well-trained discrete-space neural network sequence-to-sequence (seq2seq) model to generate egregious outputs (aggressive, malicious, attacking, etc.). And if such inputs exist, how to find them efficiently. We adopt an empirical methodology, in which we first create lists of egregious output sequences, and then design a discrete optimization algorithm to find input sequences that will cause the model to generate them. Moreover, the optimization algorithm is enhanced for large vocabulary search and constrained to search for input sequences that are likely to be input by real-world users. In our experiments, we apply this approach to dialogue response generation models trained on three real-world dialogue data-sets: Ubuntu, Switchboard and OpenSubtitles, testing whether the model can generate malicious responses. We demonstrate that given the trigger inputs our algorithm finds, a significant number of malicious sentences are assigned large probability by the model, which reveals an undesirable consequence of standard seq2seq training. ""","""This work examines how to craft adversarial examples that will lead trained seq2seq models to generate undesired outputs (here defined as, assigning higher-than-average probability to undesired outputs). Making a model safe for deployment is an important unsolved problem and this work is looking at it from an interesting angle, and all reviewers agree that the paper is clear, well-presented, and offering useful observations. While the paper does not provide ways to fix the problem of egregious outputs being probable, as pointed out by reviewers, it is still a valuable study of the behavior of trained models and an interesting way to ""probe"" them, that would likely be of high interest to many people at ICLR."""
424,"""Expressiveness in Deep Reinforcement Learning""",[],"""Representation learning in reinforcement learning (RL) algorithms focuses on extracting useful features for choosing good actions. Expressive representations are essential for learning well-performed policies. In this paper, we study the relationship between the state representation assigned by the state extractor and the performance of the RL agent. We observe that representations assigned by the better state extractor are more scattered than which assigned by the worse one. Moreover, RL agents achieving high performances always have high rank matrices which are composed by their representations. Based on our observations, we formally define expressiveness of the state extractor as the rank of the matrix composed by representations. Therefore, we propose to promote expressiveness so as to improve algorithm performances, and we call it Expressiveness Promoted DRL. We apply our method on both policy gradient and value-based algorithms, and experimental results on 55 Atari games show the superiority of our proposed method.""","""The authors propose to define 'Expressiveness' in deep RL by the rank of a matrix comprising a number of feature vectors from propagating observations through the learnt representation, and show a correlation between higher rank and higher performance. They try 3 regularizers to increase rank and show that they improve the final score on Atari games compared to A3C or DQN. The AC and reviewers agree that the paper is interesting and novel and could have general significance for the RL field. Also, the authors were very responsive to the reviewers and added more details, plus several experiments and analyses to support their claims. However, the reviewers were concerned about a number of aspects and have recommended that the authors clean up their presentation and analysis a bit more. In particular, the fact that the regularization coefficient is tuned for each Atari game makes it very hard to compare to DQN/A3C which are very careful to keep the same hyperparameters across every game."""
425,"""ADef: an Iterative Algorithm to Construct Adversarial Deformations""","['Adversarial examples', 'deformations', 'deep neural networks', 'computer vision']","""While deep neural networks have proven to be a powerful tool for many recognition and classification tasks, their stability properties are still not well understood. In the past, image classifiers have been shown to be vulnerable to so-called adversarial attacks, which are created by additively perturbing the correctly classified image. In this paper, we propose the ADef algorithm to construct a different kind of adversarial attack created by iteratively applying small deformations to the image, found through a gradient descent step. We demonstrate our results on MNIST with convolutional neural networks and on ImageNet with Inception-v3 and ResNet-101.""","""The submission proposes a method to construct adversarial attacks based on deforming an input image rather than adding small peturbations. Although deformations can also be characterized by the difference of the original and deformed image, it is qualitatively and quantitatively different as a small deformation can result in a large difference. On the positive side, this paper proposes an interesting form of adversarial attack, whose success can give additional insights on the forms of existing adversarial attacks. The experiments on MNIST and ImageNet are reasonably comprehensive and allow interesting interpretation of how the image deforms to allow the attack. The paper is also praised for its clarity, and cleaner formulation compared to Xiao et al. (see below). Additional experiments during rebuttal phase partially answered reviewer concerns, and provided more information e.g. about the effect of the smoothness of the deformation. There were some concerns that the paper primarly presents one idea, and perhaps missed an opportunity for deeper analysis (R1). R2 would have appreciated more analysis on how to defend against the attack. A controversial point is the relation / novelty with respect to Xiao et al., ICLR 2018. As e.g. pointed out by R1: ""The paper originates from a document provably written in late 2017, which is before the deposit on arXiv of another article (by different authors, early 2018) which was later accepted to ICLR 2018 [Xiao and al.]. This remark is important in that it changes my rating of the paper (being more indulgent with papers proposing new ideas, as otherwise the novelty is rather low compared to [Xiao and al.])."" On the balance, all three reviewers recommended acceptance of the paper. Regarding novelty over Xiao et al., even ignoring the arguable precedence of the current submission, the formulation is cleaner and will likely advance the analysis of adversarial attacks."""
426,"""Stochastic Adversarial Video Prediction""","['video prediction', 'GANs', 'variational autoencoder']","""Being able to predict what may happen in the future requires an in-depth understanding of the physical and causal rules that govern the world. A model that is able to do so has a number of appealing applications, from robotic planning to representation learning. However, learning to predict raw future observations, such as frames in a video, is exceedingly challengingthe ambiguous nature of the problem can cause a naively designed model to average together possible futures into a single, blurry prediction. Recently, this has been addressed by two distinct approaches: (a) latent variational variable models that explicitly model underlying stochasticity and (b) adversarially-trained models that aim to produce naturalistic images. However, a standard latent variable model can struggle to produce realistic results, and a standard adversarially-trained model underutilizes latent variables and fails to produce diverse predictions. We show that these distinct methods are in fact complementary. Combining the two produces predictions that look more realistic to human raters and better cover the range of possible futures. Our method outperforms prior works in these aspects.""","""This paper shows that combining GAN and VAE for video prediction allows to trade off diversity and realism. The paper is well-written and the experimentation is careful, as noted by reviewers. However, reviewers agree that this combination is of limited novelty (having been used for images before). Reviewers also note that the empirical performance is not very much stronger than baselines. Overall, the novelty is too slight and the empirical results are not strong enough compared to baselines to justify acceptance based solely on empirical results."""
427,"""S3TA: A Soft, Spatial, Sequential, Top-Down Attention Model""","['Attention', 'RL', 'Top-Down', 'Interpretability']","""We present a soft, spatial, sequential, top-down attention model (S3TA). This model uses a soft attention mechanism to bottleneck its view of the input. A recurrent core is used to generate query vectors, which actively select information from the input by correlating the query with input- and space-dependent key maps at different spatial locations. We demonstrate the power and interpretabilty of this model under two settings. First, we build an agent which uses this attention model in RL environments and show that we can achieve performance competitive with state-of-the-art models while producing attention maps that elucidate some of the strategies used to solve the task. Second, we use this model in supervised learning tasks and show that it also achieves competitive performance and provides interpretable attention maps that show some of the underlying logic in the model's decision making.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. The paper - tackles an interesting problem - makes a concerted effort to provide qualititative results that give insight into the models behaviour. - sufficiently cites related work. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The model architecture lacks novelty. - There was also agreement that the contributions - (i) minor modifications of existing sequential attention-based models, and (ii) application to the RL domain - are minor. - A lot of space in the paper (section 4.2) is devoted to exploring the use of this model for image classification and video action recognition. However the proposed model performed poorly compared to SOTA methods for this task and no motivation was given for why the proposed model would be useful for such tasks. All three points impacted the final decision. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. There was high agreement between the reviewers on the main drawbacks of the paper, before and after the rebuttal. The AC considered the rebuttals by the authors (in which they argued that there was sufficient contribution) but, in the end, agreed with the reviewers' assessments. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be rejected. """
428,"""Don't let your Discriminator be fooled""","['GAN', 'generative models', 'computer vision']","""Generative Adversarial Networks are one of the leading tools in generative modeling, image editing and content creation. However, they are hard to train as they require a delicate balancing act between two deep networks fighting a never ending duel. Some of the most promising adversarial models today minimize a Wasserstein objective. It is smoother and more stable to optimize. In this paper, we show that the Wasserstein distance is just one out of a large family of objective functions that yield these properties. By making the discriminator of a GAN robust to adversarial attacks we can turn any GAN objective into a smooth and stable loss. We experimentally show that any GAN objective, including Wasserstein GANs, benefit from adversarial robustness both quantitatively and qualitatively. The training additionally becomes more robust to suboptimal choices of hyperparameters, model architectures, or objective functions.""","""The paper provides a simple method for regularising and robustifying GAN training. Always appreciated contribution to GANs. :-)"""
429,"""Non-vacuous Generalization Bounds at the ImageNet Scale: a PAC-Bayesian Compression Approach""","['generalization', 'deep-learning', 'pac-bayes']","""Modern neural networks are highly overparameterized, with capacity to substantially overfit to training data. Nevertheless, these networks often generalize well in practice. It has also been observed that trained networks can often be ``compressed to much smaller representations. The purpose of this paper is to connect these two empirical observations. Our main technical result is a generalization bound for compressed networks based on the compressed size that, combined with off-the-shelf compression algorithms, leads to state-of-the-art generalization guarantees. In particular, we provide the first non-vacuous generalization guarantees for realistic architectures applied to the ImageNet classification problem. Additionally, we show that compressibility of models that tend to overfit is limited. Empirical results show that an increase in overfitting increases the number of bits required to describe a trained network.""","""The paper combines PAC-Bayes bound with network compression to derive a generalization bound for large-scale neural nets such as ImageNet. The approach is novel and interesting and the paper is well-written. The authors provided detailed replies and improvements in response to reviewers questions, and all reviewers agree this is a very nice contribution."""
430,"""A Case for Object Compositionality in Deep Generative Models of Images""","['Objects', 'Compositionality', 'Generative Models', 'GAN', 'Unsupervised Learning']","""Deep generative models seek to recover the process with which the observed data was generated. They may be used to synthesize new samples or to subsequently extract representations. Successful approaches in the domain of images are driven by several core inductive biases. However, a bias to account for the compositional way in which humans structure a visual scene in terms of objects has frequently been overlooked. In this work we propose to structure the generator of a GAN to consider objects and their relations explicitly, and generate images by means of composition. This provides a way to efficiently learn a more accurate generative model of real-world images, and serves as an initial step towards learning corresponding object representations. We evaluate our approach on several multi-object image datasets, and find that the generator learns to identify and disentangle information corresponding to different objects at a representational level. A human study reveals that the resulting generative model is better at generating images that are more faithful to the reference distribution.""","""The paper proposes a generative model that generates one object at a time, and uses a relational network to encode cross-object relationships. Similar object-centric generation and object-object relational network is proposed in ""sequential attend, infer, repeat"" of Kosiorek et al. for video generation, which first appeared on arxiv on June 5th 2018 and was officially accepted in NIPS 2018 before the submission deadline for ICLR 2019. Moreover, several recent generative models have been proposed that consider object-centric biases, which the current paper references but does not compare against, e.g., 'attend, infer, repeat' of Eslami et al., or ""DRAW: A Recurrent Neural Network For Image Generation"" of Gregor et al. . The CLEVR dataset considered, though it contains real images, the intrinsic image complexity is low because it features a small number of objects against table background. As a result, the novelty of the proposed work may not be sufficient in light of recent literature, despite the fact that the paper presents a reasonable and interesting approach for image generation. """
431,"""Advocacy Learning""","['competition', 'supervision', 'deep learning', 'adversarial', 'debate']","""We introduce advocacy learning, a novel supervised training scheme for classification problems. This training scheme applies to a framework consisting of two connected networks: 1) the Advocates, composed of one subnetwork per class, which take the input and provide a convincing class-conditional argument in the form of an attention map, and 2) a Judge, which predicts the inputs class label based on these arguments. Each Advocate aims to convince the Judge that the input example belongs to their corresponding class. In contrast to a standard network, in which all subnetworks are trained to jointly cooperate, we train the Advocates to competitively argue for their class, even when the input belongs to a different class. We also explore a variant, honest advocacy learning, where the Advocates are only trained on data corresponding to their class. Applied to several different classification tasks, we show that advocacy learning can lead to small improvements in classification accuracy over an identical supervised baseline. Through a series of follow-up experiments, we analyze when and how Advocates improve discriminative performance. Though it may seem counter-intuitive, a framework in which subnetworks are trained to competitively provide evidence in support of their class shows promise, performing as well as or better than standard approaches. This provides a foundation for further exploration into the effect of competition and class-conditional representations.""","""The paper presents a novel architecture, reminescent of mixtures-of-experts, composed of a set of advocates networks providing an attention map to a separate ""judge"" network. Reviewers have several concerns, including lack of theoretical justification, potential scaling limitations, and weak experimental results. Authors answered to several of the concerns, which did not convinced reviewers. The reviewer with the highest score was also the least confident, so overall I will recommend to reject the paper."""
432,"""Efficient Dictionary Learning with Gradient Descent""","['dictionary learning', 'nonconvex optimization']","""Randomly initialized first-order optimization algorithms are the method of choice for solving many high-dimensional nonconvex problems in machine learning, yet general theoretical guarantees cannot rule out convergence to critical points of poor objective value. For some highly structured nonconvex problems however, the success of gradient descent can be understood by studying the geometry of the objective. We study one such problem -- complete orthogonal dictionary learning, and provide converge guarantees for randomly initialized gradient descent to the neighborhood of a global optimum. The resulting rates scale as low order polynomials in the dimension even though the objective possesses an exponential number of saddle points. This efficient convergence can be viewed as a consequence of negative curvature normal to the stable manifolds associated with saddle points, and we provide evidence that this feature is shared by other nonconvex problems of importance as well. ""","""It seems that the reviewers reached a consensus that the paper is not ready for publication in ICLR. (see more details in the reviews below. )"""
433,"""Language Modeling with Graph Temporal Convolutional Networks""","['Graph Neural Network', 'Language Modeling', 'Convolution']","""Recently, there have been some attempts to use non-recurrent neural models for language modeling. However, a noticeable performance gap still remains. We propose a non-recurrent neural language model, dubbed graph temporal convolutional network (GTCN), that relies on graph neural network blocks and convolution operations. While the standard recurrent neural network language models encode sentences sequentially without modeling higher-level structural information, our model regards sentences as graphs and processes input words within a message propagation framework, aiming to learn better syntactic information by inferring skip-word connections. Specifically, the graph network blocks operate in parallel and learn the underlying graph structures in sentences without any additional annotation pertaining to structure knowledge. Experiments demonstrate that the model without recurrence can achieve comparable perplexity results in language modeling tasks and successfully learn syntactic information.""","""Though the overall direction is interesting, the reviewers are in consensus that the work is not ready for publication (better / larger scale evaluation is needed, comparison with other non-autoregressive architectures should be provided, esp Transformer as there is a close relation between the methods). """
434,"""On Computation and Generalization of Generative Adversarial Networks under Spectrum Control""",[],"""Generative Adversarial Networks (GANs), though powerful, is hard to train. Several recent works (Brock et al., 2016; Miyato et al., 2018) suggest that controlling the spectra of weight matrices in the discriminator can significantly improve the training of GANs. Motivated by their discovery, we propose a new framework for training GANs, which allows more flexible spectrum control (e.g., making the weight matrices of the discriminator have slow singular value decays). Specifically, we propose a new reparameterization approach for the weight matrices of the discriminator in GANs, which allows us to directly manipulate the spectra of the weight matrices through various regularizers and constraints, without intensively computing singular value decompositions. Theoretically, we further show that the spectrum control improves the generalization ability of GANs. Our experiments on CIFAR-10, STL-10, and ImgaeNet datasets confirm that compared to other competitors, our proposed method is capable of generating images with better or equal quality by utilizing spectral normalization and encouraging the slow singular value decay.""","""All the reviewers agree that the paper has an interesting idea on regularizing the spectral norm of the weight matrices in GANs, and a generalization bound has been shown. The empirical result shows that indeed regularization improves the performance of the GANs. Based on these the AC suggested acceptance. """
435,"""Neural Model-Based Reinforcement Learning for Recommendation""","['Generative adversarial user model', 'Recommendation system', 'combinatorial recommendation policy', 'model-based reinforcement learning', 'deep Q-networks']","""There are great interests as well as many challenges in applying reinforcement learning (RL) to recommendation systems. In this setting, an online user is the environment; neither the reward function nor the environment dynamics are clearly defined, making the application of RL challenging. In this paper, we propose a novel model-based reinforcement learning framework for recommendation systems, where we develop a generative adversarial network to imitate user behavior dynamics and learn her reward function. Using this user model as the simulation environment, we develop a novel DQN algorithm to obtain a combinatorial recommendation policy which can handle a large number of candidate items efficiently. In our experiments with real data, we show this generative adversarial user model can better explain user behavior than alternatives, and the RL policy based on this model can lead to a better long-term reward for the user and higher click rate for the system.""","""This paper formulates the recommendation as a model-based reinforcement learning problem. Major concerns of the paper include: paper writing needs improvement; many decisions in experimental design were not justified; lack of sufficient baselines; results not convincing. Overall, this paper cannot be published in its current form. """
436,"""RelWalk -- A Latent Variable Model Approach to Knowledge Graph Embedding""","['relation representations', 'natural language processing', 'theoretical analysis', 'knowledge graphs']","""Knowledge Graph Embedding (KGE) is the task of jointly learning entity and relation embeddings for a given knowledge graph. Existing methods for learning KGEs can be seen as a two-stage process where (a) entities and relations in the knowledge graph are represented using some linear algebraic structures (embeddings), and (b) a scoring function is defined that evaluates the strength of a relation that holds between two entities using the corresponding relation and entity embeddings. Unfortunately, prior proposals for the scoring functions in the first step have been heuristically motivated, and it is unclear as to how the scoring functions in KGEs relate to the generation process of the underlying knowledge graph. To address this issue, we propose a generative account of the KGE learning task. Specifically, given a knowledge graph represented by a set of relational triples (h, R, t), where the semantic relation R holds between the two entities h (head) and t (tail), we extend the random walk model (Arora et al., 2016a) of word embeddings to KGE. We derive a theoretical relationship between the joint probability p(h, R, t) and the embeddings of h, R and t. Moreover, we show that marginal loss minimisation, a popular objective used by much prior work in KGE, follows naturally from the log-likelihood ratio maximisation under the probabilities estimated from the KGEs according to our theoretical relationship. We propose a learning objective motivated by the theoretical analysis to learn KGEs from a given knowledge graph. The KGEs learnt by our proposed method obtain state-of-the-art performance on FB15K237 and WN18RR benchmark datasets, providing empirical evidence in support of the theory. ""","""This paper proposes a new scoring function for link prediction model that is based on a generative model for the knowledge graph, based on a random-walk model previously used for word embeddings. The new scoring function, as it is accompanied by the generative model, provides interesting theoretical results that the reviewers also appreciate. Finally, the results are quite strong, as they obtain state-of-art on the primary benchmarks for the task. Based on the submitted version, the reviewers and AC note the following potential weaknesses: (1) the reviewers felt that the proposed work is a direct application of the random-walk model from Arora et al. and thus limited in novelty, (2) given the generative model, the reviewers felt that the paper would benefit from an analysis of the learned embeddings, and their difference from ones from existing approaches, (3) The reviewers noted that the authors were using an incorrect version of FB15k and WN18, (4) the authors were not providing results for all the metrics, (5) the coverage of related work is quite limited. The authors addressed many of the concerns raised by the reviewers in their comments and revision, in particular, they obtained state-of-art results for the corrected versions of the benchmarks. Further, they clarified the assumptions made in their modeling and revised the related work to include the papers that the reviewers mentioned. However, the concerns regarding the lack of novelty of the proposed approach, w.r.t Arora et al 2016 and the need for further analysis of the learned embeddings, still remain. This paper comes really close to getting accepted, but ultimately the reviewers agree that the remaining concerns need to be addressed."""
437,"""A Guider Network for Multi-Dual Learning""",[],"""A large amount of parallel data is needed to train a strong neural machine translation (NMT) system. This is a major challenge for low-resource languages. Building on recent work on unsupervised and semi-supervised methods, we propose a multi-dual learning framework to improve the performance of NMT by using an almost infinite amount of available monolingual data and some parallel data of other languages. Since our framework involves multiple languages and components, we further propose a timing optimization method that uses reinforcement learning (RL) to optimally schedule the different components in order to avoid imbalanced training. Experimental results demonstrate the validity of our model, and confirm its superiority to existing dual learning methods.""","""All reviewers agree in their assessment that this paper is not ready for acceptance into ICLR and the authors did not respond during the rebuttal phase."""
438,"""Deep Generative Models for learning Coherent Latent Representations from Multi-Modal Data""","['Multi-Modal Deep Generative Models', 'Sensor Fusion', 'Data Generation', 'VAE']","""The application of multi-modal generative models by means of a Variational Auto Encoder (VAE) is an upcoming research topic for sensor fusion and bi-directional modality exchange. This contribution gives insights into the learned joint latent representation and shows that expressiveness and coherence are decisive properties for multi-modal datasets. Furthermore, we propose a multi-modal VAE derived from the full joint marginal log-likelihood that is able to learn the most meaningful representation for ambiguous observations. Since the properties of multi-modal sensor setups are essential for our approach but hardly available, we also propose a technique to generate correlated datasets from uni-modal ones. ""","""This paper suggests a problem with the standard ELBO for the multi-modal case, and proposes a new objective to address this problem. However, I (and some of the reviewers) disagree with the motivation. First of all, there's no reason one can't train a separate encoder for every combination of modalities available, at least when there are only 2 or 3. Failing that, one could simple optimize per-example approximate posteriors without using an encoder. Second, once you stop optimizing the ELBO, you've lost the motivating principle for training VAEs, and must justify your new objective empirically. Almost all of the results are (in my opinion) ambiguous plots of latent encodings. Finally, a point made throughout the paper and discussions was that different modalities should give the same encodings, which is plainly false. One of the reviewers made this point: ""The fact that z_a != z_b != z_{a,b} should be expected if a and b provide different information. I don't see the problem with this."", which you dismiss. Additionally, the encoder's job is to approximate the true posterior. The true posteriors will in general be different for different modalities. I would recommend focusing on ways to train the original ELBO in the presence of different modalities, instead of modifying it based on these intuitions. """
439,"""Contingency-Aware Exploration in Reinforcement Learning""","['Reinforcement Learning', 'Exploration', 'Contingency-Awareness']","""This paper investigates whether learning contingency-awareness and controllable aspects of an environment can lead to better exploration in reinforcement learning. To investigate this question, we consider an instantiation of this hypothesis evaluated on the Arcade Learning Element (ALE). In this study, we develop an attentive dynamics model (ADM) that discovers controllable elements of the observations, which are often associated with the location of the character in Atari games. The ADM is trained in a self-supervised fashion to predict the actions taken by the agent. The learned contingency information is used as a part of the state representation for exploration purposes. We demonstrate that combining actor-critic algorithm with count-based exploration using our representation achieves impressive results on a set of notoriously challenging Atari games due to sparse rewards. For example, we report a state-of-the-art score of >11,000 points on Montezuma's Revenge without using expert demonstrations, explicit high-level information (e.g., RAM states), or supervisory data. Our experiments confirm that contingency-awareness is indeed an extremely powerful concept for tackling exploration problems in reinforcement learning and opens up interesting research questions for further investigations.""","""The paper addresses the challenging and important problem of exploration in sparse-rewards settings. The authors propose a novel use of contingency awareness, i.e., the agent's understanding of the environment features that are under its direct control, in combination with a count-based approach to exploration. The model is trained using an inverse dynamics model and attention mechanism and is shown to be able to identify the controllable character. The resulting exploration approach achieves strong empirical results compared to alternative count-based exploration techniques. The reviewers note that the novel approach has potential for opening up potential fruitful directions for follow-up research. The obtained strong empirical results are another strong indication of the value of the proposed idea. The reviewers mention several potential weaknesses. First, while the proposed idea is general, the specific implementation seems targetted specifically towards Atari games. While Atari is a popular benchmark domain, this raises questions as to whether insights can be more generally applied. Second, several questions were raised regarding the motivation for some of the presented modeling choices (e.g., loss terms) as well as their impact on the empirical results. Ablation studies were recommended as a step to resolving these questions Reviewer 3 questioned whether the learned state representation could be directly used as an additional input to the agent, and if it would improve performance. Finally, several related works were suggested that should be included in the discussion of related work. The authors carefully addressed the issues raised by the reviewers, running additional comparisons and adding to the original empirical insights. Several issues of clarity were resolved in the paper and in the discussion. Reviewer 3 engaged with the authors and confirmed that they are satisfied with the resulting submission. The AC judges that the suggestions of reviewer 1 have been addressed to a satisfactory level. A remaining issue regarding results reporting was raised anonymously towards the end of the review period, and the AC encourages the authors to address this issue in their camera ready version."""
440,"""Deep Curiosity Search: Intra-Life Exploration Can Improve Performance on Challenging Deep Reinforcement Learning Problems""",[],"""Traditional exploration methods in reinforcement learning (RL) require agents to perform random actions to find rewards. But these approaches struggle on sparse-reward domains like Montezumas Revenge where the probability that any random action sequence leads to reward is extremely low. Recent algorithms have performed well on such tasks by encouraging agents to visit new states or perform new actions in relation to all prior training episodes (which we call across-training novelty). But such algorithms do not consider whether an agent exhibits intra-life novelty: doing something new within the current episode, regardless of whether those behaviors have been performed in previous episodes. We hypothesize that across-training novelty might discourage agents from revisiting initially non-rewarding states that could become important stepping stones later in traininga problem remedied by encouraging intra-life novelty. We introduce Curiosity Search for deep reinforcement learning, or Deep Curiosity Search (DeepCS), which encourages intra-life exploration by rewarding agents for visiting as many different states as possible within each episode, and show that DeepCS matches the performance of current state-of-the-art methods on Montezumas Revenge. We further show that DeepCS improves exploration on Amidar, Freeway, Gravitar, and Tutankham (many of which are hard exploration games). Surprisingly, DeepCS also doubles A2C performance on Seaquest, a game we would not have expected to benefit from intra-life exploration because the arena is small and already easily navigated by naive exploration techniques. In one run, DeepCS achieves a maximum training score of 80,000 points on Seaquesthigher than any methods other than Ape-X. The strong performance of DeepCS on these sparse- and dense-reward tasks suggests that encouraging intra-life novelty is an interesting, new approach for improving performance in Deep RL and motivates further research into hybridizing across-training and intra-life exploration methods.""","""Pros: - novel idea of intra-life curiosity that encourages diverse behavior within each episode rather than across episodes. Cons: - privileged/ad-hoc information (RAM state, distinguishing rooms) - lack of sufficient ablations/analysis - insufficient revision/rebuttal The reviewers reached consensus that the paper should be rejected in its current form."""
441,"""Generating Realistic Stock Market Order Streams""","['application in finance', 'stock markets', 'generative models']","""We propose an approach to generate realistic and high-fidelity stock market data based on generative adversarial networks. We model the order stream as a stochastic process with finite history dependence, and employ a conditional Wasserstein GAN to capture history dependence of orders in a stock market. We test our approach with actual market and synthetic data on a number of different statistics, and find the generated data to be close to real data. ""","""The reviewers raised a number of major concerns including the incremental novelty of the proposed (WGANs are applied to a new domain), and, most importantly, insufficient and unconvincing experimental evaluation presented (including the lack of comparative studies). The authors rebuttal failed to fully alleviate reviewers concerns. Hence, I cannot suggest this paper for presentation at ICLR."""
442,"""Multi-objective training of Generative Adversarial Networks with multiple discriminators""","['Generative Adversarial Networks', 'Multi-objective optimization', 'Generative models']","""Recent literature has demonstrated promising results on the training of Generative Adversarial Networks by employing a set of discriminators, as opposed to the traditional game involving one generator against a single adversary. Those methods perform single-objective optimization on some simple consolidation of the losses, e.g. an average. In this work, we revisit the multiple-discriminator approach by framing the simultaneous minimization of losses provided by different models as a multi-objective optimization problem. Specifically, we evaluate the performance of multiple gradient descent and the hypervolume maximization algorithm on a number of different datasets. Moreover, we argue that the previously proposed methods and hypervolume maximization can all be seen as variations of multiple gradient descent in which the update direction computation can be done efficiently. Our results indicate that hypervolume maximization presents a better compromise between sample quality and diversity, and computational cost than previous methods.""","""The reviewers found that paper is well written, clear and that the authors did a good job placing the work in the relevant literature. The proposed method for using multiple discriminators in a multi-objective setting to train GANs seems interesting and compelling. However, all the reviewers found the paper to be on the borderline. The main concern was the significance of the work in the context of existing literature. Specifically, the reviewers did not find the experimental results significant enough to be convinced that this work presents a major advance in GAN training. """
443,"""Discrete flow posteriors for variational inference in discrete dynamical systems""","['normalising flow', 'variational inference', 'discrete latent variable']","""Each training step for a variational autoencoder (VAE) requires us to sample from the approximate posterior, so we usually choose simple (e.g. factorised) approximate posteriors in which sampling is an efficient computation that fully exploits GPU parallelism. However, such simple approximate posteriors are often insufficient, as they eliminate statistical dependencies in the posterior. While it is possible to use normalizing flow approximate posteriors for continuous latents, there is nothing analogous for discrete latents. The most natural approach to model discrete dependencies is an autoregressive distribution, but sampling from such distributions is inherently sequential and thus slow. We develop a fast, parallel sampling procedure for autoregressive distributions based on fixed-point iterations which enables efficient and accurate variational inference in discrete state-space models. To optimize the variational bound, we considered two ways to evaluate probabilities: inserting the relaxed samples directly into the pmf for the discrete distribution, or converting to continuous logistic latent variables and interpreting the K-step fixed-point iterations as a normalizing flow. We found that converting to continuous latent variables gave considerable additional scope for mismatch between the true and approximate posteriors, which resulted in biased inferences, we thus used the former approach. We tested our approach on the neuroscience problem of inferring discrete spiking activity from noisy calcium-imaging data, and found that it gave accurate connectivity estimates in an order of magnitude less time.""","""The paper presents an original approach to replace inefficient discrete autoregressive posterior sampling by a parallel sampling procedure based on fixed-point iterations reminiscent of normalizing flow, but for discrete variables. All reviewers liked the idea, and found that it was an original and promising approach. But all agreed the paper was poorly written and very unclear. All also found the experimental section lacking, in clarity and scope. Authors did not provide a rebuttal. Overall a potentially really promising idea, but the paper is not yet ripe. """
444,"""Unsupervised Multi-Target Domain Adaptation: An Information Theoretic Approach""",[],"""Unsupervised domain adaptation (uDA) models focus on pairwise adaptation settings where there is a single, labeled, source and a single target domain. However, in many real-world settings one seeks to adapt to multiple, but somewhat similar, target domains. Applying pairwise adaptation approaches to this setting may be suboptimal, as they would fail to leverage shared information among the multiple domains. In this work we propose an information theoretic approach for domain adaptation in the novel context of multiple target domains with unlabeled instances and one source domain with labeled instances. Our model aims to find a shared latent space common to all domains, while simultaneously accounting for the remaining private, domain-specific factors. Disentanglement of shared and private information is accomplished using a unified information-theoretic approach, which also serves to provide a stronger link between the latent representations and the observed data. The resulting single model, accompanied by an efficient optimization algorithm, allows simultaneous adaptation from a single source to multiple target domains. We test our approach on three publicly-available datasets, showing that it outperforms several popular domain adaptation methods.""","""The paper proposes the unique setting of adapting to multiple target domains. The idea being that their approach may leverage commonality across domains to improve adaptation while maintaining domain specific parameters where needed. This idea and general approach is interesting and worth exploring. The authors' rebuttal and paper edits significantly improved the draft and clarified some details missing from the original presentation. There is an ablation study showing that each part of the model contributes to the overall performance. However, the approach provides only modest improvements over comparative methods which were not designed to learn from multiple target domains. In addition, comparison against the latest approaches is missing so it is likely that the performance reported here is below state-of-the-art. Overall, given the modest experimental gains combined with incremental improvement over single source information theoretic methods, this paper is not yet ready for publication."""
445,"""Purchase as Reward : Session-based Recommendation by Imagination Reconstruction""","['recommender systems', 'reinforcement learning', 'predictive learning', 'self-supervised RL', 'model-based planning']","""One of the key challenges of session-based recommender systems is to enhance users purchase intentions. In this paper, we formulate the sequential interactions between user sessions and a recommender agent as a Markov Decision Process (MDP). In practice, the purchase reward is delayed and sparse, and may be buried by clicks, making it an impoverished signal for policy learning. Inspired by the prediction error minimization (PEM) and embodied cognition, we propose a simple architecture to augment reward, namely Imagination Reconstruction Network (IRN). Specically, IRN enables the agent to explore its environment and learn predictive representations via three key components. The imagination core generates predicted trajectories, i.e., imagined items that users may purchase. The trajectory manager controls the granularity of imagined trajectories using the planning strategies, which balances the long-term rewards and short-term rewards. To optimize the action policy, the imagination-augmented executor minimizes the intrinsic imagination error of simulated trajectories by self-supervised reconstruction, while maximizing the extrinsic reward using model-free algorithms. Empirically, IRN promotes quicker adaptation to user interest, and shows improved robustness to the cold-start scenario and ultimately higher purchase performance compared to several baselines. Somewhat surprisingly, IRN using only the purchase reward achieves excellent next-click prediction performance, demonstrating that the agent can ""guess what you like"" via internal planning.""","""This paper addresses the problem of recommendations within user sessions from a reinforcement learning perspective. The problem is naturally modeled as an RL problem, given its sequential nature and inherent uncertainty of any model over user preferences. The problem suffers from delayed and sparse rewards, which the authors propose to address using self-supervised prediction. The approach is empirically validated in a simulated setting, using data from the 2015 ACM RecSys Challenge. The reviewers and AC note that the problem studied is an important application area where RL has high potential to improve over current research results and industry practice. The proposed idea is interesting, and the strong empirical evaluation on a publicly available data set is highlighted. R1 also commends the authors' decision to address the challenging cold-start problem. The reviewers and AC also note several potential weaknesses. The choice of addressing the problem from a reinforcement learning perspective is not clearly motivated. This is needed, as many supervised learning (and other types) approaches to the problem exist. A performance comparison to current state-of-the-art RL baselines is missing. The proposed approach is related to both imagination augmented (I2A, Racaniere et al. 2017) and agents with auxiliary rewards (UNREAL, Jaderberg et al. 2016), but does not compare to either method. Neither does the related work section sufficiently clarify why the proposed approach is expected to improve over these prior approaches. A thorough comparison to these baselines in a real-world application like session-based recommendation would be a strong contribution in itself, but without the contributions of the paper are hard to assess. Reviewers also noted lack of clarity. Some concerns are addressed by the authors, but the consensus is that the paper would benefit from a major revision to clearly work out the method, as well as it's conceptual and empirical differences from existing reinforcement learning approaches. R3 mentions missing related work, some of which the authors include in the revision. The AC recommends also following up on references in cited papers to ensure a future revision of the paper is well placed in the context of prior work on recommender systems, especially when modeled as a reinforcement learning problem. Overall, the paper was assessed as borderline by the reviewers. The ACs view is that there are too many concerns for acceptance at ICLR in the present form, and that the paper will benefit from a thorough revision."""
446,"""Low Latency Privacy Preserving Inference""","['privacy', 'classification', 'homomorphic encryption', 'neural networks']","""When applying machine learning to sensitive data one has to balance between accuracy, information leakage, and computational-complexity. Recent studies have shown that Homomorphic Encryption (HE) can be used for protecting against information leakage while applying neural networks. However, this comes with the cost of limiting the kind of neural networks that can be used (and hence the accuracy) and with latency of the order of several minutes even for relatively simple networks. In this study we improve on previous results both in the kind of networks that can be applied and in terms of the latency. Most of the improvement is achieved by novel ways to represent the data to make better use of the capabilities of the encryption scheme.""","""The paper proposes improvements on the area of neural network inference with homomorphically encrypted data. Existing applications typically have high computational cost, and this paper provides some solutions to these problems. Some of the improvements are due to better ""engineering"" (the use of the faster SAEL 3.2.1 over CryptoNet). The idea of using pre-trained AlexNet features is new, but pretty standard practice. The presentation has been greatly improved in the updated version, however the paper could benefit from additional discussions and experiments. For example, when a practitioner wants to solve a new problem with some design need (e.g. accuracy, latency vs. bandwidth trade-off), what network modules should be used and how should they be represented? To summarize, the problem considered is important, however, as pointed out by the reviewers, both the empirical and the theoretical results appear to be incremental with respect to the existing literature. """
447,"""Shallow Learning For Deep Networks""","['CNN', 'greedy learning']","""Shallow supervised 1-hidden layer neural networks have a number of favorable properties that make them easier to interpret, analyze, and optimize than their deep counterparts, but lack their representational power. Here we use 1-hidden layer learning problems to sequentially build deep networks layer by layer, which can inherit properties from shallow networks. Contrary to previous approaches using shallow networks, we focus on problems where deep learning is reported as critical for success. We thus study CNNs on image recognition tasks using the large-scale Imagenet dataset and the CIFAR-10 dataset. Using a simple set of ideas for architecture and training we find that solving sequential 1-hidden-layer auxiliary problems leads to a CNN that exceeds AlexNet performance on ImageNet. Extending our training methodology to construct individual layers by solving 2-and-3-hidden layer auxiliary problems, we obtain an 11-layer network that exceeds VGG-11 on ImageNet obtaining 89.8% top-5 single crop. To our knowledge, this is the first competitive alternative to end-to-end training of CNNs that can scale to ImageNet. We conduct a wide range of experiments to study the properties this induces on the intermediate layers.""","""The paper discusses layer-wise training of deep networks. The authors show that it's possible to achieve reasonable performance by training deep nets layer by layer, as opposed to now widely adopted end-to-end training. While such a training procedure is not novel, the authors argue that this is an interesting result, considering that such a training procedure is often dismissed as sub-optimal and leading to inferior results. However, the results show exactly that, as the performance of the models is significantly worse than the state of the art, and it is unclear what other advantages such a training scheme can offer. The authors mention that layer-wise training could be useful for theoretical understanding of deep nets, but they dont really perform such analysis in this submission, and its also unclear whether conclusions of such analysis would extend to deep nets trained end-to-end. In its current form, the paper is not ready for acceptance. I encourage the authors to make a more clear case for the method: either by improving results to match end-to-end training, or by actually demonstrating that layer-wise training has certain advantages over end-to-end learning. """
448,"""Residual Non-local Attention Networks for Image Restoration""","['Non-local network', 'attention network', 'image restoration', 'residual learning']","""In this paper, we propose a residual non-local attention network for high-quality image restoration. Without considering the uneven distribution of information in the corrupted images, previous methods are restricted by local convolutional operation and equal treatment of spatial- and channel-wise features. To address this issue, we design local and non-local attention blocks to extract features that capture the long-range dependencies between pixels and pay more attention to the challenging parts. Specifically, we design trunk branch and (non-)local mask branch in each (non-)local attention block. The trunk branch is used to extract hierarchical features. Local and non-local mask branches aim to adaptively rescale these hierarchical features with mixed attentions. The local mask branch concentrates on more local structures with convolutional operations, while non-local attention considers more about long-range dependencies in the whole feature map. Furthermore, we propose residual local and non-local attention learning to train the very deep network, which further enhance the representation ability of the network. Our proposed method can be generalized for various image restoration applications, such as image denoising, demosaicing, compression artifacts reduction, and super-resolution. Experiments demonstrate that our method obtains comparable or better results compared with recently leading methods quantitatively and visually.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - strong qualitative and quantitative results - a good ablative analysis of the proposed method. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - clarity could be improved (and was much improved in the revision). - somewhat limited novelty. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. No major points of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be accepted. """
449,"""A CASE STUDY ON OPTIMAL DEEP LEARNING MODEL FOR UAVS""","['Energy Efficiency', 'Autonomous Flying', 'Trail Detection']","""Over the passage of time Unmanned Autonomous Vehicles (UAVs), especially Autonomous flying drones grabbed a lot of attention in Artificial Intelligence. Since electronic technology is getting smaller, cheaper and more efficient, huge advancement in the study of UAVs has been observed recently. From monitoring floods, discerning the spread of algae in water bodies to detecting forest trail, their application is far and wide. Our work is mainly focused on autonomous flying drones where we establish a case study towards efficiency, robustness and accuracy of UAVs where we showed our results well supported through experiments. We provide details of the software and hardware architecture used in the study. We further discuss about our implementation algorithms and present experiments that provide a comparison between three different state-of-the-art algorithms namely TrailNet, InceptionResnet and MobileNet in terms of accuracy, robustness, power consumption and inference time. In our study, we have shown that MobileNet has produced better results with very less computational requirement and power consumption. We have also reported the challenges we have faced during our work as well as a brief discussion on our future work to improve safety features and performance.""","""The paper compared between different CNNs for UAV trail guidance. The reviewers arrived at a consensus on rejection due to lack of new ideas, and the paper is not well polished. """
450,"""Understand the dynamics of GANs via Primal-Dual Optimization""","['non-convex optimization', 'generative adversarial network', 'primal dual algorithm']","""Generative adversarial network (GAN) is one of the best known unsupervised learning techniques these days due to its superior ability to learn data distributions. In spite of its great success in applications, GAN is known to be notoriously hard to train. The tremendous amount of time it takes to run the training algorithm and its sensitivity to hyper-parameter tuning have been haunting researchers in this area. To resolve these issues, we need to first understand how GANs work. Herein, we take a step toward this direction by examining the dynamics of GANs. We relate a large class of GANs including the Wasserstein GANs to max-min optimization problems with the coupling term being linear over the discriminator. By developing new primal-dual optimization tools, we show that, with a proper stepsize choice, the widely used first-order iterative algorithm in training GANs would in fact converge to a stationary solution with a sublinear rate. The same framework also applies to multi-task learning and distributional robust learning problems. We verify our analysis on numerical examples with both synthetic and real data sets. We hope our analysis shed light on future studies on the theoretical properties of relevant machine learning problems.""","""The paper studies the convergence of a primal-dual algorithm on a special min-max problem in WGAN where the maximization is with respect to linear variables (linear discriminator) and minimization is over non-convex generators. Experiments with both simulated and real world data are conducted to show that the algorithm works for WGANs and multi-task learning. The major concern of reviewers lies in that the linear discriminator assumption in WGAN is too restrictive to general non-convex mini-max saddle point problem in GANs. Linear discriminator implies that the maximization part in min-max problem is concave, and it is thus not surprise that under this assumption the paper converts the original problem to a non-convex optimization instance and proves its first order convergence with descent lemma. This technique however cant be applied to general non-convex saddle point problem in GANs. Also the experimental studies are also not strong enough. Therefore, current version of the paper is proposed as borderline lean reject. """
451,"""Graph Learning Network: A Structure Learning Algorithm""","['graph prediction', 'graph structure learning', 'graph neural network']","""Graph prediction methods that work closely with the structure of the data, e.g., graph generation, commonly ignore the content of its nodes. On the other hand, the solutions that consider the nodes information, e.g., classification, ignore the structure of the whole. And some methods exist in between, e.g., link prediction, but predict the structure piece-wise instead of considering the graph as a whole. We hypothesize that by jointly predicting the structure of the graph and its nodes features, we can improve both tasks. We propose the Graph Learning Network (GLN), a simple yet effective process to learn node embeddings and structure prediction functions. Our model uses graph convolutions to propose expected node features, and predict the best structure based on them. We repeat these steps sequentially to enhance the prediction and the embeddings. In contrast to existing generation methods that rely only on the structure of the data, we use the feature on the nodes to predict better relations, similar to what link prediction methods do. However, we propose an holistic approach to process the whole graph for our predictions. Our experiments show that our method predicts consistent structures across a set of problems, while creating meaningful node embeddings.""","""The paper addresses an important problem of supervised learning for predicting graph connectivity using both node features and the overall graph structure. The paper is clearly written, and the presented approach produces promising results on synthetic data. However, all reviewers agree that the paper could be improved by including more comparison with prior art and related work discussion, and strengthening empirical results by including real-life data and more through evaluation; they also find the novelty and significance of the proposed approach somewhat limited. We hope the authors will use the suggestions of the reviewers to further improve the paper. """
452,"""Preventing Posterior Collapse with delta-VAEs""","['Posterior Collapse', 'VAE', 'Autoregressive Models']","""Due to the phenomenon of posterior collapse, current latent variable generative models pose a challenging design choice that either weakens the capacity of the decoder or requires altering the training objective. We develop an alternative that utilizes the most powerful generative models as decoders, optimize the variational lower bound, and ensures that the latent variables preserve and encode useful information. Our proposed -VAEs achieve this by constraining the variational family for the posterior to have a minimum distance to the prior. For sequential latent variable models, our approach resembles the classic representation learning approach of slow feature analysis. We demonstrate our methods efficacy at modeling text on LM1B and modeling images: learning representations, improving sample quality, and achieving state of the art log-likelihood on CIFAR-10 and ImageNet 32 32.""","""Strengths: The proposed method is relatively principled. The paper also demonstrates a new ability: training VAEs with autoregressive decoders that have meaningful latents. The paper is clear and easy to read. Weaknesses: I wasn't entirely convinced by the causal/anticausal formulation, and it's a bit unfortunate that the decoder couldn't have been copied without modification from another paper. Points of contention: It's not clear how general the proposed approach is, or how important the causal/anti-causal idea was, although the authors added an ablation study to check this last question. Consensus: All reviewers rated the paper above the bar, and the objections of the two 6's seem to have been satisfactorily addressed by the rebuttal and paper update."""
453,"""InfoBot: Transfer and Exploration via the Information Bottleneck""","['Information bottleneck', 'policy transfer', 'policy generalization', 'exploration']","""A central challenge in reinforcement learning is discovering effective policies for tasks where rewards are sparsely distributed. We postulate that in the absence of useful reward signals, an effective exploration strategy should seek out {\it decision states}. These states lie at critical junctions in the state space from where the agent can transition to new, potentially unexplored regions. We propose to learn about decision states from prior experience. By training a goal-conditioned model with an information bottleneck, we can identify decision states by examining where the model accesses the goal state through the bottleneck. We find that this simple mechanism effectively identifies decision states, even in partially observed settings. In effect, the model learns the sensory cues that correlate with potential subgoals. In new environments, this model can then identify novel subgoals for further exploration, guiding the agent through a sequence of potential decision states and through new regions of the state space.""","""The paper presents the use of information bottlenecks as a way to identify key ""decision states"" in exploration, in a goal-conditioned model. The concept of ""decision states"" is actually common in RL, states where exploring can lead to very diverse/new states. The implementation of the ""information bottleneck"" is done by adding a regularizing term, the conditional mutual information I(A;G|S). The main weaknesses of the paper were its lack of clarity and the experimental section. It seems to me that the rebuttals, and the additional experiments and details, made the paper worthy of publication. The authors cleared enough of the gray areas and showcased the relative merits of the methods."""
454,"""Dimension-Free Bounds for Low-Precision Training""","['low precision', 'stochastic gradient descent']","""Low-precision training is a promising way of decreasing the time and energy cost of training machine learning models. Previous work has analyzed low-precision training algorithms, such as low-precision stochastic gradient descent, and derived theoretical bounds on their convergence rates. These bounds tend to depend on the dimension of the model pseudo-formula in that the number of bits needed to achieve a particular error bound increases as pseudo-formula increases. This is undesirable because a motivating application for low-precision training is large-scale models, such as deep learning, where pseudo-formula can be huge. In this paper, we prove dimension-independent bounds for low-precision training algorithms that use fixed-point arithmetic, which lets us better understand what affects the convergence of these algorithms as parameters scale. Our methods also generalize naturally to let us prove new convergence bounds on low-precision training with other quantization schemes, such as low-precision floating-point computation and logarithmic quantization.""","""As the reviewers pointed out, the strength of the paper mostly comes from the analysis of the non-linear quantization which depends on the double log of the Lipschitz constants and other parameters. The AC and reviewers agree with the dimension-independent nature of the bounds, but also note that dimension-independent gound may not necessarily be significantly stronger than the dimension-dependent bounds as the metric of measuring the difficulty of the problem also matters. Although the paper does seem to lack result that shows the empirical benefit of the non-linear quantization. In considering the author response and reviewer comments, the AC decided that this comparison was indeed important for understanding the contribution in this work, and it is difficult to assess the scope of the contribution without such a comparison. """
455,"""Variational Domain Adaptation""","['domain adaptation', 'variational inference', 'multi-domain']","""This paper proposes variational domain adaptation, a unified, scalable, simple framework for learning multiple distributions through variational inference. Unlike the existing methods on domain transfer through deep generative models, such as StarGAN (Choi et al., 2017) and UFDN (Liu et al., 2018), the variational domain adaptation has three advantages. Firstly, the samples from the target are not required. Instead, the framework requires one known source as a prior pseudo-formula and binary discriminators, pseudo-formula , discriminating the target domain pseudo-formula from others. Consequently, the framework regards a target as a posterior that can be explicitly formulated through the Bayesian inference, \propto p(\mathcal{D}_i|x)p(x) as exhibited by a further proposed model of dual variational autoencoder (DualVAE). Secondly, the framework is scablable to large-scale domains. As well as VAE encodes a sample pseudo-formula as a mode on a latent space: \in \mathcal{Z} DualVAE encodes a domain pseudo-formula as a mode on the dual latent space \in \mathcal{Z}^* named domain embedding. It reformulates the posterior with a natural paring \rangle: \mathcal{Z} \times \mathcal{Z}^* \rightarrow \Real$, which can be expanded to uncountable infinite domains such as continuous domains as well as interpolation. Thirdly, DualVAE fastly converges without sophisticated automatic/manual hyperparameter search in comparison to GANs as it requires only one additional parameter to VAE. Through the numerical experiment, we demonstrate the three benefits with multi-domain image generation task on CelebA with up to 60 domains, and exhibits that DualVAE records the state-of-the-art performance outperforming StarGAN and UFDN.""","""This paper proposes using conditional VAEs for multi-domain transfer and presents results on CelebA and SCUT. As mentioned by reviewers, the presentation and clarity of the work could be improved. It is quite difficult to determine the new/proposed aspects of the work from a first read through. Though we recognize and appreciate that the authors updated their manuscript to improve its clarity, another edit pass with particular focus on clarifying prior work on conditional VAEs and their proposed new application to domain transfer would be beneficial. In addition, as DIS is the main metric for comparison to prior work and for evaluation of the final approach, the conclusions about the effectiveness of this method would be easier to see if a more detailed description of the metric and analysis of the results were provided. Given the limited technical novelty and discussion amongst reviewers of the desire for more experimental evidence, this work is not quite ready for publication."""
456,"""CBOW Is Not All You Need: Combining CBOW with the Compositional Matrix Space Model""","['Text representation learning', 'Sentence embedding', 'Efficient training scheme', 'word2vec']","""Continuous Bag of Words (CBOW) is a powerful text embedding method. Due to its strong capabilities to encode word content, CBOW embeddings perform well on a wide range of downstream tasks while being efficient to compute. However, CBOW is not capable of capturing the word order. The reason is that the computation of CBOW's word embeddings is commutative, i.e., embeddings of XYZ and ZYX are the same. In order to address this shortcoming, we propose a learning algorithm for the Continuous Matrix Space Model, which we call Continual Multiplication of Words (CMOW). Our algorithm is an adaptation of word2vec, so that it can be trained on large quantities of unlabeled text. We empirically show that CMOW better captures linguistic properties, but it is inferior to CBOW in memorizing word content. Motivated by these findings, we propose a hybrid model that combines the strengths of CBOW and CMOW. Our results show that the hybrid CBOW-CMOW-model retains CBOW's strong ability to memorize word content while at the same time substantially improving its ability to encode other linguistic information by 8%. As a result, the hybrid also performs better on 8 out of 11 supervised downstream tasks with an average improvement of 1.2%.""","""This paper presents CMOWan unsupervised sentence representation learning method that treats sentences as the product of their word matrices. This method is not entirely novel, as the authors acknowledge, but it has not been successfully applied to downstream tasks before. This paper presents methods for successfully training it, and shows results on the SentEval benchmark suite for sentence representations and an associated set of analysis tasks. All three reviewers agree that the results are unimpressive: CMOW is no better than the faster CBOW baseline on most tasks, and the combination of the two is only marginally better than CBOW. However, CMOW does show some real advantages on the analysis tasks. No reviewer has any major correctness concerns that I can see. As I see it, this paper is borderline, but narrowly worth accepting: As a methods paper, it presents weak results, and it's not likely that many practitioners will leap to use the method. However, the method is so appealingly simple and well known that there is some value in seeing this as an analysis paper that thoroughly evaluates it. Because it is so simple, it will likely be of interest to researchers beyond just the NLP domain in which it is tested (as CBOW-style models have been), so ICLR seems like an appropriate venue. It seems like it's in the community's best interest to see a method like this be evaluated, and since this paper appears to offer a thorough and sound evaluation, I recommend acceptance."""
457,"""Temporal Gaussian Mixture Layer for Videos""",[],"""We introduce a new convolutional layer named the Temporal Gaussian Mixture (TGM) layer and present how it can be used to efficiently capture longer-term temporal information in continuous activity videos. The TGM layer is a temporal convolutional layer governed by a much smaller set of parameters (e.g., location/variance of Gaussians) that are fully differentiable. We present our fully convolutional video models with multiple TGM layers for activity detection. The experiments on multiple datasets including Charades and MultiTHUMOS confirm the effectiveness of TGM layers, outperforming the state-of-the-arts.""","""The reviewers raised a number of major concerns including lack of explanations, lack of baseline comparisons, and lack of discussion on pros and cons of the main contribution of this work -- the presented Temporal Gaussian Mixture (TGM) layer. The authors rebuttal addressed some of the reviewers comments but failed to address all concerns (especially when it comes to the success of TGMs; it remains unclear whether this could be attributed solely to the way TGMs are applied rather than to their fundamental methodological advantage). Having said that, I cannot suggest this paper for presentation at ICLR."""
458,"""Uncertainty-guided Lifelong Learning in Bayesian Networks""","['lifelong learning', 'continual learning', 'sequential learning']","""Sequentially learning of tasks arriving in a continuous stream is a complex problem and becomes more challenging when the model has a fixed capacity. Lifelong learning aims at learning new tasks without forgetting previously learnt ones as well as freeing up capacity for learning future tasks. We argue that identifying the most influential parameters in a representation learned for one task plays a critical role to decide on \textit{what to remember} for continual learning. Motivated by the statistically-grounded uncertainty defined in Bayesian neural networks, we propose to formulate a Bayesian lifelong learning framework, \texttt{BLLL}, that addresses two lifelong learning directions: 1) completely eliminating catastrophic forgetting using weight pruning, where a hard selection mask freezes the most certain parameters (\texttt{BLLL-PRN}) and 2) reducing catastrophic forgetting by adaptively regularizing the learning rates using the parameter uncertainty (\texttt{BLLL-REG}). While \texttt{BLLL-PRN} is by definition a zero-forgetting guaranteed method, \texttt{BLLL-REG}, despite exhibiting some small forgetting, is a task-agnostic lifelong learner, which does not require to know when a new task arrives. This feature makes \texttt{BLLL-REG} a more convenient candidate for applications such as robotics or on-line learning in which such information is not available. We evaluate our Bayesian learning approaches extensively on diverse object classification datasets in short and long sequences of tasks and perform superior or marginally better than the existing approaches.""","""Reviewers are in a consensus and recommended to reject. However, the reviewers did not engage at all with the authors, and did not acknowledge whether their concerns have been answered. I therefore lean to reject, and would recommend the authors to resubmit. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit. """
459,"""Continual Learning via Explicit Structure Learning""","['continuous learning', 'catastrophic forgetting', 'architecture learning']","""Despite recent advances in deep learning, neural networks suffer catastrophic forgetting when tasks are learned sequentially. We propose a conceptually simple and general framework for continual learning, where structure optimization is considered explicitly during learning. We implement this idea by separating the structure and parameter learning. During structure learning, the model optimizes for the best structure for the current task. The model learns when to reuse or modify structure from previous tasks, or create new ones when necessary. The model parameters are then estimated with the optimal structure. Empirically, we found that our approach leads to sensible structures when learning multiple tasks continuously. Additionally, catastrophic forgetting is also largely alleviated from explicit learning of structures. Our method also outperforms all other baselines on the permuted MNIST and split CIFAR datasets in continual learning setting.""","""The paper presents a promising approach for continual learning with no access to data from the previous tasks. For learning the current task, the authors propose to find an optimal structure of the neural network model first (select either to reuse, adapt previously learned layers or to train new layers) and then to learn its parameters. While acknowledging the originality of the method and the importance of the problem that it tries to address, all reviewers and AC agreed that they would like to see more intensive empirical evaluations and comparisons to state-of-the-art models for continual learning using more datasets and in-depth analysis of the results see details comments of all reviewers before and after rebuttal. The authors have tried to address some of these concerns during rebuttal, but an in-depth analysis of the results (evaluation in terms on accuracy, efficiency, memory demand) using different datasets still remains a critical issue. Two other requests to further strengthen the manuscript: 1) an ablation study on the three choices for structural learning (R3), and especially the importance of adaptation (R3 and R1) The authors have tried to address this verbally in their responses but a proper ablation study would be desirable to strengthen the evaluation. 2) Readability and proofreading of the manuscript is still unsatisfying after revision. """
460,"""Diminishing Batch Normalization""","['deep learning', 'learning theory', 'convergence analysis', 'batch normalization']","""In this paper, we propose a generalization of the BN algorithm, diminishing batch normalization (DBN), where we update the BN parameters in a diminishing moving average way. Batch normalization (BN) is very effective in accelerating the convergence of a neural network training phase that it has become a common practice. Our proposed DBN algorithm remains the overall structure of the original BN algorithm while introduces a weighted averaging update to some trainable parameters. We provide an analysis of the convergence of the DBN algorithm that converges to a stationary point with respect to trainable parameters. Our analysis can be easily generalized for original BN algorithm by setting some parameters to constant. To the best knowledge of authors, this analysis is the first of its kind for convergence with Batch Normalization introduced. We analyze a two-layer model with arbitrary activation function. The primary challenge of the analysis is the fact that some parameters are updated by gradient while others are not. The convergence analysis applies to any activation function that satisfies our common assumptions. For the analysis, we also show the sufficient and necessary conditions for the stepsizes and diminishing weights to ensure the convergence. In the numerical experiments, we use more complex models with more layers and ReLU activation. We observe that DBN outperforms the original BN algorithm on Imagenet, MNIST, NI and CIFAR-10 datasets with reasonable complex FNN and CNN models.""","""The paper introduces a modification of batch normalization technique. In contrast to the original batch normalization that normalizes minibatch examples using their mean and standard deviation, this modification uses weighted average of mean and standard deviation from the current and all previous minibatches. The authors then provide some theoretical justification for the superiority of their variant of BatchNorm. Unfortunately, the empirical demonstration of the improved performance seems not sufficient and thus fairly unconvincing. """
461,"""A Multi-modal one-class generative adversarial network for anomaly detection in manufacturing""","['Anomaly detection', 'one-class model', 'GAN']","""One class anomaly detection on high-dimensional data is one of the critical issue in both fundamental machine learning research area and manufacturing applica- tions. A good anomaly detection should accurately discriminate anomalies from normal data. Although most previous anomaly detection methods achieve good performances, they do not perform well on high-dimensional imbalanced data- set 1) with a limited amount of data; 2) multi-modal distribution; 3) few anomaly data. In this paper, we develop a multi-modal one-class generative adversarial net- work based detector (MMOC-GAN) to distinguish anomalies from normal data (products). Apart from a domain-specific feature extractor, our model leverage a generative adversarial network(GAN). The generator takes in a modified noise vector using a pseudo latent prior and generate samples at the low-density area of the given normal data to simulate the anomalies. The discriminator then is trained to distinguish the generate samples from the normal samples. Since the generated samples simulate the low density area for each modal, the discriminator could directly detect anomalies from normal data. Experiments demonstrate that our model outperforms the state-of-the-art one-class classification models and other anomaly detection methods on both normal data and anomalies accuracy, as well as the F1 score. Also, the generated samples can fully capture the low density area of different types of products. ""","""The authors propose a GAN-based anomaly detection method based on simulating anomalies (low density regions of the data space) in order to train an anomaly classifier. While the paper addresses an interesting take on an important problem, there are many concerns raised by reviewers including novelty, clarity, attribution, reproducibility, the use of exclusively proprietary data, and a multitude of textual mistakes. Overall, the paper shows promise but does not seem to be a mature and polished piece of work. As there has been no rebuttal or update to the paper I have no choice but to concur with the reviewers' initial assessments and reject."""
462,"""LEARNING FACTORIZED REPRESENTATIONS FOR OPEN-SET DOMAIN ADAPTATION""",['Open Set Domain Adaptation'],"""Domain adaptation for visual recognition has undergone great progress in the past few years. Nevertheless, most existing methods work in the so-called closed-set scenario, assuming that the classes depicted by the target images are exactly the same as those of the source domain. In this paper, we tackle the more challenging, yet more realistic case of open-set domain adaptation, where new, unknown classes can be present in the target data. While, in the unsupervised scenario, one cannot expect to be able to identify each specific new class, we aim to automatically detect which samples belong to these new classes and discard them from the recognition process. To this end, we rely on the intuition that the source and target samples depicting the known classes can be generated by a shared subspace, whereas the target samples from unknown classes come from a different, private subspace. We therefore introduce a framework that factorizes the data into shared and private parts, while encouraging the shared representation to be discriminative. Our experiments on standard benchmarks evidence that our approach significantly outperforms the state-of-the-art in open-set domain adaptation.""","""This paper proposes a new approach to domain adaptation based on sub-spacing, such that outliers are filtered out. While similar ideas have been used e.g. in multi-view learning, their application to domain adaptation makes it a novel and interesting approach. While the above is considered by the AC an adequate contribution to ICLR, the authors are encouraged to investigate further the implications of the assumptions made, in a way that the derived criteria seem less heuristic, as R1 pointed out. There had been some concerns regarding the experiments, but the authors have been very active in the rebuttal period and addressed these concerns satisfactorily. """
463,"""Learning Global Additive Explanations for Neural Nets Using Model Distillation""","['global interpretability', 'additive explanations', 'model distillation', 'neural nets', 'tabular data']","""Interpretability has largely focused on local explanations, i.e. explaining why a model made a particular prediction for a sample. These explanations are appealing due to their simplicity and local fidelity. However, they do not provide information about the general behavior of the model. We propose to leverage model distillation to learn global additive explanations that describe the relationship between input features and model predictions. These global explanations take the form of feature shapes, which are more expressive than feature attributions. Through careful experimentation, we show qualitatively and quantitatively that global additive explanations are able to describe model behavior and yield insights about models such as neural nets. A visualization of our approach applied to a neural net as it is trained is available at pseudo-url""","""This paper introduces a distillation approach for black-box classifiers that trains generalized additive models (GAM), an additive model over feature shapes, thus providing global explanations for the model. Given the importance of interpretability, the reviewers appreciated the focus of this work. The reviewers also found the experiments, both on real and synthetic datasets, extremely thorough and were impressed by the results. Finally, they also mentioned that the paper was clearly well-written. The reviewers and AC note the following potential weaknesses: (1) The primary concern, raised by all of the reviewers, is the lack of novelty;the proposed approach is a straightforward application of GAMs to model distillation, where black box output is the training data of the GAM, (2) The reviewers are also concern that the proposed approach is limited in scope to tabular datasets, and would not work for more interesting, complex domains like text or images, and (3) The reviewers are concerned that the interpretability of GAMS is assumed, without describing the limitations, for example, if there are correlated features, the shapes would affect each other in uninterpretable ways. Amongst other concerns, the reviewers were concerned about the formatting of the plots and tables in the paper, which made it difficult to read them, and the lack of a user study to verify the interpretability claims. In response to these criticisms, the authors provided comments and a substantial revision to the papers, heavily restructuring the paper to fit extra experiments (comparison to other global explanation techniques, including a user study) and make the figures and tables readable. While the paper was much improved by these changes, and two of the reviewers increased their scores accordingly, concerns about the limited novelty and scope still remained. Ultimately, the reviewers did not reach a conclusion, but the concerns of novelty and scope overwhelmed the clear benefits of the approach and the strong results. This paper was very close to getting accepted, and we strongly urge the authors to submit it to other premier ML conferences."""
464,"""ACIQ: Analytical Clipping for Integer Quantization of neural networks""","['quantization', 'reduced precision', 'training', 'inference', 'activation']","""We analyze the trade-off between quantization noise and clipping distortion in low precision networks. We identify the statistics of various tensors, and derive exact expressions for the mean-square-error degradation due to clipping. By optimizing these expressions, we show marked improvements over standard quantization schemes that normally avoid clipping. For example, just by choosing the accurate clipping values, more than 40\% accuracy improvement is obtained for the quantization of VGG-16 to 4-bits of precision. Our results have many applications for the quantization of neural networks at both training and inference time. ""","""The paper describes a clipping method to improve the performance of quantization. The reviewers have a consensus on rejection due to the contribution is not significant. """
465,"""Text Embeddings for Retrieval from a Large Knowledge Base""","['Text Embeddings', 'Document Ranking', 'Improving Retrieval', 'Question-Answering', 'Learning to Rank']","""Text embedding representing natural language documents in a semantic vector space can be used for document retrieval using nearest neighbor lookup. In order to study the feasibility of neural models specialized for retrieval in a semantically meaningful way, we suggest the use of the Stanford Question Answering Dataset (SQuAD) in an open-domain question answering context, where the first task is to find paragraphs useful for answering a given question. First, we compare the quality of various text-embedding methods on the performance of retrieval and give an extensive empirical comparison on the performance of various non-augmented base embedding with, and without IDF weighting. Our main results are that by training deep residual neural models specifically for retrieval purposes can yield significant gains when it is used to augment existing embeddings. We also establish that deeper models are superior to this task. The best base baseline embeddings augmented by our learned neural approach improves the top-1 recall of the system by 14% in terms of the question side, and by 8% in terms of the paragraph side.""","""I have to agree with the reviewers here and unfortunately recommend a rejection. The methodology and task are not clear. Authors have reformulated QA in SQUAD as as ranking and never compared the results of the proposed model with other QA systems. If authors want to solve a pure ranking problem why they do not compare their methods with other ranking methods/datasets."""
466,"""RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space""","['knowledge graph embedding', 'knowledge graph completion', 'adversarial sampling']","""We study the problem of learning representations of entities and relations in knowledge graphs for predicting missing links. The success of such a task heavily relies on the ability of modeling and inferring the patterns of (or between) the relations. In this paper, we present a new approach for knowledge graph embedding called RotatE, which is able to model and infer various relation patterns including: symmetry/antisymmetry, inversion, and composition. Specifically, the RotatE model defines each relation as a rotation from the source entity to the target entity in the complex vector space. In addition, we propose a novel self-adversarial negative sampling technique for efficiently and effectively training the RotatE model. Experimental results on multiple benchmark knowledge graphs show that the proposed RotatE model is not only scalable, but also able to infer and model various relation patterns and significantly outperform existing state-of-the-art models for link prediction.""","""This paper proposes a knowledge graph completion approach that represents relations as rotations in a complex space; an idea that the reviewers found quite interesting and novel. The authors provide analysis to show how this model can capture symmetry/assymmetry, inversions, and composition. The authors also introduce a separate contribution of self-adversarial negative sampling, which, combined with complex rotational embeddings, obtains state of the art results on the benchmarks for this task. The reviewers and the AC identified a number of potential weaknesses in the initial paper: (1) the evaluation only showed the final performance of the approach, and thus it was not clear how much benefit was obtained from adversarial sampling vs the scoring model, or further, how good the results would be for the baselines if the same sampling was used, (2) citation and comparison to a closely related approach (TorusE), and (3) a number of presentation issues early on in the paper. The reviewers appreciated the author's comments and the revision, which addressed all of the concerns by including (1) additional experiments to performance with and without self-adversarial sampling, and comparisons to TorusE, (2) improved presentation. With the revision, the reviewers agreed that this is a worthy paper to include in the conference. """
467,"""Graph Matching Networks for Learning the Similarity of Graph Structured Objects""","['Similarity learning', 'structured objects', 'graph matching networks']","""This paper addresses the challenging problem of retrieval and matching of graph structured objects, and makes two key contributions. First, we demonstrate how Graph Neural Networks (GNN), which have emerged as an effective model for various supervised prediction problems defined on structured data, can be trained to produce embedding of graphs in vector spaces that enables efficient similarity reasoning. Second, we propose a novel Graph Matching Network model that, given a pair of graphs as input, computes a similarity score between them by jointly reasoning on the pair through a new cross-graph attention-based matching mechanism. We demonstrate the effectiveness of our models on different domains including the challenging problem of control-flow-graph based function similarity search that plays an important role in the detection of vulnerabilities in software systems. The experimental analysis demonstrates that our models are not only able to exploit structure in the context of similarity learning but they can also outperform domain-specific baseline systems that have been carefully hand-engineered for these problems.""","""This is a tough choice as it is a reasonably strong paper. I am similar to another reviewer quite confused how this graph matching can ""only focus on important nodes in the graph"" This seems counter-intuitive and the only reason given in the rebuttal is that other people have done it also.. Relatedly: ""In graph matching, we not only care about the overall similarity of two graphs but also are interested in finding the correspondence between the nodes of two graphs"" I am sorry for the authors and hope they will get it accepted at the next conference."""
468,"""Boosting Robustness Certification of Neural Networks""","['Robustness certification', 'Adversarial Attacks', 'Abstract Interpretation', 'MILP Solvers', 'Verification of Neural Networks']","""We present a novel approach for the certification of neural networks against adversarial perturbations which combines scalable overapproximation methods with precise (mixed integer) linear programming. This results in significantly better precision than state-of-the-art verifiers on challenging feedforward and convolutional neural networks with piecewise linear activation functions.""","""The paper addresess an important problem of neural net robustness verification, and presents a novel approach outperforming state of art; author provided details rebuttals which clarified their contributions over the state of art and highlighted scalability; this work appears to be a solid and useful contribution to the field. """
469,"""Dynamic Planning Networks""","['reinforcement learning', 'planning', 'deep learning']","""We introduce Dynamic Planning Networks (DPN), a novel architecture for deep reinforcement learning, that combines model-based and model-free aspects for online planning. Our architecture learns to dynamically construct plans using a learned state-transition model by selecting and traversing between simulated states and actions to maximize valuable information before acting. In contrast to model-free methods, model-based planning lets the agent efficiently test action hypotheses without performing costly trial-and-error in the environment. DPN learns to efficiently form plans by expanding a single action-conditional state transition at a time instead of exhaustively evaluating each action, reducing the required number of state-transitions during planning by up to 96%. We observe various emergent planning patterns used to solve environments, including classical search methods such as breadth-first and depth-first search. Learning To Plan shows improved data efficiency, performance, and generalization to new and unseen domains in comparison to several baselines.""",""" pros: - Good quantitative results showing clear improvement over other model-based methods in sample efficiency and computational cost (though see Reviewer 2's concerns about the need for more experiments on computational cost). - Cool qualitative results showing discovery of BFS and DFS - Potentially novel approach (see cons) cons: - Lack of clarity especially concerning equation (1). Both Reviewers 1 and 3 were unsure of the rationale for this equation which lies at the heart of the method. It looks to me like a combination of surprise and value but the motivation is not clear. There are a number of other such places pointed out by the reviewers where model choices were made that seem ad hoc or not well motivated. - In general it's hard to understand which factors are important in driving the results you report. As Reviewer 3 points out, more ablation studies and analysis would help here. Providing more motivation, explanation and analysis would help the reader understand better the reasons for the performance of the model. The results are nice and the method is intriguing. I think this potentially a very nice paper and if you can address the above concerns but isn't quite up to the acceptance bar for ICLR this year. """
470,"""The effectiveness of layer-by-layer training using the information bottleneck principle""",[],"""The recently proposed information bottleneck (IB) theory of deep nets suggests that during training, each layer attempts to maximize its mutual information (MI) with the target labels (so as to allow good prediction accuracy), while minimizing its MI with the input (leading to effective compression and thus good generalization). To date, evidence of this phenomenon has been indirect and aroused controversy due to theoretical and practical complications. In particular, it has been pointed out that the MI with the input is theoretically innite in many cases of interest, and that the MI with the target is fundamentally difcult to estimate in high dimensions. As a consequence, the validity of this theory has been questioned. In this paper, we overcome these obstacles by two means. First, as previously suggested, we replace the MI with the input by a noise-regularized version, which ensures it is nite. As we show, this modied penalty in fact acts as a form of weight decay regularization. Second, to obtain accurate (noise regularized) MI estimates between an intermediate representation and the input, we incorporate the strong prior-knowledge we have about their relation, into the recently proposed MI estimator of Belghazi et al. (2018). With this scheme, we are able to stably train each layer independently to explicitly optimize the IB functional. Surprisingly, this leads to enhanced prediction accuracy, thus directly validating the IB theory of deep nets for the rst time.""","""This paper does two things. First, it proposes an approach to estimating the mutual information between the input, X, or target label, Y, and an internal representation in a deep neural network, L, using MINE (for I(Y;L)) or a variation on MINE (for I(X;L)) and noise regularization (estimating I(X;L+), where is isotropic Gaussian white noise) to avoid the problem that I(X;L) is infinite for deterministic networks and continuous X. Second, it attempts to validate the information bottleneck theory of deep learning (Tishby and Zaslavsky, 2015) by exploring an approach to training DNNs that optimizes the information bottleneck Lagrangian, I(Y;L) I(X;L+), layerwise instead of using cross-entropy and backpropagation. Experiments on MNIST and CIFAR-10 show improvements for the layerwise training over cross-entropy training. The penalty on I(X;L+) is described as being analogous to weight decay. The reviewers raised a number of concerns about the paper, the most serious of which is that the claim that the layerwise training results validate the information bottleneck theory of deep learning is too strong. In the AC's opinion, R1's critique that ""[i]f the true mutual information is infinite and the noise regularized estimator is only meant for comparative purposes, why then are the results of the training trajectories interpreted so literally as estimates of the true mutual information?"" is critical, and the authors' reply that ""this quantity is in fact a more appropriate measure for compactness or complexity than the mutual information itself"" undermines their claim that they are validating the information bottleneck theory of deep nets because the information bottleneck theory claims to be using mutual information. The AC also suggests that if the authors wish to continue this work and submit it to another venue, they (1) discuss the fact that MINE estimates only a lower bound that may be quite loose in practice and (2) say in their experimental section whether or not the variance of the regularizing noise was tuned as a hyperparameter, and if so, how results varied with different amounts of noise. Finally, the AC regrets that only one reviewer participated in the discussion (in a very minimal way), despite the reviewers' receiving several reminders that the discussion is a defining feature of the ICLR review process."""
471,"""Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design ""","['Deep Generative Models', 'Normalizing Flows', 'RealNVP', 'Density Estimation']","""Flow-based generative models are powerful exact likelihood models with efficient sampling and inference. Despite their computational efficiency, flow-based models generally have much worse density modeling performance compared to state-of-the-art autoregressive models. In this paper, we investigate and improve upon three limiting design choices employed by flow-based models in prior work: the use of uniform noise for dequantization, the use of inexpressive affine flows, and the use of purely convolutional conditioning networks in coupling layers. Based on our findings, we propose Flow++, a new flow-based model that is now the state-of-the-art non-autoregressive model for unconditional density estimation on standard image benchmarks. Our work has begun to close the significant performance gap that has so far existed between autoregressive models and flow-based models.""","""Strengths: -------------- This paper was clearly written, contained novel technical insights, and had SOTA results. In particular, the explanation of the generalized dequantization trick was enlightening and I expect will be useful in this entire family of methods. The paper also contained ablation experiments. Weaknesses: ------------------ The paper went for a grab-bag approach, when it might have been better to focus on one contribution and explore it in more detail (e.g. show that the learned pdf is smoother when using variational quantization, or showing the different in ELBO when using uniform q as suggested by R2). Also, the main text contains many references to experiments that hadn't converged at submission time, but the submission wasn't updated during the initial discussion period. Why not? Points of contention: ----------------------------- Everyone agrees that the contributions are novel and useful. The only question is whether the exposition is detailed enough to reproduce the new methods (the authors say they will provide code), and whether the experiments, which meet basic standards, of a high enough standard for publication, because there was little investigation into the causes of the difference in performance between models. Consensus: ---------------- The consensus was that this paper was slightly below the bar."""
472,"""Boosting Trust Region Policy Optimization by Normalizing flows Policy""","['Reinforcement Learning', 'Normalizing Flows']","""We propose to improve trust region policy search with normalizing flows policy. We illustrate that when the trust region is constructed by KL divergence constraint, normalizing flows policy can generate samples far from the 'center' of the previous policy iterate, which potentially enables better exploration and helps avoid bad local optima. We show that normalizing flows policy significantly improves upon factorized Gaussian policy baseline, with both TRPO and ACKTR, especially on tasks with complex dynamics such as Humanoid.""","""This work proposes to improve trust region policy search (TRPO) by using normalizing flow policies. This idea is a straightforward combination of two existing techniques and is not super surprising in terms of novelty. In this case, really strong experiments are needed to support the work; this is , unfortunately, is the not the case. For example, it was notice by the reviewers that the Mujoco TRPO experiments does not use the best implementation of TRPO, which makes it difficult to judge the strength of the work compared with state of the art. """
473,"""Von Mises-Fisher Loss for Training Sequence to Sequence Models with Continuous Outputs""","['Language Generation', 'Regression', 'Word Embeddings', 'Machine Translation']","""The Softmax function is used in the final layer of nearly all existing sequence-to-sequence models for language generation. However, it is usually the slowest layer to compute which limits the vocabulary size to a subset of most frequent types; and it has a large memory footprint. We propose a general technique for replacing the softmax layer with a continuous embedding layer. Our primary innovations are a novel probabilistic loss, and a training and inference procedure in which we generate a probability distribution over pre-trained word embeddings, instead of a multinomial distribution over the vocabulary obtained via softmax. We evaluate this new class of sequence-to-sequence models with continuous outputs on the task of neural machine translation. We show that our models obtain upto 2.5x speed-up in training time while performing on par with the state-of-the-art models in terms of translation quality. These models are capable of handling very large vocabularies without compromising on translation quality. They also produce more meaningful errors than in the softmax-based models, as these errors typically lie in a subspace of the vector space of the reference translations.""","""this is a meta-review with the recommendation, but i will ultimately leave the final call to the programme chairs, as this submission has a number of valid concerns. the proposed approach is one of the early, principled one to using (fixed) dense vectors for computing the predictive probability without resorting to softmax, that scales better than and work almost as well as softmax in neural sequence modelling. the reviewers as well as public commentators have noticed some (potentially significant) short comings, such as instability of learning due to numerical precision and the inability of using beam search (perhaps due to the sub-optimal calibration of probabilities under vMF.) however, i believe these two issues should be addressed as separate follow-up work not necessarily by the authors themselves but by a broader community who would find this approach appealing for their own work, which would only be possible if the authors presented this work and had a chance to discuss it with the community at the conference. therefore, i recommend it be accepted. """
474,"""Stochastic Prediction of Multi-Agent Interactions from Partial Observations""","['Dynamics modeling', 'partial observations', 'multi-agent interactions', 'predictive models']","""We present a method which learns to integrate temporal information, from a learned dynamics model, with ambiguous visual information, from a learned vision model, in the context of interacting agents. Our method is based on a graph-structured variational recurrent neural network, which is trained end-to-end to infer the current state of the (partially observed) world, as well as to forecast future states. We show that our method outperforms various baselines on two sports datasets, one based on real basketball trajectories, and one generated by a soccer game engine.""","""This paper proposes a unified approach for performing state estimation and future forecasting for agents interacting within a multi-agent system. The method relies on a graph-structured recurrent neural network trained on temporal and visual (pixel) information. The paper is well-written, with a convincing motivation and a set of novel ideas. The reviewers pointed to a few caveats in the methodology, such as quality of trajectories (AnonReviewer2) and expensive learning of states (AnonReviewer3). However, these issues do not discount much of the papers' quality. Besides, the authors have rebutted satisfactorily some of those comments. More importantly, all three reviewers were not convinced by the experimental evaluation. AnonReviewer1 believes that the idea has a lot of potential, but is hindered by the insufficient exposition of the experiments. AnonReviewer3 similarly asks for more consistency in the experiments. Overall, all reviewers agree on a score ""marginally above the threshold"". While this is not a particularly strong score, the AC weighted all opinions that, despite some caveats, indicate that the developed model and considered application fit nicely in a coherent and convincing story. The authors are strongly advised to work further on the experimental section (which they already started doing as is evident from the rebuttal) to further improve their paper."""
475,"""Attention, Learn to Solve Routing Problems!""","['learning', 'routing problems', 'heuristics', 'attention', 'reinforce', 'travelling salesman problem', 'vehicle routing problem', 'orienteering problem', 'prize collecting travelling salesman problem']","""The recently presented idea to learn heuristics for combinatorial optimization problems is promising as it can save costly development. However, to push this idea towards practical implementation, we need better models and better ways of training. We contribute in both directions: we propose a model based on attention layers with benefits over the Pointer Network and we show how to train this model using REINFORCE with a simple baseline based on a deterministic greedy rollout, which we find is more efficient than using a value function. We significantly improve over recent learned heuristics for the Travelling Salesman Problem (TSP), getting close to optimal results for problems up to 100 nodes. With the same hyperparameters, we learn strong heuristics for two variants of the Vehicle Routing Problem (VRP), the Orienteering Problem (OP) and (a stochastic variant of) the Prize Collecting TSP (PCTSP), outperforming a wide range of baselines and getting results close to highly optimized and specialized algorithms.""","""The paper presents a new deep learning approach for combinatorial optimization problems based on the Transformer architecture. The paper is well written and several experiments are provided. A reviewer asked for more intuition to the proposed approach and authors have responded accordingly. Reviewers are also concerned with scalability and theoretical basis. Overall, all reviewers were positives in their scores, and I recommend accepting the paper."""
476,"""Composing Complex Skills by Learning Transition Policies""","['reinforcement learning', 'hierarchical reinforcement learning', 'continuous control', 'modular framework']","""Humans acquire complex skills by exploiting previously learned skills and making transitions between them. To empower machines with this ability, we propose a method that can learn transition policies which effectively connect primitive skills to perform sequential tasks without handcrafted rewards. To efficiently train our transition policies, we introduce proximity predictors which induce rewards gauging proximity to suitable initial states for the next skill. The proposed method is evaluated on a set of complex continuous control tasks in bipedal locomotion and robotic arm manipulation which traditional policy gradient methods struggle at. We demonstrate that transition policies enable us to effectively compose complex skills with existing primitive skills. The proposed induced rewards computed using the proximity predictor further improve training efficiency by providing more dense information than the sparse rewards from the environments. We make our environments, primitive skills, and code public for further research at pseudo-url .""","""Strengths: The paper tackles a novel, well-motivated problem related to options & HRL. The problem is that of learning transition policies, and the paper proposes a novel and simple solution to that problem, using learned proximity predictors and transition policies that can leverage those. Solid evaluations are done on simulated locomotion and manipulation tasks. The paper is well written. Weaknesses: Limitations were not originally discussed in any depth. There is related work related to sub-goal generation in HRL. AC: The physics of the 2D walker simulations looks to be unrealistic; the character seems to move in a low-gravity environment, and can lean forwards at extreme angles without falling. It would be good to see this explained. There is a consensus among reviewers and AC that the paper would make an excellent ICLR contribution. AC: I suggest a poster presentation; it could also be considered for oral presentation based on the very positive reception by reviewers."""
477,"""RedSync : Reducing Synchronization Traffic for Distributed Deep Learning""","['Data parallel', 'Deep Learning', 'Multiple GPU system', 'Communication Compression', 'Sparsification', 'Quantization']","""Data parallelism has become a dominant method to scale Deep Neural Network (DNN) training across multiple nodes. Since the synchronization of the local models or gradients can be a bottleneck for large-scale distributed training, compressing communication traffic has gained widespread attention recently. Among several recent proposed compression algorithms, Residual Gradient Compression (RGC) is one of the most successful approaches---it can significantly compress the transmitting message size (0.1% of the gradient size) of each node and still preserve accuracy. However, the literature on compressing deep networks focuses almost exclusively on achieving good compression rate, while the efficiency of RGC in real implementation has been less investigated. In this paper, we develop an RGC method that achieves significant training time improvement in real-world multi-GPU systems. Our proposed RGC system design called RedSync, introduces a set of optimizations to reduce communication bandwidth while introducing limited overhead. We examine the performance of RedSync on two different multiple GPU platforms, including a supercomputer and a multi-card server. Our test cases include image classification on Cifar10 and ImageNet, and language modeling tasks on Penn Treebank and Wiki2 datasets. For DNNs featured with high communication to computation ratio, which has long been considered with poor scalability, RedSync shows significant performance improvement.""","""This paper proposed Residual Gradient Compression as a promising approach to reduce the synchronization cost of gradients in a distributed settings. It provides a useful approach that works for a number of models. The reviewers have a consensus that the quality is below acceptance standard due to practicality of experiments and lack of contribution."""
478,"""Image Score: how to select useful samples""",[],"""There has long been debates on how we could interpret neural networks and understand the decisions our models make. Specifically, why deep neural networks tend to be error-prone when dealing with samples that output low softmax scores. We present an efficient approach to measure the confidence of decision-making steps by statistically investigating each unit's contribution to that decision. Instead of focusing on how the models react on datasets, we study the datasets themselves given a pre-trained model. Our approach is capable of assigning a score to each sample within a dataset that measures the frequency of occurrence of that sample's chain of activation. We demonstrate with experiments that our method could select useful samples to improve deep neural networks in a semi-supervised leaning setting.""","""Reviewers are in full agreement for rejection."""
479,"""signSGD with Majority Vote is Communication Efficient and Fault Tolerant""","['large-scale learning', 'distributed systems', 'communication efficiency', 'convergence rate analysis', 'robust optimisation']","""Training neural networks on large datasets can be accelerated by distributing the workload over a network of machines. As datasets grow ever larger, networks of hundreds or thousands of machines become economically viable. The time cost of communicating gradients limits the effectiveness of using such large machine counts, as may the increased chance of network faults. We explore a particularly simple algorithm for robust, communication-efficient learning---signSGD. Workers transmit only the sign of their gradient vector to a server, and the overall update is decided by a majority vote. This algorithm uses 32x less communication per iteration than full-precision, distributed SGD. Under natural conditions verified by experiment, we prove that signSGD converges in the large and mini-batch settings, establishing convergence for a parameter regime of Adam as a byproduct. Aggregating sign gradients by majority vote means that no individual worker has too much power. We prove that unlike SGD, majority vote is robust when up to 50% of workers behave adversarially. The class of adversaries we consider includes as special cases those that invert or randomise their gradient estimate. On the practical side, we built our distributed training system in Pytorch. Benchmarking against the state of the art collective communications library (NCCL), our framework---with the parameter server housed entirely on one machine---led to a 25% reduction in time for training resnet50 on Imagenet when using 15 AWS p3.2xlarge machines.""","""The Reviewers noticed that the paper undergone many editions and raise concern about the content. They encourage improving experimental section further and strengthening the message of the paper. """
480,"""Learning Procedural Abstractions and Evaluating Discrete Latent Temporal Structure""","['learning procedural abstractions', 'latent variable modeling', 'evaluation criteria']","""Clustering methods and latent variable models are often used as tools for pattern mining and discovery of latent structure in time-series data. In this work, we consider the problem of learning procedural abstractions from possibly high-dimensional observational sequences, such as video demonstrations. Given a dataset of time-series, the goal is to identify the latent sequence of steps common to them and label each time-series with the temporal extent of these procedural steps. We introduce a hierarchical Bayesian model called Prism that models the realization of a common procedure across multiple time-series, and can recover procedural abstractions with supervision. We also bring to light two characteristics ignored by traditional evaluation criteria when evaluating latent temporal labelings (temporal clusterings) -- segment structure, and repeated structure -- and develop new metrics tailored to their evaluation. We demonstrate that our metrics improve interpretability and ease of analysis for evaluation on benchmark time-series datasets. Results on benchmark and video datasets indicate that Prism outperforms standard sequence models as well as state-of-the-art techniques in identifying procedural abstractions.""","""While the reviews of this paper were somewhat mixed (7,6,4), I ended up favoring acceptance because of the thorough author responses, and the novelty of what is being examined. The reviewer with a score of 4, argues that this work is not a good fit for iclr, but, although tailoring new metrics may not be a common area that is explored, I don't believe that it's outside the range of iclr's interest, and therefore also more unique."""
481,"""A Priori Estimates of the Generalization Error for Two-layer Neural Networks""","['Over-parameterization', 'A priori estimates', 'Path norm', 'Neural networks', 'Generalization error', 'Approximation error']","""New estimates for the generalization error are established for a nonlinear regression problem using a two-layer neural network model. These new estimates are a priori in nature in the sense that the bounds depend only on some norms of the underlying functions to be fitted, not the parameters in the model. In contrast, most existing results for neural networks are a posteriori in nature in the sense that the bounds depend on some norms of the model parameters. The error rates are comparable to that of the Monte Carlo method in terms of the size of the dataset. Moreover, these bounds are equally effective in the over-parametrized regime when the network size is much larger than the size of the dataset. ""","""I enjoyed reading the paper myself and agree with some of the criticisms raised by the reviewers, but not all of them. In particular, I don't think it's a major issues that this work studies an explicit regularization scheme BECAUSE the state of our understanding of generalization in deep learning is so embarrassingly poor!! Unlike a lot of work, this work is engaging with the *approximation* error and developing risk bounds (called ""generalization error"" here ... not my favorite term for the risk!) rather than just controlling the generalization gap. The simple proof in the bounded noiseless case was nice to see. On the other hand, not being familiar with the work of Klusowski and Barron (2016), I'm not willing to overrule the reviewers on judgments that this work is not novel enough. I would suggest the authors take control of this and paint a more detailed picture of how these two bodies of work relate, including how the proof techniques and arguments overlap. Some other comments: 1. your theorem requires \lambda > 4, but then you're using \lambda = 0.1. this seems problematic to me. 2. your ""nonvacuous upper bound"" is path-norm/sqrt(n) ... but do the numbers in the table include the constants? looking at the constants that are likely to show up, (4Bn sqrt(2 log 2d), they are easily contributing a factor greater than 10 which would make these bounds vacuous as well. you need to explain how you are calculating these numbers more carefully. 3. several times Arora et al and Neyshabur et al are cited when reference is being made to numerical experiments to show that existing bounds are vacuously large. But Dziugaite and Roy, who you cite for the term ""nonvacuous"", made an earlier analysis of path-norm bounds in their appendix and point out that they are vacuous. 4. the paper does not really engage with the fact that you are unlikely to be exactly minimizing the functional J. any hope of bridging this gap? 5. the experiments in general are a bit too vaguely described. also, you control squared error but then only report classification error. would be interested to see both."""
482,"""Dual Skew Divergence Loss for Neural Machine Translation""",[],"""For neural sequence model training, maximum likelihood (ML) has been commonly adopted to optimize model parameters with respect to the corresponding objective. However, in the case of sequence prediction tasks like neural machine translation (NMT), training with the ML-based cross entropy loss would often lead to models that overgeneralize and plunge into local optima. In this paper, we propose an extended loss function called dual skew divergence (DSD), which aims to give a better tradeoff between generalization ability and error avoidance during NMT training. Our empirical study indicates that switching to DSD loss after the convergence of ML training helps the model skip the local optimum and stimulates a stable performance improvement. The evaluations on WMT 2014 English-German and English-French translation tasks demonstrate that the proposed loss indeed helps bring about better translation performance than several baselines.""","""This paper proposes a new loss function that can be used in place of the standard maximum likelihood objective in training NMT models. This leads to a small improvement in training MT systems. There were some concerns about the paper though: one was that the method itself seemed somewhat heuristic without a clear mathematical explanation. The second was that the baselines seemed relatively dated, although one reviewer noted that this seemed like a bit of a lesser concern. Finally, the improvements afforded were relatively small. Given the high number of good papers submitted to ICLR this year, it seems that this one falls short of the acceptance threshold."""
483,"""Gradient-based Training of Slow Feature Analysis by Differentiable Approximate Whitening""","['Slow Feature Analysis', 'Deep Learning', 'Spectral Embedding', 'Temporal Coherence']","""We propose Power Slow Feature Analysis, a gradient-based method to extract temporally slow features from a high-dimensional input stream that varies on a faster time-scale, as a variant of Slow Feature Analysis (SFA). While displaying performance comparable to hierarchical extensions to the SFA algorithm, such as Hierarchical Slow Feature Analysis, for a small number of output-features, our algorithm allows fully differentiable end-to-end training of arbitrary differentiable approximators (e.g., deep neural networks). We provide experimental evidence that PowerSFA is able to extract meaningful and informative low-dimensional features in the case of (a) synthetic low-dimensional data, (b) visual data, and also for (c) a general dataset for which symmetric non-temporal relations between points can be defined.""","""This paper proposes to unroll power iterations within a Slow-Feature-Analysis learning objective in order to obtain a fully differentiable slow feature learning system. Experiments on several datasets are reported. This is a borderline submissions, with reviewers torn between acceptance and rejection. They were generally positive about the clarity and simplicity of the presentation, whereas they raised concerns about the relative lack of novelty (especially related to the recent SpIN model), as well as the current limitations of the approach on large-scale problems. Reviewers also found authors to be responsive and diligent during the rebuttal phase. The AC agrees with this assessment, and therefore recommends rejection at this time, encouraging the authors to resubmit to the next conference cycle after addressing the above points. """
484,"""Learning From the Experience of Others: Approximate Empirical Bayes in Neural Networks""","['Empirical Bayes', 'Bayesian Deep Learning']","""Learning deep neural networks could be understood as the combination of representation learning and learning halfspaces. While most previous work aims to diversify representation learning by data augmentations and regularizations, we explore the opposite direction through the lens of empirical Bayes method. Specifically, we propose a matrix-variate normal prior whose covariance matrix has a Kronecker product structure to capture the correlations in learning different neurons through backpropagation. The prior encourages neurons to learn from the experience of others, hence it provides an effective regularization when training large networks on small datasets. To optimize the model, we design an efficient block coordinate descent algorithm with analytic solutions. Empirically, we show that the proposed method helps the network converge to better local optima that also generalize better, and we verify the effectiveness of the approach on both multiclass classification and multitask regression problems with various network structures. ""","""This paper proposes a method called approximate empirical Bayes to learn both the weights and hyperparameters. Reviewers have had a mixed feeling about this paper. Reviewers agree that the novelty of this paper is limited since AEB is already a well known method (in fact, iterative conditional modes is a well known algorithm). Unfortunately, the paper completely ignores the huge literature on this topic; the previous reference to use AEB is from McInerney (2017). Another issue is that the paper seems to be unaware of any issues that this type of approach might have. Here is a reference that discusses some problems with this type of approach: ""Deterministic Latent Variable Models and their Pitfalls"" Max Welling Chaitanya Chemudugunta, Nathan Sutter, 2008 The experiments presented in the paper are interesting, but then are not really doing a good job to assess why the method works well here even though in theory it should not be as good as the exact empirical Bayes method. This paper does not meet the bar for acceptance at ICLR and therefore I recommend a reject for this paper. """
485,"""Calibration of neural network logit vectors to combat adversarial attacks""","['Adversarial attacks', 'calibration', 'probability', 'adversarial defence']","""Adversarial examples remain an issue for contemporary neural networks. This paper draws on Background Check (Perello-Nieto et al., 2016), a technique in model calibration, to assist two-class neural networks in detecting adversarial examples, using the one dimensional difference between logit values as the underlying measure. This method interestingly tends to achieve the highest average recall on image sets that are generated with large perturbation vectors, which is unlike the existing literature on adversarial attacks (Cubuk et al., 2017). The proposed method does not need knowledge of the attack parameters or methods at training time, unlike a great deal of the literature that uses deep learning based methods to detect adversarial examples, such as Metzen et al. (2017), imbuing the proposed method with additional flexibility.""","""The reviewers and AC note the potential weaknesses of the paper in various aspects, and decided that the authors need more works to publish. """
486,"""In Your Pace: Learning the Right Example at the Right Time""","['Curriculum Learning', 'Transfer Learning', 'Self-Paced Learning', 'Image Recognition']","""Training neural networks is traditionally done by sequentially providing random mini-batches sampled uniformly from the entire dataset. In our work, we show that sampling mini-batches non-uniformly can both enhance the speed of learning and improve the final accuracy of the trained network. Specifically, we decompose the problem using the principles of curriculum learning: first, we sort the data by some difficulty measure; second, we sample mini-batches with a gradually increasing level of difficulty. We focus on CNNs trained on image recognition. Initially, we define the difficulty of a training image using transfer learning from some competitive ""teacher"" network trained on the Imagenet database, showing improvement in learning speed and final performance for both small and competitive networks, using the CIFAR-10 and the CIFAR-100 datasets. We then suggest a bootstrap alternative to evaluate the difficulty of points using the same network without relying on a ""teacher"" network, thus increasing the applicability of our suggested method. We compare this approach to a related version of Self-Paced Learning, showing that our method benefits learning while SPL impairs it.""","""This paper presents an interesting strategy of curriculum learning for training neural networks, where mini-batches of samples are formed with a gradually increasing level of difficulty. While reviewers acknowledge the importance of studying the curriculum learning and the potential usefulness of the proposed approach for training neural networks, they raised several important concerns that place this paper bellow the acceptance bar: (1) empirical results are not convincing (R2, R3); comparisons on other datasets (large-scale) and with state-of-the-art methods would substantially strengthen the evaluation (R3); see also R2s concerns regarding the comprehensive study; (2) important references and baseline methods are missing see R2s suggestions how to improve; (3) limited technical novelty -- R1 has provided a very detailed review questioning novelty of the proposed approach w.r.t. Weinshall et al, 2018. Another suggestions to further strengthen and extend the manuscript is to consider curriculum and anti-curriculum learning for increasing performance (R1). The authors provided additional experiment on a subset of 7 classes from the ImageNet dataset, but this does not show the advantage of the proposed model in a large-scale learning setting. The AC decided that addressing (1)-(3) is indeed important for understanding the contribution in this work, and it is difficult to assess the scope of the contribution without addressing them. """
487,"""Smoothing the Geometry of Probabilistic Box Embeddings""","['embeddings', 'order embeddings', 'knowledge graph embedding', 'relational learning']","""There is growing interest in geometrically-inspired embeddings for learning hierarchies, partial orders, and lattice structures, with natural applications to transitive relational data such as entailment graphs. Recent work has extended these ideas beyond deterministic hierarchies to probabilistically calibrated models, which enable learning from uncertain supervision and inferring soft-inclusions among concepts, while maintaining the geometric inductive bias of hierarchical embedding models. We build on the Box Lattice model of Vilnis et al. (2018), which showed promising results in modeling soft-inclusions through an overlapping hierarchy of sets, parameterized as high-dimensional hyperrectangles (boxes). However, the hard edges of the boxes present difficulties for standard gradient based optimization; that work employed a special surrogate function for the disjoint case, but we find this method to be fragile. In this work, we present a novel hierarchical embedding model, inspired by a relaxation of box embeddings into parameterized density functions using Gaussian convolutions over the boxes. Our approach provides an alternative surrogate to the original lattice measure that improves the robustness of optimization in the disjoint case, while also preserving the desirable properties with respect to the original lattice. We demonstrate increased or matching performance on WordNet hypernymy prediction, Flickr caption entailment, and a MovieLens-based market basket dataset. We show especially marked improvements in the case of sparse data, where many conditional probabilities should be low, and thus boxes should be nearly disjoint.""","""The manuscript presents a promising new algorithm for learning geometrically-inspired embeddings for learning hierarchies, partial orders, and lattice structures. The manuscript builds on the build on the box lattice model, extending prior work by relaxing the box embeddings via Gaussian convolutions. This is shown to be particularly effective for non-overlapping boxes, where the previous method fail. The primary weakness identified by reviewers was the writing, which was thought to be lacking some context, and may be difficult to approach for the non-domain expert. This can be improved by including an additional general introduction. Otherwise, the manuscript was well written. Overall, reviewers and AC agree that the general problem statement is timely and interesting, and well executed. In our opinion, this paper is a clear accept."""
488,"""StrokeNet: A Neural Painting Environment""","['image generation', 'differentiable model', 'reinforcement learning', 'deep learning', 'model based']","""We've seen tremendous success of image generating models these years. Generating images through a neural network is usually pixel-based, which is fundamentally different from how humans create artwork using brushes. To imitate human drawing, interactions between the environment and the agent is required to allow trials. However, the environment is usually non-differentiable, leading to slow convergence and massive computation. In this paper we try to address the discrete nature of software environment with an intermediate, differentiable simulation. We present StrokeNet, a novel model where the agent is trained upon a well-crafted neural approximation of the painting environment. With this approach, our agent was able to learn to write characters such as MNIST digits faster than reinforcement learning approaches in an unsupervised manner. Our primary contribution is the neural simulation of a real-world environment. Furthermore, the agent trained with the emulated environment is able to directly transfer its skills to real-world software.""","""The paper proposes a novel differential way to output brush strokes, taking a few ideas from model-based learning. The method is efficient in that one can train it in an unsupervised manner and does not require paired data. The strengths of the paper are the qualitative results that demonstrate nice interpolations among other things, on a number of datasets (esp. post-rebuttal). The weaknesses of the paper are the writing (which I think is relatively easy to improve if the authors make an honest effort) and some of the quantitative evaluation. I would encourage the authors to get in touch with the SPIRAL paper authors in order to get access to the SPIRAL generated MNIST test data and then perhaps the classification metric could be updated. In summary, from the discussion, the major points of contention were the somewhat lacking initial evaluation (which was fixed to a large extent) and the quality of writing (which could be fixed more). I believe the submission is genuinely novel, interesting (esp. the usage of world model-like techniques) and valuable for the ICLR audience so I recommend acceptance."""
489,"""Generalized Capsule Networks with Trainable Routing Procedure""","['Capsule networks', 'generalization', 'scalability', 'adversarial robustness']","""CapsNet (Capsule Network) was first proposed by Sabour et al. (2017) and lateranother version of CapsNet was proposed by Hinton et al. (2018). CapsNet hasbeen proved effective in modeling spatial features with much fewer parameters.However, the routing procedures (dynamic routing and EM routing) in both pa-pers are not well incorporated into the whole training process, and the optimalnumber for the routing procedure has to be found manually. We propose Gen-eralized GapsNet (G-CapsNet) to overcome this disadvantages by incorporatingthe routing procedure into the optimization. We implement two versions of G-CapsNet (fully-connected and convolutional) on CAFFE (Jia et al. (2014)) andevaluate them by testing the accuracy on MNIST & CIFAR10, the robustness towhite-box & black-box attack, and the generalization ability on GAN-generatedsynthetic images. We also explore the scalability of G-CapsNet by constructinga relatively deep G-CapsNet. The experiment shows that G-CapsNet has goodgeneralization ability and scalability. ""","""The paper proposes to replace dynamic routing in Capsule networks with a trainable layer that produces routing coefficients. The goal is to improve their scalability. This is promising as a research direction but reviewers have raised several concerns about unclear contributions and lack of a thorough evaluation of the approach. There is also a recent relevant work pointed out by Reviewer 1 that should be discussed. Given these concerns, the paper is not suitable for publication in its current form, however I encourage the authors to use reviewers' comments for improving the paper and resubmit in next venues."""
490,"""A Convergent Variant of the Boltzmann Softmax Operator in Reinforcement Learning""","['Reinforcement Learning', 'Boltzmann Softmax Operator', 'Value Function Estimation']","""The Boltzmann softmax operator can trade-off well between exploration and exploitation according to current estimation in an exponential weighting scheme, which is a promising way to address the exploration-exploitation dilemma in reinforcement learning. Unfortunately, the Boltzmann softmax operator is not a non-expansion, which may lead to unstable or even divergent learning behavior when used in estimating the value function. The non-expansion is a vital and widely-used sufficient condition to guarantee the convergence of value iteration. However, how to characterize the effect of such non-expansive operators in value iteration remains an open problem. In this paper, we propose a new technique to analyze the error bound of value iteration with the the Boltzmann softmax operator. We then propose the dynamic Boltzmann softmax(DBS) operator to enable the convergence to the optimal value function in value iteration. We also present convergence rate analysis of the algorithm. Using Q-learning as an application, we show that the DBS operator can be applied in a model-free reinforcement learning algorithm. Finally, we demonstrate the effectiveness of the DBS operator in a toy problem called GridWorld and a suite of Atari games. Experimental results show that outperforms DQN substantially in benchmark games.""","""Pros: - a method that obtains convergence results using a using time-dependent (not fixed or state-dependent) softmax temperature. Cons: - theoretical contribution is not very novel - some theoretical results are dubious - mismatch of Boltzmann updates and epsilon-greedy exploration - the authors seem to have intended to upload a revised version of the paper, but unfortunately, they changed only title and abstract, not the pdf -- and consequently the reviewers did not change their scores. The reviewers agree that the paper should be rejected in the submitted form."""
491,"""Knowledge Flow: Improve Upon Your Teachers""","['Transfer Learning', 'Reinforcement Learning']","""A zoo of deep nets is available these days for almost any given task, and it is increasingly unclear which net to start with when addressing a new task, or which net to use as an initialization for fine-tuning a new model. To address this issue, in this paper, we develop knowledge flow which moves knowledge from multiple deep nets, referred to as teachers, to a new deep net model, called the student. The structure of the teachers and the student can differ arbitrarily and they can be trained on entirely different tasks with different output spaces too. Upon training with knowledge flow the student is independent of the teachers. We demonstrate our approach on a variety of supervised and reinforcement learning tasks, outperforming fine-tuning and other knowledge exchange methods. ""","""The authors have taken inspiration from recent publications that demonstrate transfer learning over sequential RL tasks and have proposed a method that trains individual learners from experts using layerwise connections, gradually forcing the features to distill into the student with a hard-coded annealing of coeffiecients. The authors have done thorough experiments and the value of the approach seems clear, especially compared against progressive nets and pathnets. The paper is well-written and interesting, and the approach is novel. The reviewers have discussed the paper in detail and agree, with the AC, that it should be accepted."""
492,"""Prototypical Examples in Deep Learning: Metrics, Characteristics, and Utility""","['prototypes', 'curriculum learning', 'interpretability', 'differential privacy', 'adversarial robustness']","""Machine learning (ML) research has investigated prototypes: examples that are representative of the behavior to be learned. We systematically evaluate five methods for identifying prototypes, both ones previously introduced as well as new ones we propose, finding all of them to provide meaningful but different interpretations. Through a human study, we confirm that all five metrics are well matched to human intuition. Examining cases where the metrics disagree offers an informative perspective on the properties of data and algorithms used in learning, with implications for data-corpus construction, efficiency, adversarial robustness, interpretability, and other ML aspects. In particular, we confirm that the ""train on hard"" curriculum approach can improve accuracy on many datasets and tasks, but that it is strictly worse when there are many mislabeled or ambiguous examples.""","""This paper considers ""prototypes"" in machine learning, in which a small subset of a dataset is selected as representative of the behavior of the models. The authors propose a number of desiderata, and outline the connections to existing approaches. Further, they carry out evaluation with user studies to compare them with human intuition, and empirical experiments to compare them to each other. The reviewers agreed that the search for more concrete definitions of prototypes is a worthy one, and they appreciated the user studies. The reviewers and AC note the following potential weaknesses: (1) the specific description of prototypes that the authors are using is not provided precisely, (2) the desiderata was found to be informal, leading to considerable confusion regarding the choices that are made and their compatibility with each other, (3) concerns in the evaluation regarding the practicality and the appropriateness of the user study for the goals of the paper. Although the authors provided detailed responses to these concerns, most of them still remained. Both reviewer 1 and reviewer 2 encourage the authors to define the prototypes defined more precisely, providing motivation for the various choices therein. Even though some of the concerns raised by reviewer 3 were addressed, it still remains to be seen how scalable the approach is for real-world applications. For these reasons, the reviewers and the AC feel that the authors would need to make substantial improvements for the paper to be accepted."""
493,"""Learning concise representations for regression by evolving networks of trees""","['regression', 'stochastic optimization', 'evolutionary compution', 'feature engineering']","""We propose and study a method for learning interpretable representations for the task of regression. Features are represented as networks of multi-type expression trees comprised of activation functions common in neural networks in addition to other elementary functions. Differentiable features are trained via gradient descent, and the performance of features in a linear model is used to weight the rate of change among subcomponents of each representation. The search process maintains an archive of representations with accuracy-complexity trade-offs to assist in generalization and interpretation. We compare several stochastic optimization approaches within this framework. We benchmark these variants on 100 open-source regression problems in comparison to state-of-the-art machine learning approaches. Our main finding is that this approach produces the highest average test scores across problems while producing representations that are orders of magnitude smaller than the next best performing method (gradient boosting). We also report a negative result in which attempts to directly optimize the disentanglement of the representation result in more highly correlated features.""","""The reviewers all feel that the paper should be accepted to the conference. The main strengths that they noted were the quality of writing, the wide applicability of the proposed method and the strength of the empirical evaluation. It's nice to see experiments across a large number of problems (100), with corresponding code, where baselines were hyperparameter tuned as well. This helps to give some assurance that the method will generalize to new problems and datasets. Some weaknesses noted by the reviewers were computational cost (the method is significantly slower than the baselines) and they weren't entirely convinced that having more concise representations would directly lead to the claimed interpretability of the approach. Nevertheless, they found it would make for a solid contribution to the conference."""
494,"""Cramer-Wold AutoEncoder""","['autoencoder', 'generative models', 'deep neural networks']","""Assessing distance betweeen the true and the sample distribution is a key component of many state of the art generative models, such as Wasserstein Autoencoder (WAE). Inspired by prior work on Sliced-Wasserstein Autoencoders (SWAE) and kernel smoothing we construct a new generative model Cramer-Wold AutoEncoder (CWAE). CWAE cost function, based on introduced Cramer-Wold distance between samples, has a simple closed-form in the case of normal prior. As a consequence, while simplifying the optimization procedure (no need of sampling necessary to evaluate the distance function in the training loop), CWAE performance matches quantitatively and qualitatively that of WAE-MMD (WAE using maximum mean discrepancy based distance function) and often improves upon SWAE.""","""The reviewers in general like the idea of using the Cramer-Wold kernel, noting that its heavy tails and closed form solution are appealing properties that lead to increased stability and improved training. The main concern was novelty, as this paper can be seen as simply changing the kernel in WAE-MMD. One suggestion is to more heavily highlight the CW-distance, and in particular to find another useful application for it outside of WAE-MMD. The paper emphasizes frequently that the closed-form loss function is a critical feature of this approach, however I dont see any experiments that optimize WAE-MMD under the CW-distance while sampling from the Gaussian. This is important to measure the degree to which any improvement is attributable to a closed-form solution, or to the distance measure itself. """
495,"""Learning Information Propagation in the Dynamical Systems via Information Bottleneck Hierarchy""","['compact representation', 'perception', 'dynamical systems', 'information bottleneck']","""Extracting relevant information, causally inferring and predicting the future states with high accuracy is a crucial task for modeling complex systems. The endeavor to address these tasks is made even more challenging when we have to deal with high-dimensional heterogeneous data streams. Such data streams often have higher-order inter-dependencies across spatial and temporal dimensions. We propose to perform a soft-clustering of the data and learn its dynamics to produce a compact dynamical model while still ensuring the original objectives of causal inference and accurate predictions. To efficiently and rigorously process the dynamics of soft-clustering, we advocate for an information theory inspired approach that incorporates stochastic calculus and seeks to determine a trade-off between the predictive accuracy and compactness of the mathematical representation. We cast the model construction as a maximization of the compression of the state variables such that the predictive ability and causal interdependence (relatedness) constraints between the original data streams and the compact model are closely bounded. We provide theoretical guarantees concerning the convergence of the proposed learning algorithm. To further test the proposed framework, we consider a high-dimensional Gaussian case study and describe an iterative scheme for updating the new model parameters. Using numerical experiments, we demonstrate the benefits on compression and prediction accuracy for a class of dynamical systems. Finally, we apply the proposed algorithm to the real-world dataset of multimodal sentiment intensity and show improvements in prediction with reduced dimensions.""","""The reviewers reached a consensus that the paper is not ready for publication in ICLR. (see more details in the reviews below. )"""
496,"""IB-GAN: Disentangled Representation Learning with Information Bottleneck GAN""","['Unsupervised disentangled representation learning', 'GAN', 'Information Bottleneck', 'Variational Inference']","""We present a novel architecture of GAN for a disentangled representation learning. The new model architecture is inspired by Information Bottleneck (IB) theory thereby named IB-GAN. IB-GAN objective is similar to that of InfoGAN but has a crucial difference; a capacity regularization for mutual information is adopted, thanks to which the generator of IB-GAN can harness a latent representation in disentangled and interpretable manner. To facilitate the optimization of IB-GAN in practice, a new variational upper-bound is derived. With experiments on CelebA, 3DChairs, and dSprites datasets, we demonstrate that the visual quality of samples generated by IB-GAN is often better than those by -VAEs. Moreover, IB-GAN achieves much higher disentanglement metrics score than -VAEs or InfoGAN on the dSprites dataset.""","""Strengths: This paper introduces a clever construction to build a more principled disentanglement objective for GANs than the InfoGAN. The paper is relatively clearly written. This method provides the possibility of combining the merits of GANs with the useful information-theoretic quantities that can be used to regularize VAEs. Weaknesses: The quantitative experiments are based entirely around the toy dSprites dataset, on which they perform comparably to other methods. Additionally, the qualitative results look pretty bad (in my subjective opinion). They may still be better than a naive VAE, but the authors could have demonstrated the ability of their model by comparing their models against other models both qualitatively and quantitatively on problems hard enough to make the VAEs fail. Points of contention: The quantitative baselines are taken from another paper which did zero hyperparameter search. However the authors provided an updated results table based on numbers from other papers in a comment. Consensus: Everyone agreed that the idea was good and the experiments were lacking. Some of the comments about experiments were addressed in the updated version but not all."""
497,"""Large Scale Graph Learning From Smooth Signals""","['Graph learning', 'Graph signal processing', 'Network inference']","""Graphs are a prevalent tool in data science, as they model the inherent structure of the data. Typically they are constructed either by connecting nearest samples, or by learning them from data, solving an optimization problem. While graph learning does achieve a better quality, it also comes with a higher computational cost. In particular, the current state-of-the-art model cost is O(n^2) for n samples. In this paper, we show how to scale it, obtaining an approximation with leading cost of O(n log(n)), with quality that approaches the exact graph learning model. Our algorithm uses known approximate nearest neighbor techniques to reduce the number of variables, and automatically selects the correct parameters of the model, requiring a single intuitive input: the desired edge density.""","""The paper is proposed as probable accept based on current ratings with a majority accept (7,7,5)."""
498,"""An Information-Theoretic Metric of Transferability for Task Transfer Learning""","['transfer learning', 'task transfer learning', 'H-score', 'transferability']","""An important question in task transfer learning is to determine task transferability, i.e. given a common input domain, estimating to what extent representations learned from a source task can help in learning a target task. Typically, transferability is either measured experimentally or inferred through task relatedness, which is often defined without a clear operational meaning. In this paper, we present a novel metric, H-score, an easily-computable evaluation function that estimates the performance of transferred representations from one task to another in classification problems. Inspired by a principled information theoretic approach, H-score has a direct connection to the asymptotic error probability of the decision function based on the transferred feature. This formulation of transferability can further be used to select a suitable set of source tasks in task transfer learning problems or to devise efficient transfer learning policies. Experiments using both synthetic and real image data show that not only our formulation of transferability is meaningful in practice, but also it can generalize to inference problems beyond classification, such as recognition tasks for 3D indoor-scene understanding.""","""The paper proposes an information theoretic quantity to measure the performance of transferred representations with an operational appeal, easier computation, and empirical validation. The relation of the proposed measure to test accuracy is not considered. The operational meaning holds exactly only in the special case of linear fine tuning layers. The paper seems to import heavily from previous works. Reviewers found it difficult to understand whether the proposed method makes sense, that the computation of relevant quantities might be difficult in general, and that the comparison with mutual information was not clear. The revision addresses these points, adding experiments and explanations. Yet, none of the reviewers gives the paper a rating beyond marginally above acceptance threshold. All reviewers found the paper interesting and relevant, but none of them found the paper particularly strong. This is a borderline case of a sound and promising paper, which nonetheless seems to be missing a clear selling point. I would suggest that developing the program laid out in the conclusions could make the contributions more convincing, in particular the development of more scalable algorithms and the application of the proposed measure to the design of hierarchies for transfer learning. """
499,"""Learning Factorized Multimodal Representations""","['multimodal learning', 'representation learning']","""Learning multimodal representations is a fundamentally complex research problem due to the presence of multiple heterogeneous sources of information. Although the presence of multiple modalities provides additional valuable information, there are two key challenges to address when learning from multimodal data: 1) models must learn the complex intra-modal and cross-modal interactions for prediction and 2) models must be robust to unexpected missing or noisy modalities during testing. In this paper, we propose to optimize for a joint generative-discriminative objective across multimodal data and labels. We introduce a model that factorizes representations into two sets of independent factors: multimodal discriminative and modality-specific generative factors. Multimodal discriminative factors are shared across all modalities and contain joint multimodal features required for discriminative tasks such as sentiment prediction. Modality-specific generative factors are unique for each modality and contain the information required for generating data. Experimental results show that our model is able to learn meaningful multimodal representations that achieve state-of-the-art or competitive performance on six multimodal datasets. Our model demonstrates flexible generative capabilities by conditioning on independent factors and can reconstruct missing modalities without significantly impacting performance. Lastly, we interpret our factorized representations to understand the interactions that influence multimodal learning.""","""This paper offers a novel perspective for learning latent multimodal representations. The idea of segmenting the information into multimodal discriminative and modality-specific generating factors is found to be intriguing by all reviewers and the AC. The technical derivations allow for an efficient implementation of this idea. There have been some concerns regarding the experimental section, but they have all been addressed adequately during the rebuttal period. Therefore the AC suggests this paper for acceptance. It is an overall nice and well-thought work. """
500,"""Geomstats: a Python Package for Riemannian Geometry in Machine Learning""","['Riemannian geometry', 'Python package', 'machine learning', 'deep learning']","""We introduce geomstats, a Python package for Riemannian modelization and optimization over manifolds such as hyperspheres, hyperbolic spaces, SPD matrices or Lie groups of transformations. Our contribution is threefold. First, geomstats allows the flexible modeling of many a machine learning problem through an efficient and extensively unit-tested implementations of these manifolds, as well as the set of useful Riemannian metrics, exponential and logarithm maps that we provide. Moreover, the wide choice of loss functions and our implementation of the corresponding gradients allow fast and easy optimization over manifolds. Finally, geomstats is the only package to provide a unified framework for Riemannian geometry, as the operations implemented in geomstats are available with different computing backends (numpy,tensorflow and keras), as well as with a GPU-enabled mode-thus considerably facilitating the application of Riemannian geometry in machine learning. In this paper, we present geomstats through a review of the utility and advantages of manifolds in machine learning, using the concrete examples that they span to show the efficiency and practicality of their implementation using our package""","""Learning on Riemannian manifolds can be easily done with this Python package. Considering the recent work on these in latent-variable models, the package can be quite a useful approach. But its novelty is disputed. In particular Pymanopt is a package that does mostly the same, even though that may be computationally more expensive. The merits of Geomstats vs. Pymanopt is not clarified. But be that as it may, there is interest amongst the reviewers for the software package. In the end, too, it's not uniformly agreed upon that a software-describing paper fits ICLR."""
501,"""On Generalization Bounds of a Family of Recurrent Neural Networks""","['Recurrent Neural Networks', 'MGU', 'LSTM', 'Generalization Bound', 'PAC-Learning']","""Recurrent Neural Networks (RNNs) have been widely applied to sequential data analysis. Due to their complicated modeling structures, however, the theory behind is still largely missing. To connect theory and practice, we study the generalization properties of vanilla RNNs as well as their variants, including Minimal Gated Unit (MGU) and Long Short Term Memory (LSTM) RNNs. Specifically, our theory is established under the PAC-Learning framework. The generalization bound is presented in terms of the spectral norms of the weight matrices and the total number of parameters. We also establish refined generalization bounds with additional norm assumptions, and draw a comparison among these bounds. We remark: (1) Our generalization bound for vanilla RNNs is significantly tighter than the best of existing results; (2) We are not aware of any other generalization bounds for MGU and LSTM in the exiting literature; (3) We demonstrate the advantages of these variants in generalization.""","""Some expert reviewers have raised novelty issues, that the authors have addressed in detail. Still, these expert reviewers are not entirely convinced. If this were a journal, I would recommend a major revision or reject-and-resubmit in order to allow the authors to anticipate the reviewers' concerns in the body of the paper and get some fresh reviews. I compliment the authors on the diligence they have put into the rebuttal stage, and look forward to reading the next version of the work. I will note that the bounds by Bartlett, Foster, and Telgarsky (and then the PAC-Bayes versions by Neyshabur et al.) are numerically vacuous empirically, and so whether those bounds or these bounds for RNNs explain generalization is up for debate."""
502,"""Learning from Positive and Unlabeled Data with a Selection Bias""","['PU learning', 'deep learning', 'machine learning', 'anomaly detection', 'sampling bias']","""We consider the problem of learning a binary classifier only from positive data and unlabeled data (PU learning). Recent methods of PU learning commonly assume that the labeled positive data are identically distributed as the unlabeled positive data. However, this assumption is unrealistic in many instances of PU learning because it fails to capture the existence of a selection bias in the labeling process. When the data has a selection bias, it is difficult to learn the Bayes optimal classifier by conventional methods of PU learning. In this paper, we propose a method to partially identify the classifier. The proposed algorithm learns a scoring function that preserves the order induced by the class posterior under mild assumptions, which can be used as a classifier by setting an appropriate threshold. Through experiments, we show that the method outperforms previous methods for PU learning on various real-world datasets.""","""This manuscript proposes a new algorithm for learning from positive and unlabeled data. The motivation for this work includes cases of selection bias, where the positive label is correlated with observation. The resulting procedure is shown to learn a scoring function that preserves the class-posterior ordering, and can thus be thresholded to obtain a classifier. The problem addressed is interesting, and the approach sounds reasonable. The writing seems to be well done, particularly after the rebuttal when the work was better placed in context. The reviewers and AC note issues with the evaluation of the proposed method. In particular, the authors do not provide a sufficiently convincing empirical evaluation on real data. """
503,"""Guided Exploration in Deep Reinforcement Learning""","['deep reinforcement learning', 'guided exploration', 'RL training speed up']","""This paper proposes a new method to drastically speed up deep reinforcement learning (deep RL) training for problems that have the property of \textit{state-action permissibility} (SAP). Two types of permissibility are defined under SAP. The first type says that after an action pseudo-formula is performed in a state pseudo-formula and the agent reaches the new state pseudo-formula , the agent can decide whether the action pseudo-formula is \textit{permissible} or \textit{not permissible} in state pseudo-formula . The second type says that even without performing the action pseudo-formula in state pseudo-formula , the agent can already decide whether pseudo-formula is permissible or not in pseudo-formula . An action is not permissible in a state if the action can never lead to an optimal solution and thus should not be tried. We incorporate the proposed SAP property into two state-of-the-art deep RL algorithms to guide their state-action exploration. Results show that the SAP guidance can markedly speed up training.""","""The paper presents a simple and interesting idea to improve exploration efficiency, using the notion of action permissibility. Experiments in two problems (lane keeping, and flappy bird) show that exploration can be improved over baselines like DQN and DDPG. However, action permissibility appears to be very strong domain knowledge that limits the use in complex problems. Rephrasing one of reviewers, action permissibility essentially implies that some one-step information can be used to rule out suboptimal actions, while a defining challenge in RL is that the agent needs to learn/plan/reason over multiple steps to decide whether an action is suboptimal or not. Indeed, the two problems in the experiments have such a property that a myopic agent can solve the tasks pretty well. The paper would be stronger if the AP function can be defined for more common RL benchmarks, with similar benefits demonstrated."""
504,"""Meta-learning with differentiable closed-form solvers""","['few-shot learning', 'one-shot learning', 'meta-learning', 'deep learning', 'ridge regression', 'classification']","""Adapting deep networks to new concepts from a few examples is challenging, due to the high computational requirements of standard fine-tuning procedures. Most work on few-shot learning has thus focused on simple learning techniques for adaptation, such as nearest neighbours or gradient descent. Nonetheless, the machine learning literature contains a wealth of methods that learn non-deep models very efficiently. In this paper, we propose to use these fast convergent methods as the main adaptation mechanism for few-shot learning. The main idea is to teach a deep network to use standard machine learning tools, such as ridge regression, as part of its own internal model, enabling it to quickly adapt to novel data. This requires back-propagating errors through the solver steps. While normally the cost of the matrix operations involved in such a process would be significant, by using the Woodbury identity we can make the small number of examples work to our advantage. We propose both closed-form and iterative solvers, based on ridge regression and logistic regression components. Our methods constitute a simple and novel approach to the problem of few-shot learning and achieve performance competitive with or superior to the state of the art on three benchmarks.""","""The reviewers disagree strongly on this paper. Reviewer 2 was the most positive, believing it to be an interesting contribution with strong results. Reviewer 3 however, was underwhelmed by the results. Reviewer 1 does not believe that the contribution is sufficiently novel, seeing it as too close to existing multi-task learning approaches. After considering all of the discussion so far, I have to agree with reviewer 2 on their assessment. Much of the meta learning literature involves changing the base learner *for a fixed architecture* and seeing how it affects performance. There is a temptation to chase performance by changing the architecture, adding new regularizers, etc., and while this is important for practical reasons, it does not help to shed light on the underlying fundamentals. This is best done by considering carefully controlled and well understood experimental settings. Even still, the performance is quite good relative to popular base learners. Regarding novelty, I agree it is a simple change to the base learner, using a technique that has been tried before in other settings (linear regression as opposed to classification), however its use in a meta learning setup is novel in my opinion, and the new experimental comparison regression on top of pre-trained CNN features helps to demonstrate the utility of its use in the meta-learning settings. While the novelty can certainly be debated, I want to highlight two reasons why I am opting to accept this paper: 1) simple and effective ideas are often some of the most impactful. 2) sometimes taking ideas from one area (e.g., multi-task learning) and demonstrating that they can be effective in other settings (e.g., meta-learning) can itself be a valuable contribution. I believe that the meta-learning community would benefit from reading this paper. """
505,"""Implicit Autoencoders""","['Unsupervised Learning', 'Generative Models', 'Variational Inference', 'Generative Adversarial Networks.']","""In this paper, we describe the ""implicit autoencoder"" (IAE), a generative autoencoder in which both the generative path and the recognition path are parametrized by implicit distributions. We use two generative adversarial networks to define the reconstruction and the regularization cost functions of the implicit autoencoder, and derive the learning rules based on maximum-likelihood learning. Using implicit distributions allows us to learn more expressive posterior and conditional likelihood distributions for the autoencoder. Learning an expressive conditional likelihood distribution enables the latent code to only capture the abstract and high-level information of the data, while the remaining information is captured by the implicit conditional likelihood distribution. For example, we show that implicit autoencoders can disentangle the global and local information, and perform deterministic or stochastic reconstructions of the images. We further show that implicit autoencoders can disentangle discrete underlying factors of variation from the continuous factors in an unsupervised fashion, and perform clustering and semi-supervised learning.""","""The paper proposes an original idea for training a generative model based on an objective inspired by a VAE-like evidence lower bound (ELBO), reformulated as two KL terms, which are then approximately optimized by two GANs. They thus use implicit distributions for both the posterior and the conditional likelihood. The idea is original and intriguing. But reviewers and AC found that the paper currently suffered from the following weaknesses: a) The presentation of the approach is unclear, due primarily to the fact that it doesn't throughout unambiguously enough separate the VAE-like ELBO *inspiration*, from what happens when replacing the two KL terms by GANs, i.e. the actual algorithm used. This is a big conceptual jump that would deserve being discussed and analyzed more carefully and thoroughly. b) Reviewers agreed that the paper does not sufficiently evaluate the approach in comparative experiments with alternatives, in particular its generative capabilities, in addition to the provided evaluations of the learned representation on downstream tasks. Reviewers did not reach a clear consensus on this paper, although discussion led two of them to revise their assessment score slightly towards each other's. One reviewer judged the paper currently too confusing (point a) putting more weight on this aspect than the other reviewers. Based on the paper and the review discussion thread, the AC judges that while it is an original, interesting and potentially promising approach, its presentation can and should be much clarified and improved. """
506,"""Efficient Lifelong Learning with A-GEM""","['Lifelong Learning', 'Continual Learning', 'Catastrophic Forgetting', 'Few-shot Transfer']","""In lifelong learning, the learner is presented with a sequence of tasks, incrementally building a data-driven prior which may be leveraged to speed up learning of a new task. In this work, we investigate the efficiency of current lifelong approaches, in terms of sample complexity, computational and memory cost. Towards this end, we first introduce a new and a more realistic evaluation protocol, whereby learners observe each example only once and hyper-parameter selection is done on a small and disjoint set of tasks, which is not used for the actual learning experience and evaluation. Second, we introduce a new metric measuring how quickly a learner acquires a new skill. Third, we propose an improved version of GEM (Lopez-Paz & Ranzato, 2017), dubbed Averaged GEM (A-GEM), which enjoys the same or even better performance as GEM, while being almost as computationally and memory efficient as EWC (Kirkpatrick et al., 2016) and other regularization-based methods. Finally, we show that all algorithms including A-GEM can learn even more quickly if they are provided with task descriptors specifying the classification tasks under consideration. Our experiments on several standard lifelong learning benchmarks demonstrate that A-GEM has the best trade-off between accuracy and efficiency""",""" Pros: - Great work on getting rid of the need for QP and the corresponding proof of the update rule - Mostly clear writing - Good experimental results on relevant datasets - Introduction of a more reasonable evaluation methodology for continual learning Cons: - The model is arguably a little incremental over GEM. In the end I think all the reviewers agree though that the practical value of a considerably more efficient and easy to implement approach largely outweighs this concern. I think this is a good contribution in this area and I recommend acceptance."""
507,"""Learning To Simulate""","['Simulation in machine learning', 'reinforcement learning', 'policy gradients', 'image rendering']","""Simulation is a useful tool in situations where training data for machine learning models is costly to annotate or even hard to acquire. In this work, we propose a reinforcement learning-based method for automatically adjusting the parameters of any (non-differentiable) simulator, thereby controlling the distribution of synthesized data in order to maximize the accuracy of a model trained on that data. In contrast to prior art that hand-crafts these simulation parameters or adjusts only parts of the available parameters, our approach fully controls the simulator with the actual underlying goal of maximizing accuracy, rather than mimicking the real data distribution or randomly generating a large volume of data. We find that our approach (i) quickly converges to the optimal simulation parameters in controlled experiments and (ii) can indeed discover good sets of parameters for an image rendering simulator in actual computer vision applications.""","""This paper discusses the promising idea of using RL for optimizing simulators parameters. The theme of this paper was very well received by the reviewers. Initial concerns about insufficient experimentation were justified, however the amendments done during the rebuttal period ameliorated this issue. The authors argue that due to considered domain and status of existing literature, extensive comparisons are difficult. The AC sympathizes with this argument, however it is still advised that the experiments are conducted in a more conclusive way, for example by disentangling the effects of the different choices made by the proposed model. For example, how would different sampling strategies for optimization perform? Are there more natural black-box optimization methods to use? The reviewers believe that the methodology followed has a lot of space for improvement. However, the paper presents some fresh and intriguing ideas, which make it overall a relevant work for presentation at ICLR."""
508,"""Wasserstein Barycenter Model Ensembling""",['Wasserstein barycenter model ensembling'],"""In this paper we propose to perform model ensembling in a multiclass or a multilabel learning setting using Wasserstein (W.) barycenters. Optimal transport metrics, such as the Wasserstein distance, allow incorporating semantic side information such as word embeddings. Using W. barycenters to find the consensus between models allows us to balance confidence and semantics in finding the agreement between the models. We show applications of Wasserstein ensembling in attribute-based classification, multilabel learning and image captioning generation. These results show that the W. ensembling is a viable alternative to the basic geometric or arithmetic mean ensembling.""","""The paper proposes a novel way to ensemble multi-class or multi-label models based on a Wasserstein barycenter approach. The approach is theoretically justified and obtains good results. Reviewers were concerned with time complexity, and authors provided a clear breakdown of the complexity. Overall, all reviewers were positives in their scores, and I recommend accepting the paper."""
509,"""Combining adaptive algorithms and hypergradient method: a performance and robustness study""","['optimization', 'adaptive methods', 'learning rate decay']","""Wilson et al. (2017) showed that, when the stepsize schedule is properly designed, stochastic gradient generalizes better than ADAM (Kingma & Ba, 2014). In light of recent work on hypergradient methods (Baydin et al., 2018), we revisit these claims to see if such methods close the gap between the most popular optimizers. As a byproduct, we analyze the true benefit of these hypergradient methods compared to more classical schedules, such as the fixed decay of Wilson et al. (2017). In particular, we observe they are of marginal help since their performance varies significantly when tuning their hyperparameters. Finally, as robustness is a critical quality of an optimizer, we provide a sensitivity analysis of these gradient based optimizers to assess how challenging their tuning is.""","""The paper is a premature submission that needs significant improvement in terms of conceptual, theoretical, and empirical aspects."""
510,"""Attentive Task-Agnostic Meta-Learning for Few-Shot Text Classification""","['meta-learning', 'learning to learn', 'few-shot learning']","""Current deep learning based text classification methods are limited by their ability to achieve fast learning and generalization when the data is scarce. We address this problem by integrating a meta-learning procedure that uses the knowledge learned across many tasks as an inductive bias towards better natural language understanding. Inspired by the Model-Agnostic Meta-Learning framework (MAML), we introduce the Attentive Task-Agnostic Meta-Learning (ATAML) algorithm for text classification. The proposed ATAML is designed to encourage task-agnostic representation learning by way of task-agnostic parameterization and facilitate task-specific adaptation via attention mechanisms. We provide evidence to show that the attention mechanism in ATAML has a synergistic effect on learning performance. Our experimental results reveal that, for few-shot text classification tasks, gradient-based meta-learning approaches ourperform popular transfer learning methods. In comparisons with models trained from random initialization, pretrained models and meta trained MAML, our proposed ATAML method generalizes better on single-label and multi-label classification tasks in miniRCV1 and miniReuters-21578 datasets.""","""This paper describes an incorporation of attention into model agnostic meta learning. The reviewers found that the paper was rather confusing in its presentation of both the method and the tasks. While the results seemed interesting, it was difficult to frame them due to lack of clarity as to what the task is, and the relation between attention and MAML. It sounds like this paper needs a bit more work, and thus is not suitable for publication at this time. It is disappointing that the reviews were so short, but as the authors did not challenge them, unfortunately the AC must decide on the basis of the first set of comments by reviewers."""
511,"""Cohen Welling bases & SO(2)-Equivariant classifiers using Tensor nonlinearity.""","['group representations', 'group equivariant networks', 'tensor product nonlinearity']","""In this paper we propose autoencoder architectures for learning a Cohen-Welling (CW)-basis for images and their rotations. We use the learned CW-basis to build a rotation equivariant classifier to classify images. The autoencoder and classi- fier architectures use only tensor product nonlinearity. The model proposed by Cohen & Welling (2014) uses ideas from group representation theory, and extracts a basis exposing irreducible representations for images and their rotations. We give several architectures to learn CW-bases including a novel coupling AE archi- tecture to learn a coupled CW-bases for images in different scales simultaneously. Our use of tensor product nonlinearity is inspired from recent work of Kondor (2018a). Our classifier has very good accuracy and we use fewer parameters. Even when the sample complexity to learn a good CW-basis is low we learn clas- sifiers which perform impressively. We show that a coupled CW-bases in one scale can be deployed to classify images in a classifier trained and tested on images in a different scale with only a marginal dip in performance.""","""This paper studies group equivariant neural network representations by building on the work by [Cohen and Welling, '14], which introduced learning of group irreducible representations, and [Kondor'18], who introduced tensor product non-linearities operating directly in the group Fourier domain. Reviewers highlighted the significance of the approach, but were also unanimously concerned by the lack of clarity of the current manuscript, making its widespread impact within ICLR difficult, and the lack of a large-scale experiment that corroborates the usefulness of the approach. They were also very positive about the improvements of the paper during the author response phase. The AC completely agrees with this assessment of the paper. Therefore, the paper cannot be accepted at this time, but the AC strongly encourages the authors to resubmit their work in the next conference cycle by addressing the above remarks (improve clarity of presentation and include a large-scale experiment). """
512,"""Actor-Attention-Critic for Multi-Agent Reinforcement Learning""","['multi-agent', 'reinforcement learning', 'attention', 'actor-critic']","""Reinforcement learning in multi-agent scenarios is important for real-world applications but presents challenges beyond those seen in single-agent settings. We present an actor-critic algorithm that trains decentralized policies in multi-agent settings, using centrally computed critics that share an attention mechanism which selects relevant information for each agent at every timestep. This attention mechanism enables more effective and scalable learning in complex multi-agent environments, when compared to recent approaches. Our approach is applicable not only to cooperative settings with shared rewards, but also individualized reward settings, including adversarial settings, and it makes no assumptions about the action spaces of the agents. As such, it is flexible enough to be applied to most multi-agent learning problems""","""The authors propose an approach for a learnt attention mechanism to be used for selecting agents in a multi agent RL setting. The attention mechanism is learnt by a central critic, and it scales linearly with the number of agents rather than quadratically. There is some novelty in the proposed method, and the authors clearly explain and motivate the approach. However the empirical evaluation feels quite limited and does not show conclusively that the method is superior to the others. Moreover, the simple empirical results don't give any evidence how the attention mechanism is working or whether it is truly the attention that is affecting the results. The reviewers were split on their recommendation and did not come to a consensus. The AC feels that the paper is not quite strong enough and encourages the authors to broaden the work with additional experiments and analysis."""
513,"""Pumpout: A Meta Approach for Robustly Training Deep Neural Networks with Noisy Labels""","['Noisy Labels', 'Deep Learning', 'Meta Approach']","""It is challenging to train deep neural networks robustly on the industrial-level data, since labels of such data are heavily noisy, and their label generation processes are normally agnostic. To handle these issues, by using the memorization effects of deep neural networks, we may train deep neural networks on the whole dataset only the first few iterations. Then, we may employ early stopping or the small-loss trick to train them on selected instances. However, in such training procedures, deep neural networks inevitably memorize some noisy labels, which will degrade their generalization. In this paper, we propose a meta algorithm called Pumpout to overcome the problem of memorizing noisy labels. By using scaled stochastic gradient ascent, Pumpout actively squeezes out the negative effects of noisy labels from the training model, instead of passively forgetting these effects. We leverage Pumpout to upgrade two representative methods: MentorNet and Backward Correction. Empirical results on benchmark vision and text datasets demonstrate that Pumpout can significantly improve the robustness of representative methods.""","""The paper presents an approach to mitigate the presence of noisy labels during training by trying to forget wrong labels. Reviewers pointed out a few concerns, including lack of novelty, lack of enough experimental support, and lack of theoretical support. Authors have added some experiments and details about the experimental section, but reviewers still think it's not enough for acceptance. I concur with the reviewers to reject the paper."""
514,"""Sparse Binary Compression: Towards Distributed Deep Learning with minimal Communication""",[],"""Currently, progressively larger deep neural networks are trained on ever growing data corpora. In result, distributed training schemes are becoming increasingly relevant. A major issue in distributed training is the limited communication bandwidth between contributing nodes or prohibitive communication cost in general. %These challenges become even more pressing, as the number of computation nodes increases. To mitigate this problem we propose Sparse Binary Compression (SBC), a compression framework that allows for a drastic reduction of communication cost for distributed training. SBC combines existing techniques of communication delay and gradient sparsification with a novel binarization method and optimal weight update encoding to push compression gains to new limits. By doing so, our method also allows us to smoothly trade-off gradient sparsity and temporal sparsity to adapt to the requirements of the learning task. %We use tools from information theory to reason why SBC can achieve the striking compression rates observed in the experiments. Our experiments show, that SBC can reduce the upstream communication on a variety of convolutional and recurrent neural network architectures by more than four orders of magnitude without significantly harming the convergence speed in terms of forward-backward passes. For instance, we can train ResNet50 on ImageNet in the same number of iterations to the baseline accuracy, using 3531 less bits or train it to a pseudo-formula lower accuracy using 37208$ less bits. In the latter case, the total upstream communication required is cut from 125 terabytes to 3.35 gigabytes for every participating client. Our method also achieves state-of-the-art compression rates in a Federated Learning setting with 400 clients.""","""This paper proposes a sparse binary compression method for distributed training of neural networks with minimal communication cost. Unfortunately, the proposed approach is not novel, nor supported by strong experiments. The authors did not provide a rebuttal for reviewers' concerns. """
515,"""No Training Required: Exploring Random Encoders for Sentence Classification""",['sentence embeddings'],"""We explore various methods for computing sentence representations from pre-trained word embeddings without any training, i.e., using nothing but random parameterizations. Our aim is to put sentence embeddings on more solid footing by 1) looking at how much modern sentence embeddings gain over random methods---as it turns out, surprisingly little; and by 2) providing the field with more appropriate baselines going forward---which are, as it turns out, quite strong. We also make important observations about proper experimental protocol for sentence classification evaluation, together with recommendations for future research.""","""This paper provides a new family of untrained/randomly initialized sentence encoder baselines for a standard suite of NLP evaluation tasks, and shows that it does surprisingly wellvery close to widely-used methods for some of the tasks. All three reviewers acknowledge that this is a substantial contribution, and none see any major errors or fatal flaws. One reviewer had initially argued the experiments and discussion are not as thorough as would be typical for a strong paper. In particular, the results are focused on a single set of word embeddings and a narrow class of architectures. I'm sympathetic to this concern, but since there don't seem to be any outstanding concerns about the correctness of the paper, and since the other reviewers see the contribution as quite important, I recommend acceptance. [Update: This reviewer has since revised their review to make it more positive.] (As a nit, I'd ask the authors to ensure that the final version of the paper fits within the margins.)"""
516,"""Denoise while Aggregating: Collaborative Learning in Open-Domain Question Answering""","['natural language processing', 'open-domain question answering', 'semi-supervised learning']","""The open-domain question answering (OpenQA) task aims to extract answers that match specific questions from a distantly supervised corpus. Unlike supervised reading comprehension (RC) datasets where questions are designed for particular paragraphs, background sentences in OpenQA datasets are more prone to noise. We observe that most existing OpenQA approaches are vulnerable to noise since they simply regard those sentences that contain the answer span as ground truths and ignore the plausible correlation between the sentences and the question. To address this deficiency, we introduce a unified and collaborative model that leverages alignment information from query-sentence pairs in a small-scale supervised RC dataset and aggregates relevant evidence from distantly supervised corpus to answer open-domain questions. We evaluate our model on several real-world OpenQA datasets, and experimental results show that our collaborative learning methods outperform the existing baselines significantly.""","""This paper presents a model for question answering, where the idea is to have a collaborative model that aligns queries and sentences on a small supervised dataset and also uses semi-supervised information from a weakly supervised corpus to answer open domain questions resulting in short answer spans. The main criticism of the paper is regarding its novelty, and reviewers cite the similarities with prior work such as Chen et al. and Min et al. There is relative consensus between the reviewers that further work using the semi-supervised outlook with stronger results could strengthen the paper further."""
517,"""GRAPH TRANSFORMATION POLICY NETWORK FOR CHEMICAL REACTION PREDICTION""","['Chemical Reaction', 'Graph Transformation', 'Reinforcement Learning']","""We address a fundamental problem in chemistry known as chemical reaction product prediction. Our main insight is that the input reactant and reagent molecules can be jointly represented as a graph, and the process of generating product molecules from reactant molecules can be formulated as a sequence of graph transformations. To this end, we propose Graph Transformation Policy Network (GTPN) - a novel generic method that combines the strengths of graph neural networks and reinforcement learning to learn the reactions directly from data with minimal chemical knowledge. Compared to previous methods, GTPN has some appealing properties such as: end-to-end learning, and making no assumption about the length or the order of graph transformations. In order to guide model search through the complex discrete space of sets of bond changes effectively, we extend the standard policy gradient loss by adding useful constraints. Evaluation results show that GTPN improves the top-1 accuracy over the current state-of-the-art method by about 3% on the large USPTO dataset. Our model's performances and prediction errors are also analyzed carefully in the paper.""","""The reviewers and authors participated in modest discussion, with the authors providing direct responses to reviewer comments. However, this did not appreciably change the overall ratings of the paper (one reviewer raised their rating, while another grew more concerned), and in aggregate the reviewers do not recommend that the paper meets the bar for acceptance."""
518,"""Cutting Down Training Memory by Re-fowarding""","['deep learning', 'training memory', 'computation-memory trade off', 'optimal solution']","""Deep Neutral Networks(DNNs) require huge GPU memory when training on modern image/video databases. Unfortunately, the GPU memory as a hardware resource is always finite, which limits the image resolution, batch size, and learning rate that could be used for better DNN performance. In this paper, we propose a novel training approach, called Re-forwarding, that substantially reduces memory usage in training. Our approach automatically finds a subset of vertices in a DNN computation graph, and stores tensors only at these vertices during the first forward. During backward, extra local forwards (called the Re-forwarding process) are conducted to compute the missing tensors between the subset of vertices. The total memory cost becomes the sum of (1) the memory cost at the subset of vertices and (2) the maximum memory cost among local re-forwards. Re-forwarding trades training time overheads for memory and does not compromise any performance in testing. We propose theories and algorithms that achieve the optimal memory solutions for DNNs with either linear or arbitrary computation graphs. Experiments show that Re-forwarding cuts down up-to 80% of training memory on popular DNNs such as Alexnet, VGG, ResNet, Densenet and Inception net.""","""This paper proposed a method to reduce the memory of training neural nets, in exchange for additional training time. The paper is simple and looks reasonable. It's a natural followup with previous work. The improvement over previous work is not significant, with some overhead incurred in training time. This is a borderline paper but given the <30% acceptance rate, I need to downgrade the paper to reject. """
519,"""Modular Deep Probabilistic Programming""",[],"""Modularity is a key feature of deep learning libraries but has not been fully exploited for probabilistic programming. We propose to improve modularity of probabilistic programming language by offering not only plain probabilistic distributions but also sophisticated probabilistic model such as Bayesian non-parametric models as fundamental building blocks. We demonstrate this idea by presenting a modular probabilistic programming language MXFusion, which includes a new type of re-usable building blocks, called probabilistic modules. A probabilistic module consists of a set of random variables with associated probabilistic distributions and dedicated inference methods. Under the framework of variational inference, the pre-specified inference methods of individual probabilistic modules can be transparently used for inference of the whole probabilistic model. We demonstrate the power and convenience of probabilistic modules in MXFusion with various examples of Gaussian process models, which are evaluated with experiments on real data.""","""This paper presents a probabilistic programming language where models are constructed out of building blocks which specify both the distribution and an inference procedure. As a demonstration, they show how a GP-LVM can be implemented. The paper spends a lot of space arguing for the benefits of modularity. Modularity is of course hard to argue with, and the benefits are already understood in the PPL community. But, as the reviewers point out, various other PPLs have already adopted various strategies to enable modular definition of models, and (in cases like Venture) special-purpose higher-level inference algorithms. This paper contains little discussion of other PPLs and how the specific design decisions relate to theirs, so it's hard to judge whether this paper really covers new ground. Such discussion wasn't added to the revised paper, even though multiple reviewers asked for it. I can't recommend acceptance. """
520,"""Unsupervised Domain Adaptation for Distance Metric Learning""","['domain adaptation', 'distance metric learning', 'face recognition']","""Unsupervised domain adaptation is a promising avenue to enhance the performance of deep neural networks on a target domain, using labels only from a source domain. However, the two predominant methods, domain discrepancy reduction learning and semi-supervised learning, are not readily applicable when source and target domains do not share a common label space. This paper addresses the above scenario by learning a representation space that retains discriminative power on both the (labeled) source and (unlabeled) target domains while keeping representations for the two domains well-separated. Inspired by a theoretical analysis, we first reformulate the disjoint classification task, where the source and target domains correspond to non-overlapping class labels, to a verification one. To handle both within and cross domain verifications, we propose a Feature Transfer Network (FTN) to separate the target feature space from the original source space while aligned with a transformed source space. Moreover, we present a non-parametric multi-class entropy minimization loss to further boost the discriminative power of FTNs on the target domain. In experiments, we first illustrate how FTN works in a controlled setting of adapting from MNIST-M to MNIST with disjoint digit classes between the two domains and then demonstrate the effectiveness of FTNs through state-of-the-art performances on a cross-ethnicity face recognition problem. ""","""This paper proposes a new solution for tackling domain adaptation across disjoint label spaces. Two of the reviewers agree that the main technical approach is interesting and novel. The final reviewer asked for clarification of the problem setting which the authors have provided in their rebuttal. We encourage the authors to include this in the final version. However, there is also a consensus that more experimental evaluation would improve the manuscript and complete experimental details are needed for reliable reproduction."""
521,"""Context-adaptive Entropy Model for End-to-end Optimized Image Compression""","['image compression', 'deep learning', 'entropy model']","""We propose a context-adaptive entropy model for use in end-to-end optimized image compression. Our model exploits two types of contexts, bit-consuming contexts and bit-free contexts, distinguished based upon whether additional bit allocation is required. Based on these contexts, we allow the model to more accurately estimate the distribution of each latent representation with a more generalized form of the approximation models, which accordingly leads to an enhanced compression performance. Based on the experimental results, the proposed method outperforms the traditional image codecs, such as BPG and JPEG2000, as well as other previous artificial-neural-network (ANN) based approaches, in terms of the peak signal-to-noise ratio (PSNR) and multi-scale structural similarity (MS-SSIM) index. The test code is publicly available at pseudo-url.""","""This paper proposes an algorithm for end-to-end image compression outperforming previously proposed ANN-based techniques and typical image compression standards like JPEG. Strengths - All reviewers agreed that this a well written paper, with careful analysis and results. Weaknesses - One of the points raised during the review process was that 2 very recent publications propose very similar algorithms. Since these works appeared very close to ICLR paper submission deadline (within 30 days), the program committee decided to treat this as concurrent work. The authors also clarified the differences and similarities with prior work, and included additional experiments to clarify some of the concerns raised during the review process. Overall the paper is a solid contribution towards improving image compression, and is therefore recommended to be accepted. """
522,"""CONTROLLING COVARIATE SHIFT USING EQUILIBRIUM NORMALIZATION OF WEIGHTS""","['normalization', 'optimization']","""We introduce a new normalization technique that exhibits the fast convergence properties of batch normalization using a transformation of layer weights instead of layer outputs. The proposed technique keeps the contribution of positive and negative weights to the layer output in equilibrium. We validate our method on a set of standard benchmarks including CIFAR-10/100, SVHN and ILSVRC 2012 ImageNet.""","""This paper introduces a technique called EquiNorm, which normalizes the weights of convolutional layers in order to control covariate shift. The paper is well-written and the reviewers agree that the solution idea is elegant. However, the reviewers also agree that the experiments presented in the work were insufficient to prove the method's superiority. Reviewer 2 also expressed concerns about the poor results on ImageNet, which calls into question the significance of the proposed method. """
523,"""Open-Ended Content-Style Recombination Via Leakage Filtering""",[],"""We consider visual domains in which a class label specifies the content of an image, and class-irrelevant properties that differentiate instances constitute the style. We present a domain-independent method that permits the open-ended recombination of style of one image with the content of another. Open ended simply means that the method generalizes to style and content not present in the training data. The method starts by constructing a content embedding using an existing deep metric-learning technique. This trained content encoder is incorporated into a variational autoencoder (VAE), paired with a to-be-trained style encoder. The VAE reconstruction loss alone is inadequate to ensure a decomposition of the latent representation into style and content. Our method thus includes an auxiliary loss, leakage filtering, which ensures that no style information remaining in the content representation is used for reconstruction and vice versa. We synthesize novel images by decoding the style representation obtained from one image with the content representation from another. Using this method for data-set augmentation, we obtain state-of-the-art performance on few-shot learning tasks.""","""The paper is on the borderline. From my reading, the paper presents a reasonable idea with quite good results on novel image generation and one-shot learning. On the other hand, the comparison against the prior work (both generation task and one-shot classification task) is not convincing. I also feel that there are many work with similar ideas (I listed some below, but these are not exhaustive/comprehensive list), but they are not cited or compared, I am not sure about if the proposed concept is novel in high-level. Although some implementation details of this method may provide advantages over other related work, such comparison is not clear to me. Disentangling factors of variation in deep representations using adversarial training pseudo-url NIPS 2016 Rethinking Style and Content Disentanglement in Variational Autoencoders pseudo-url ICLR 2018 workshop Disentangling Factors of Variation by Mixing Them pseudo-url CVPR 2018 Separating Style and Content for Generalized Style Transfer pseudo-url Finally, I feel that the writing needs improvement. Although the method is intuitive and has simple idea, the paper seems to lack full details (e.g., principled derivation of the model as a variant of the VAE formulation) and precise definitions of terms (e.g., second term of LF loss). """
524,"""Adversarial Domain Adaptation for Stable Brain-Machine Interfaces""","['Brain-Machine Interfaces', 'Domain Adaptation', 'Adversarial Networks']","""Brain-Machine Interfaces (BMIs) have recently emerged as a clinically viable option to restore voluntary movements after paralysis. These devices are based on the ability to extract information about movement intent from neural signals recorded using multi-electrode arrays chronically implanted in the motor cortices of the brain. However, the inherent loss and turnover of recorded neurons requires repeated recalibrations of the interface, which can potentially alter the day-to-day user experience. The resulting need for continued user adaptation interferes with the natural, subconscious use of the BMI. Here, we introduce a new computational approach that decodes movement intent from a low-dimensional latent representation of the neural data. We implement various domain adaptation methods to stabilize the interface over significantly long times. This includes Canonical Correlation Analysis used to align the latent variables across days; this method requires prior point-to-point correspondence of the time series across domains. Alternatively, we match the empirical probability distributions of the latent variables across days through the minimization of their Kullback-Leibler divergence. These two methods provide a significant and comparable improvement in the performance of the interface. However, implementation of an Adversarial Domain Adaptation Network trained to match the empirical probability distribution of the residuals of the reconstructed neural signals outperforms the two methods based on latent variables, while requiring remarkably few data points to solve the domain adaptation problem.""","""BMIs need per-patient and per-session calibration, and this paper seeks to amend that. Using VAEs and RNNs, it relates sEEG to sEMG, in principle a ten-year old approach, but do so using a novel adversarial approach that seems to work. The reviewers agree the approach is nice, the statements in the paper are too strong, but publication is recommended. Clinical evaluation is an important next step."""
525,"""An experimental study of layer-level training speed and its impact on generalization""","['generalization', 'optimization', 'vanishing gradients', 'experimental', 'fundamental research']","""How optimization influences the generalization ability of a DNN is still an active area of research. This work aims to unveil and study a factor of influence: the speed at which each layer trains. In our preliminary work, we develop a visualization technique and an optimization algorithm to monitor and control the layer rotation rate, a tentative measure of layer-level training speed, and show that it has a remarkably consistent and substantial impact on generalization. Our experiments further suggest that weight decay's and adaptive gradients methods' impact on both generalization performance and speed of convergence are solely due to layer rotation rate changes compared to vanilla SGD, offering a novel interpretation of these widely used techniques, and providing supplementary evidence that layer-level training speed indeed impacts generalization. Besides these fundamental findings, we also expect that on a practical level, the tools we introduce will reduce the meta-parameter tuning required to get the best generalization out of a deep network.""","""Dear authors, The reviewers all appreciated the question you are asking and the study of the impact of each layer is definitely an interesting one. They were however uncertain about the actual metrics you used to emphasize your points. Further, as you noted, there were quite a few presentation issues that led to skepticism of the reviewers, despite them spending quite a bit of time reading the paper and engaging in discussion. Hence, I regret to inform you that your work is not yet ready for publication. A more focused analysis would be a great addition to the questions you raise."""
526,"""Variational Sparse Coding""","['Variational Auto-Encoders', 'Sparse Coding', 'Variational Inference']","""Variational auto-encoders (VAEs) offer a tractable approach when performing approximate inference in otherwise intractable generative models. However, standard VAEs often produce latent codes that are disperse and lack interpretability, thus making the resulting representations unsuitable for auxiliary tasks (e.g. classication) and human interpretation. We address these issues by merging ideas from variational auto-encoders and sparse coding, and propose to explicitly model sparsity in the latent space of a VAE with a Spike and Slab prior distribution. We derive the evidence lower bound using a discrete mixture recognition function thereby making approximate posterior inference as computational efcient as in the standard VAE case. With the new approach, we are able to infer truly sparse representations with generally intractable non-linear probabilistic models. We show that these sparse representations are advantageous over standard VAE representations on two benchmark classication tasks (MNIST and Fashion-MNIST) by demonstrating improved classication accuracy and signicantly increased robustness to the number of latent dimensions. Furthermore, we demonstrate qualitatively that the sparse elements capture subjectively understandable sources of variation.""","""The paper develops and investigates the use of a spike-and-slab prior and approximate posterior for a VAE. It uses a continuous relaxation for the discrete binary component in the reconstruction term of the ELBO, and an analytic expression for the KL term between the spike-and-slab prior and approximate posterior. Experiments on MNIS, Fashion-MNIST and CelebA convincingly show that the approach works to learn sparse representations with improved interpretability that also yield more robust classification All reviewers agreed that this approach to sparsity in VAEs is well motivated and sound, that the paper is well written and clear, and the experiments interesting. One reviewer noted that the accuracy on MNIST remains really poor, so the approach does not cure VAEs yielding subpar representations for classification (although not the goal of this research). The reviewers and the AC however all judged that it currently constitutes a too limited contribution because a) the approach is a straightforward application of vanilla VAEs with a different prior/posterior, and is thus rather incremental. b) the scope of the paper is rather limited, in particular as it does not sufficiently discuss and does not empirically compare with other (VAE-related) approaches from the literature that were developed for sparse latent representations. """
527,"""Structured Neural Summarization""","['Summarization', 'Graphs', 'Source Code']","""Summarization of long sequences into a concise statement is a core problem in natural language processing, requiring non-trivial understanding of the input. Based on the promising results of graph neural networks on highly structured data, we develop a framework to extend existing sequence encoders with a graph component that can reason about long-distance relationships in weakly structured data such as text. In an extensive evaluation, we show that the resulting hybrid sequence-graph models outperform both pure sequence models as well as pure graph models on a range of summarization tasks.""","""This paper examines ways of encoding structured input such as source code or parsed natural language into representations that are conducive for summarization. Specifically, the innovation is to not use only a sequence model, nor only a tree model, but both. Empirical evaluation is extensive, and it is exhaustively demonstrated that combining both models provides the best results. The major perceived issue of the paper is the lack of methodological novelty, which the authors acknowledge. In addition, there are other existing graph-based architectures that have not been compared to. However, given that the experimental results are informative and convincing, I think that the paper is a reasonable candidate to be accepted to the conference."""
528,"""Evaluating GANs via Duality""","['Generative Adversarial Networks', 'GANs', 'game theory']","""Generative Adversarial Networks (GANs) have shown great results in accurately modeling complex distributions, but their training is known to be difficult due to instabilities caused by a challenging minimax optimization problem. This is especially troublesome given the lack of an evaluation metric that can reliably detect non-convergent behaviors. We leverage the notion of duality gap from game theory in order to propose a novel convergence metric for GANs that has low computational cost. We verify the validity of the proposed metric for various test scenarios commonly used in the literature. ""","""All reviewers still argue for rejection for the submitted paper. The AC thinks that this paper should be published at some point, but for now it is a ""revise and resubmit""."""
529,"""Guiding Physical Intuition with Neural Stethoscopes""","['Deep Learning', 'Intuitive Physics', 'Stability Prediction', 'Adversarial Training', 'Auxiliary Training', 'Multi-Task Learning']","""Model interpretability and systematic, targeted model adaptation present central challenges in deep learning. In the domain of intuitive physics, we study the task of visually predicting stability of block towers with the goal of understanding and influencing the model's reasoning. Our contributions are two-fold. Firstly, we introduce neural stethoscopes as a framework for quantifying the degree of importance of specific factors of influence in deep networks as well as for actively promoting and suppressing information as appropriate. In doing so, we unify concepts from multitask learning as well as training with auxiliary and adversarial losses. Secondly, we deploy the stethoscope framework to provide an in-depth analysis of a state-of-the-art deep neural network for stability prediction, specifically examining its physical reasoning. We show that the baseline model is susceptible to being misled by incorrect visual cues. This leads to a performance breakdown to the level of random guessing when training on scenarios where visual cues are inversely correlated with stability. Using stethoscopes to promote meaningful feature extraction increases performance from 51% to 90% prediction accuracy. Conversely, training on an easy dataset where visual cues are positively correlated with stability, the baseline model learns a bias leading to poor performance on a harder dataset. Using an adversarial stethoscope, the network is successfully de-biased, leading to a performance increase from 66% to 88%.""","""This submission proposes an interesting new approach on how to evaluate what features are the most useful during training. The paper is interesting and the proposed approach has the potential to be deployed in many applications, however the work as currently presented is demonstrated in a very narrow domain (stability prediction), as noted by all reviewers. Authors are encouraged to provide stronger experimental validation over more domains to show that their approach can truly improve over existing multitask frameworks."""
530,"""Monge-Amp\`ere Flow for Generative Modeling""","['generative modeling', 'Monge-Amp\\`ere equation', 'dynamical system', 'optimal transport', 'density estimation', 'free energy calculation']","""We present a deep generative model, named Monge-Amp\`ere flow, which builds on continuous-time gradient flow arising from the Monge-Amp\`ere equation in optimal transport theory. The generative map from the latent space to the data space follows a dynamical system, where a learnable potential function guides a compressible fluid to flow towards the target density distribution. Training of the model amounts to solving an optimal control problem. The Monge-Amp\`ere flow has tractable likelihoods and supports efficient sampling and inference. One can easily impose symmetry constraints in the generative model by designing suitable scalar potential functions. We apply the approach to unsupervised density estimation of the MNIST dataset and variational calculation of the two-dimensional Ising model at the critical point. This approach brings insights and techniques from Monge-Amp\`ere equation, optimal transport, and fluid dynamics into reversible flow-based generative models. ""","""This paper develops a generative density model based on continuous-time flows on a potential field. Strengths: The paper contains interesting ideas and connections to physics, in particular the enforcement of symmetry in a computationally cheap way. Weaknesses: The main quantitative results of this paper are undercut by the numerical error introduced by the approximate, fixed-step integrator used. In the paper, the authors did not check the degree of numerical error (or to what extend their reported likelihoods do not normalize) as a function of the step size. This was partially addressed in a comment below. There does seem to be some novelty but the lack of concrete experiments is a letdown. One could e.g. verify that the samples have similar properties (e.g. moments) to the ground truth, which are known for the Ising model. Regarding clarity, the symmetry constraints are never clearly specified. This paper contains many ideas that would have been novel, but were scooped by [1] which was put on arXiv 3 months before the ICLR submission date. The authors have added appropriate references to this paper, but this still undercuts the originality of the contribution. The explanation of how and which symmetries are enforced is a little bit buried and unclear. Points of contention: Two of the reviewers didn't seem to be aware that the main mathematical results of the model are special cases of results from [1]. Consensus: All reviewers agreed that there were interesting ideas in the paper, and that it was close to the bar. [1] Chen, Tian Qi, et al. ""Neural Ordinary Differential Equations."""""
531,"""One-Shot High-Fidelity Imitation: Training Large-Scale Deep Nets with RL""","['Imitation Learning', 'Deep Learning']","""Humans are experts at high-fidelity imitation -- closely mimicking a demonstration, often in one attempt. Humans use this ability to quickly solve a task instance, and to bootstrap learning of new tasks. Achieving these abilities in autonomous agents is an open problem. In this paper, we introduce an off-policy RL algorithm (MetaMimic) to narrow this gap. MetaMimic can learn both (i) policies for high-fidelity one-shot imitation of diverse novel skills, and (ii) policies that enable the agent to solve tasks more efficiently than the demonstrators. MetaMimic relies on the principle of storing all experiences in a memory and replaying these to learn massive deep neural network policies by off-policy RL. This paper introduces, to the best of our knowledge, the largest existing neural networks for deep RL and shows that larger networks with normalization are needed to achieve one-shot high-fidelity imitation on a challenging manipulation task. The results also show that both types of policy can be learned from vision, in spite of the task rewards being sparse, and without access to demonstrator actions. ""","""The paper introduces a setting called high-fidelity imitation where the goal one-shot generalization to new trajectories in a given environment. The authors contrast this with more standard one-shot imitation approaches where one-shot generalization is to a task rather than a precise trajectory. The authors propose a technique that works off of only state information, which is coupled with an RL algorithm that learns from a replay buffer that is populated by the imitator. The authors emphasize that their approach can leverage very large deep learning models, and demonstrate strong empirical performance in a (simulated) robotics setting. A key weakness of the paper is its clarity. All reviewers were unclear about the precise setting as well as relation to prior work in one-shot imitation learning. As a result, there were substantial challenges in assessing the technical contribution of the paper. There were many requests for clarification, including for the motivation, difference between the present setting and those addressed in previous work, algorithmic details, and experiment details. I believe that a further concern was the lack of a wide range of baselines. The authors construct several baselines that are relevant in the given setting, but did not consider ""naive baseline"" approaches proposed by the reviewers. For example, behavior cloning is mentioned as a potential baseline several times. The authors argue that this is not applicable as it would require expert actions. Instead of considering it a baseline, BC could be used as an ""oracle"" - performance that could be achieved if demonstration actions were known. As long as the access to additional information is clearly marked, such a comparison with a privileged oracle can be properly placed by the reader. Without including such commonly known reference approaches, it is very challenging to assess the proposed method's performance in the context of the difficulty of the task. Generally, whenever a paper introduces both a new task and a new approach, a lot of care needs to be taken to build up insights into whether the task appropriately reflects the domain / challenge the paper claims to address, how challenging the task is in comparison to those addressed in prior work, and to place the performance of the novel proposed method in the context of prior work. In the present paper, on top of the task and approach being novel, the pure RL baseline D4PG is not yet widely known in the community and it's performance relative to common approaches is not well understood. Including commonly known RL approaches would help put all these results in context. The authors took great care to respond to the reviewer comments, providing thorough discussion of related work and clarifications of the task and approach, and these were very helpful to the AC to understand the paper. The AC believes that the paper has excellent potential. At the same time, a much more thorough empirical evaluation is needed to demonstrate the value of the proposed approach in this novel setting, as well as to provide additional conceptual insights into why and in what kinds of settings the algorithm performance well, or where its limitations are. """
532,"""Neural TTS Stylization with Adversarial and Collaborative Games""","['Text-To-Speech synthesis', 'GANs']","""The modeling of style when synthesizing natural human speech from text has been the focus of significant attention. Some state-of-the-art approaches train an encoder-decoder network on paired text and audio samples (x_txt, x_aud) by encouraging its output to reconstruct x_aud. The synthesized audio waveform is expected to contain the verbal content of x_txt and the auditory style of x_aud. Unfortunately, modeling style in TTS is somewhat under-determined and training models with a reconstruction loss alone is insufficient to disentangle content and style from other factors of variation. In this work, we introduce an end-to-end TTS model that offers enhanced content-style disentanglement ability and controllability. We achieve this by combining a pairwise training procedure, an adversarial game, and a collaborative game into one training scheme. The adversarial game concentrates the true data distribution, and the collaborative game minimizes the distance between real samples and generated samples in both the original space and the latent space. As a result, the proposed model delivers a highly controllable generator, and a disentangled representation. Benefiting from the separate modeling of style and content, our model can generate human fidelity speech that satisfies the desired style conditions. Our model achieves start-of-the-art results across multiple tasks, including style transfer (content and style swapping), emotion modeling, and identity transfer (fitting a new speaker's voice).""","""The paper proposes using GANs for disentangling style information from speech content, and thereby improve style transfer in TTS. The review and responses for this paper have been especially thorough! The authors significantly improved the paper during the review process, as pointed out by the reviewers. Inclusion of additional baselines, evaluations and ablation analysis helped improve the overall quality of the paper and helped alleviate concerns raised by the reviewers. Therefore, it is recommended that the paper be accepted for publication."""
533,"""Towards Language Agnostic Universal Representations""","['universal representations', 'language agnostic representations', 'NLP', 'GAN']","""When a bilingual student learns to solve word problems in math, we expect the student to be able to solve these problem in both languages the student is fluent in, even if the math lessons were only taught in one language. However, current representations in machine learning are language dependent. In this work, we present a method to decouple the language from the problem by learning language agnostic representations and therefore allowing training a model in one language and applying to a different one in a zero shot fashion. We learn these representations by taking inspiration from linguistics, specifically the Universal Grammar hypothesis and learn universal latent representations that are language agnostic (Chomsky, 2014; Montague, 1970). We demonstrate the capabilities of these representations by showing that the models trained on a single language using language agnostic representations achieve very similar accuracies in other languages.""","""This paper addresses a clear open problem in representation learning for language: the learning of language-agnostic representations for zero-shot cross-lingual transfer. All three reviewers agree that it makes some progress on that problem, and my understanding is that a straightforward presentation of these would likely have been accepted to this conference. However, there were serious issues with the framing and presentation of the paper. One reviewer expressed serious concerns about clarity and detail, and two others expressed serious concerns about the paper's framing. I'm more worried about the framing issue: The paper opens with a sweeping discussion about the nature of language and universal grammar and, in the original version, also claims (in vague terms) to have made substantial progress on understanding the nature of language. The most problematic claims have since been removed, but the sweeping introduction remains, and it serves as the only introduction to the paper, leaving little discussion of the substantial points that the paper is trying to make. I reluctantly have to recommend rejection. These problems should be fixable with a substantial re-write of the paper, but the reviewers were not satisfied with the progress made in that direction so far."""
534,"""q-Neurons: Neuron Activations based on Stochastic Jackson's Derivative Operators""","['q-calculus', 'neural activation function']","""We propose a new generic type of stochastic neurons, called pseudo-formula -neurons, that considers activation functions based on Jackson's pseudo-formula -derivatives, with stochastic parameters pseudo-formula . Our generalization of neural network architectures with pseudo-formula -neurons is shown to be both scalable and very easy to implement. We demonstrate experimentally consistently improved performances over state-of-the-art standard activation functions, both on training and testing loss functions. ""","""This paper proposes a new type of activations function based on q-calculus. The reviewers found that the papers is significantly lacking in its presentation, in clarity, and in its experimental evaluation. The motivation of the method raises several significant questions to the reviewers, and the proposed method is not sufficiently compared to existing approaches for (noisy) activation functions. After reviews, the authors have failed to present any updates to their paper."""
535,"""Meta Learning with Fast/Slow Learners""","['computer vision', 'meta learning']","""Meta-learning has recently achieved success in many optimization problems. In general, a meta learner g(.) could be learned for a base model f(.) on a variety of tasks, such that it can be more efficient on a new task. In this paper, we make some key modifications to enhance the performance of meta-learning models. (1) we leverage different meta-strategies for different modules to optimize them separately: we use conservative slow learners on low-level basic feature representation layers and fast learners on high-level task-specific layers; (2) Furthermore, we provide theoretical analysis on why the proposed approach works, based on a case study on a two-layer MLP. We evaluate our model on synthetic MLP regression, as well as low-shot learning tasks on Omniglot and ImageNet benchmarks. We demonstrate that our approach is able to achieve state-of-the-art performance.""","""The paper introduces an interesting idea of using different rates of learning for low level vs high level computation for meta learning. However, the experiments lack the thoroughness needed to justify the basic intuition of the approach and design choices like which layers to learn fast or slow need to be further ablated."""
536,"""Reconciling Feature-Reuse and Overfitting in DenseNet with Specialized Dropout""","['Specialized dropout', 'computer vision']","""Recently convolutional neural networks (CNNs) achieve great accuracy in visual recognition tasks. DenseNet becomes one of the most popular CNN models due to its effectiveness in feature-reuse. However, like other CNN models, DenseNets also face overfitting problem if not severer. Existing dropout method can be applied but not as effective due to the introduced nonlinear connections. In particular, the property of feature-reuse in DenseNet will be impeded, and the dropout effect will be weakened by the spatial correlation inside feature maps. To address these problems, we craft the design of a specialized dropout method from three aspects, dropout location, dropout granularity, and dropout probability. The insights attained here could potentially be applied as a general approach for boosting the accuracy of other CNN models with similar nonlinear connections. Experimental results show that DenseNets with our specialized dropout method yield better accuracy compared to vanilla DenseNet and state-of-the-art CNN models, and such accuracy boost increases with the model depth.""","""All reviewers recommend reject and there is no rebuttal. There is no basis on which to accept the paper."""
537,"""On the Margin Theory of Feedforward Neural Networks""","['generalization theory', 'implicit regularization', 'generalization', 'over-parametrization', 'theory', 'deep learning theory', 'margin']","""Past works have shown that, somewhat surprisingly, over-parametrization can help generalization in neural networks. Towards explaining this phenomenon, we adopt a margin-based perspective. We establish: 1) for multi-layer feedforward relu networks, the global minimizer of a weakly-regularized cross-entropy loss has the maximum normalized margin among all networks, 2) as a result, increasing the over-parametrization improves the normalized margin and generalization error bounds for deep networks. In the case of two-layer networks, an infinite-width neural network enjoys the best generalization guarantees. The typical infinite feature methods are kernel methods; we compare the neural net margin with that of kernel methods and construct natural instances where kernel methods have much weaker generalization guarantees. We validate this gap between the two approaches empirically. Finally, this infinite-neuron viewpoint is also fruitful for analyzing optimization. We show that a perturbed gradient flow on infinite-size networks finds a global optimizer in polynomial time.""","""This paper has received reviews from multiple experts who raise a litany of issues. These have been addressed quite convincingly by the authors, but I believe that ultimately this work needs to go through another round of reviewing, and this cannot be achieved in the context of ICLR's reviewing setup. I look forward to reading the final version of the paper in the near future."""
538,"""Unsupervised Adversarial Image Reconstruction""","['Deep Learning', 'Adversarial', 'MAP', 'GAN', 'neural networks']","""We address the problem of recovering an underlying signal from lossy, inaccurate observations in an unsupervised setting. Typically, we consider situations where there is little to no background knowledge on the structure of the underlying signal, no access to signal-measurement pairs, nor even unpaired signal-measurement data. The only available information is provided by the observations and the measurement process statistics. We cast the problem as finding the \textit{maximum a posteriori} estimate of the signal given each measurement, and propose a general framework for the reconstruction problem. We use a formulation of generative adversarial networks, where the generator takes as input a corrupted observation in order to produce realistic reconstructions, and add a penalty term tying the reconstruction to the associated observation. We evaluate our reconstructions on several image datasets with different types of corruptions. The proposed approach yields better results than alternative baselines, and comparable performance with model variants trained with additional supervision.""","""This paper proposes a GAN-based method to recover images from a noisy version of it. The paper builds upon existing works on AmbientGAN and CS-GAN. By combining the two approaches, the work finds a new method that performs better than existing approaches. The paper clearly has new interesting ideas which have been executed well. Two of the reviewers have voted in favour of acceptance, with one of the reviewer providing an extensive and detailed review. The third reviewer however has some doubts which were not resolved completely after the rebuttal. Upon reading the work myself, I am convinced that this will be interesting to the community. However, I will recommend the authors to take the comments of Reviewer 2 into account and do whatever it takes to resolve issues pointed by the reviewer. During the review process, another related work was found to be very similar to the approach discussed in this work. This work should be cited in the paper, as a prior work that the authors were unaware of. pseudo-url Please also discuss any new insights this work offers on top of this existing work. Given that the above suggestions are taken into account, I recommend to accept this paper. """
539,"""Feature-Wise Bias Amplification""","['bias', 'bias amplification', 'classification']","""We study the phenomenon of bias amplification in classifiers, wherein a machine learning model learns to predict classes with a greater disparity than the underlying ground truth. We demonstrate that bias amplification can arise via inductive bias in gradient descent methods resulting in overestimation of importance of moderately-predictive ``weak'' features if insufficient training data is available. This overestimation gives rise to feature-wise bias amplification -- a previously unreported form of bias that can be traced back to the features of a trained model. Through analysis and experiments, we show that the while some bias cannot be mitigated without sacrificing accuracy, feature-wise bias amplification can be mitigated through targeted feature selection. We present two new feature selection algorithms for mitigating bias amplification in linear models, and show how they can be adapted to convolutional neural networks efficiently. Our experiments on synthetic and real data demonstrate that these algorithms consistently lead to reduced bias without harming accuracy, in some cases eliminating predictive bias altogether while providing modest gains in accuracy.""","""The authors identify a source of bias that occurs when a model overestimates the importance of weak features in the regime where sufficient training data is not available. The bias is characterized theoretically, and demonstrated on synthetic and real datasets. The authors then present two algorithms to mitigate this bias, and demonstrate that they are effective in experimental evaluations. As noted by the reviewers, the work is well-motivated and clearly presented. Given the generally positive reviews, the AC recommends that the work be accepted. The authors should consider adding additional text describing the details concerning Figure 3 in the appendix. """
540,"""Kernel Change-point Detection with Auxiliary Deep Generative Models""","['deep kernel learning', 'generative models', 'kernel two-sample test', 'time series change-point detection']","""Detecting the emergence of abrupt property changes in time series is a challenging problem. Kernel two-sample test has been studied for this task which makes fewer assumptions on the distributions than traditional parametric approaches. However, selecting kernels is non-trivial in practice. Although kernel selection for the two-sample test has been studied, the insufficient samples in change point detection problem hinder the success of those developed kernel selection algorithms. In this paper, we propose KL-CPD, a novel kernel learning framework for time series CPD that optimizes a lower bound of test power via an auxiliary generative model. With deep kernel parameterization, KL-CPD endows kernel two-sample test with the data-driven kernel to detect different types of change-points in real-world applications. The proposed approach significantly outperformed other state-of-the-art methods in our comparative evaluation of benchmark datasets and simulation studies.""","""This paper proposes a new kernel learning framework for change point detection by using a generative model. The reviewers agree that the paper is interesting and useful for the community. One of the reviewer had some issues with the paper but those were resolved after the rebuttal. The other two reviewers have short reviews and somewhat low confidence, so it is difficult to tell how this paper stands among other that exist in the literature. Overall, given the consistent ratings from all the reviewers, I believe this paper can be accepted. """
541,"""Visual Semantic Navigation using Scene Priors""","['Visual Navigation', 'Scene Prior', 'Knowledge Graph', 'Graph Convolution Networks', 'Deep Reinforcement Learning']","""How do humans navigate to target objects in novel scenes? Do we use the semantic/functional priors we have built over years to efficiently search and navigate? For example, to search for mugs, we search cabinets near the coffee machine and for fruits we try the fridge. In this work, we focus on incorporating semantic priors in the task of semantic navigation. We propose to use Graph Convolutional Networks for incorporating the prior knowledge into a deep reinforcement learning framework. The agent uses the features from the knowledge graph to predict the actions. For evaluation, we use the AI2-THOR framework. Our experiments show how semantic knowledge improves the performance significantly. More importantly, we show improvement in generalization to unseen scenes and/or objects.""","""The authors propose an approach for visual navigation that leverages a semantic knowledge graph to ground and inform the policy of an RL agent. The agent uses a graphnet to learn relationships and support the navigation. The empirical protocol is sound and uses best practices, and the authors have added additional experiments during the revision period, in response to the reviewers' requests. However, there were some significant problems with the submission - there were no comparisons to other semantic navigation methods, the approach is somewhat convoluted and will not survive the test of time, and the authors did not conclusively show the value of their approach. The reviewers uniformly support the publication of this paper, but with a low confidence. """
542,"""Practical lossless compression with latent variables using bits back coding""","['compression', 'variational auto-encoders', 'deep latent gaussian models', 'lossless compression', 'latent variables', 'approximate inference', 'variational inference']","""Deep latent variable models have seen recent success in many data domains. Lossless compression is an application of these models which, despite having the potential to be highly useful, has yet to be implemented in a practical manner. We present '`Bits Back with ANS' (BB-ANS), a scheme to perform lossless compression with latent variable models at a near optimal rate. We demonstrate this scheme by using it to compress the MNIST dataset with a variational auto-encoder model (VAE), achieving compression rates superior to standard methods with only a simple VAE. Given that the scheme is highly amenable to parallelization, we conclude that with a sufficiently high quality generative model this scheme could be used to achieve substantial improvements in compression rate with acceptable running time. We make our implementation available open source at pseudo-url .""","""The paper proposes a novel lossless compression scheme that leverages latent-variable models such as VAEs. Its main original contribution is to improve the bits back coding scheme [B. Frey 1997] through the use of asymmetric numeral systems (ANS) instead of arithmetic coding. The developed practical algorithm is also able to use continuous latents. The paper is well written but the reader will benefit from prior familiarity with compression schemes. Resulting message bit-length is shown empirically to be close to ELBO on MNIST. The main weakness pointed out by reviewers is that the empirical evaluation is limited to MNIST and to a simple VAE, while applicability to other models (autoregressive) and data (PixelVAE on ImageNet) is only hinted to and expected bit-length merely extrapolated from previously reported log-likelihood. The work could be much more convincing if its compression was empirically demonstrated on larger and better models and larger scale data. Nevertheless reviewers agreed that it sufficiently advanced the field to warrant acceptance."""
543,"""LeMoNADe: Learned Motif and Neuronal Assembly Detection in calcium imaging videos""","['VAE', 'unsupervised learning', 'neuronal assemblies', 'calcium imaging analysis']","""Neuronal assemblies, loosely defined as subsets of neurons with reoccurring spatio-temporally coordinated activation patterns, or ""motifs"", are thought to be building blocks of neural representations and information processing. We here propose LeMoNADe, a new exploratory data analysis method that facilitates hunting for motifs in calcium imaging videos, the dominant microscopic functional imaging modality in neurophysiology. Our nonparametric method extracts motifs directly from videos, bypassing the difficult intermediate step of spike extraction. Our technique augments variational autoencoders with a discrete stochastic node, and we show in detail how a differentiable reparametrization and relaxation can be used. An evaluation on simulated data, with available ground truth, reveals excellent quantitative performance. In real video data acquired from brain slices, with no ground truth available, LeMoNADe uncovers nontrivial candidate motifs that can help generate hypotheses for more focused biological investigations.""","""This paper is about representation learning for calcium imaging and thus a bit different in scope that most ICLR submissions. But the paper is well-executed with good choices for the various parts of the model making it relevant for other similar domains."""
544,"""On the Selection of Initialization and Activation Function for Deep Neural Networks""","['Deep Neural Networks', 'Initialization', 'Gaussian Processes']","""The weight initialization and the activation function of deep neural networks have a crucial impact on the performance of the training procedure. An inappropriate selection can lead to the loss of information of the input during forward propagation and the exponential vanishing/exploding of gradients during back-propagation. Understanding the theoretical properties of untrained random networks is key to identifying which deep networks may be trained successfully as recently demonstrated by Schoenholz et al. (2017) who showed that for deep feedforward neural networks only a specific choice of hyperparameters known as the `edge of chaos' can lead to good performance. We complete this analysis by providing quantitative results showing that, for a class of ReLU-like activation functions, the information propagates indeed deeper for an initialization at the edge of chaos. By further extending this analysis, we identify a class of activation functions that improve the information propagation over ReLU-like functions. This class includes the Swish activation, = x \cdot \text{sigmoid}(x) used in Hendrycks & Gimpel (2016), Elfwing et al. (2017) and Ramachandran et al. (2017). This provides a theoretical grounding for the excellent empirical performance of pseudo-formula observed in these contributions. We complement those previous results by illustrating the benefit of using a random initialization on the edge of chaos in this context.""","""The paper attempts to extend the recent analysis of random deep networks to alternative activation functions. Unfortunately, none of the reviewers recommended the paper be accepted. The current presentation suffers from a lack of clarity and a sufficiently convincing supporting argument/evidence to satisfy the reviewers. The contribution is perceived as too incremental in light of previous work."""
545,"""Learning Neural Random Fields with Inclusive Auxiliary Generators""","['Neural random fields', 'Deep generative models', 'Unsupervised learning', 'Semi-supervised learning']","""Neural random fields (NRFs), which are defined by using neural networks to implement potential functions in undirected models, provide an interesting family of model spaces for machine learning. In this paper we develop a new approach to learning NRFs with inclusive-divergence minimized auxiliary generator - the inclusive-NRF approach, for continuous data (e.g. images), with solid theoretical examination on exploiting gradient information in model sampling. We show that inclusive-NRFs can be flexibly used in unsupervised/supervised image generation and semi-supervised classification, and empirically to the best of our knowledge, represent the best-performed random fields in these tasks. Particularly, inclusive-NRFs achieve state-of-the-art sample generation quality on CIFAR-10 in both unsupervised and supervised settings. Semi-supervised inclusive-NRFs show strong classification results on par with state-of-the-art generative model based semi-supervised learning methods, and simultaneously achieve superior generation, on the widely benchmarked datasets - MNIST, SVHN and CIFAR-10.""","""This paper proposes a method for learning neural RFs with the inclusive-divergence minimization problem. Reviewers generally agree that the experiments are sufficient and convincing, and that the method is evaluated well. Results are comparable with SOTA methods for image generation. The paper is reasonably well-written. The paper is also somewhat lacking in background; most people at ICLR will not be very familiar with this learning problem. More information on the inclusive-divergence minimization problem would be helpful. A major concern of reviewers is whether novelty of the method is sufficient for publication. """
546,"""Understanding Straight-Through Estimator in Training Activation Quantized Neural Nets""","['straight-through estimator', 'quantized activation', 'binary neuron']","""Training activation quantized neural networks involves minimizing a piecewise constant training loss whose gradient vanishes almost everywhere, which is undesirable for the standard back-propagation or chain rule. An empirical way around this issue is to use a straight-through estimator (STE) (Bengio et al., 2013) in the backward pass only, so that the ""gradient"" through the modified chain rule becomes non-trivial. Since this unusual ""gradient"" is certainly not the gradient of loss function, the following question arises: why searching in its negative direction minimizes the training loss? In this paper, we provide the theoretical justification of the concept of STE by answering this question. We consider the problem of learning a two-linear-layer network with binarized ReLU activation and Gaussian input data. We shall refer to the unusual ""gradient"" given by the STE-modifed chain rule as coarse gradient. The choice of STE is not unique. We prove that if the STE is properly chosen, the expected coarse gradient correlates positively with the population gradient (not available for the training), and its negation is a descent direction for minimizing the population loss. We further show the associated coarse gradient descent algorithm converges to a critical point of the population loss minimization problem. Moreover, we show that a poor choice of STE leads to instability of the training algorithm near certain local minima, which is verified with CIFAR-10 experiments.""","""The paper contributes to the understanding of straight-through estimation for single hidden layer neural networks, revealing advantages for ReLU and clipped ReLU over identity activations. A thorough and convincing theoretical analysis is provided to support these findings. After resolving various issues during the response period, the reviewers concluded with a unanimous recommendation of acceptance. Valid criticisms of the presentation quality were raised during the review and response period, and the authors would be well served by continuing to improve the paper's clarity."""
547,"""Exploring Curvature Noise in Large-Batch Stochastic Optimization""","['optimization', 'large-batch training', 'generalization', 'noise covariance']","""Using stochastic gradient descent (SGD) with large batch-sizes to train deep neural networks is an increasingly popular technique. By doing so, one can improve parallelization by scaling to multiple workers (GPUs) and hence leading to significant reductions in training time. Unfortunately, a major drawback is the so-called generalization gap: large-batch training typically leads to a degradation in generalization performance of the model as compared to small-batch training. In this paper, we propose to correct this generalization gap by adding diagonal Fisher curvature noise to large-batch gradient updates. We provide a theoretical analysis of our method in the convex quadratic setting. Our empirical study with state-of-the-art deep learning models shows that our method not only improves the generalization performance in large-batch training but furthermore, does so in a way where the training convergence remains desirable and the training duration is not elongated. We additionally connect our method to recent works on loss surface landscape in the experimental section. ""","""Dear authors, Your proposition of adding a noise scaling with the diagonal of the gradient covariance to the updates as a middle-ground between the identity and the full covariance is interesting and tackles the timely question of the links between optimization and generalization. However, the reviewers had concerns about the experiments that did not reveal to which extent each trick had an influence. I would like to add that, even though the term Fisher is used for both the true Fisher and tne empirical one, these two matrices encore very different kind of information. In particular, the latter is only defined when there is a dataset. Hence, your case study (section 3.2) which uses the true Fisher does not apply to the empirical Fisher. I encourage the authors to pursue in this direction but to update the experimental section in order to highlight the impact of each technique used."""
548,"""FEED: Feature-level Ensemble Effect for knowledge Distillation""","['Knowledge Distillation', 'Ensemble Effect', 'Knowledge Transfer']","""This paper proposes a versatile and powerful training algorithm named Feature-level Ensemble Effect for knowledge Distillation(FEED), which is inspired by the work of factor transfer. The factor transfer is one of the knowledge transfer methods that improves the performance of a student network with a strong teacher network. It transfers the knowledge of a teacher in the feature map level using high-capacity teacher network, and our training algorithm FEED is an extension of it. FEED aims to transfer ensemble knowledge, using either multiple teachers in parallel or multiple training sequences. Adapting the peer-teaching framework, we introduce a couple of training algorithms that transfer ensemble knowledge to the student at the feature map level, both of which help the student network find more generalized solutions in the parameter space. Experimental results on CIFAR-100 and ImageNet show that our method, FEED, has clear performance enhancements,without introducing any additional parameters or computations at test time.""","""The paper describes knowledge distillation methods. As noted by all reviewers, the methods are very similar to the prior art, so there is not enough novelty for the paper to be accepted. The reviewers' opinion didn't change after the rebuttal."""
549,"""NEURAL MALWARE CONTROL WITH DEEP REINFORCEMENT LEARNING""","['malware', 'execution', 'control', 'deep reinforcement learning']","""Antimalware products are a key component in detecting malware attacks, and their engines typically execute unknown programs in a sandbox prior to running them on the native operating system. Files cannot be scanned indefinitely so the engine employs heuristics to determine when to halt execution. Previous research has investigated analyzing the sequence of system calls generated during this emulation process to predict if an unknown file is malicious, but these models require the emulation to be stopped after executing a fixed number of events from the beginning of the file. Also, these classifiers are not accurate enough to halt emulation in the middle of the file on their own. In this paper, we propose a novel algorithm which overcomes this limitation and learns the best time to halt the file's execution based on deep reinforcement learning (DRL). Because the new DRL-based system continues to emulate the unknown file until it can make a confident decision to stop, it prevents attackers from avoiding detection by initiating malicious activity after a fixed number of system calls. Results show that the proposed malware execution control model automatically halts emulation for 91.3\% of the files earlier than heuristics employed by the engine. Furthermore, classifying the files at that time improves the true positive rate by 61.5%, at a false positive rate of 1%, compared to a baseline classifier.""","""The paper trains a classifier to decide if a program is a malware and when to halt its execution. The malware classifier is mostly composed of an RNN acting on featurized API calls (events). The presentation could be improved. The results are encouraging, but the experiments lack solid baselines, comparisons, and grounding of the task usefulness, as this is not done on an established benchmark."""
550,"""Information Theoretic lower bounds on negative log likelihood""","['latent variable modeling', 'rate-distortion theory', 'log likelihood bounds']","""In this article we use rate-distortion theory, a branch of information theory devoted to the problem of lossy compression, to shed light on an important problem in latent variable modeling of data: is there room to improve the model? One way to address this question is to find an upper bound on the probability (equivalently a lower bound on the negative log likelihood) that the model can assign to some data as one varies the prior and/or the likelihood function in a latent variable model. The core of our contribution is to formally show that the problem of optimizing priors in latent variable models is exactly an instance of the variational optimization problem that information theorists solve when computing rate-distortion functions, and then to use this to derive a lower bound on negative log likelihood. Moreover, we will show that if changing the prior can improve the log likelihood, then there is a way to change the likelihood function instead and attain the same log likelihood, and thus rate-distortion theory is of relevance to both optimizing priors as well as optimizing likelihood functions. We will experimentally argue for the usefulness of quantities derived from rate-distortion theory in latent variable modeling by applying them to a problem in image modeling.""","""Strengths: This paper gives a detailed treatment of the connections between rate distortion theory and variational lower bounds, culminating in a practical diagnostic tool. The paper is well-written. Weaknesses: Many of the theoretical results existed in older work. Points of contention: Most of the discussion was about the novelty of the lower bound. Consensus: R3 and R2 both appear to recommend acceptance (R2 in a comment), and have both clearly given the paper detailed thought."""
551,"""Hierarchical RL Using an Ensemble of Proprioceptive Periodic Policies""",[],"""In this paper we introduce a simple, robust approach to hierarchically training an agent in the setting of sparse reward tasks. The agent is split into a low-level and a high-level policy. The low-level policy only accesses internal, proprioceptive dimensions of the state observation. The low-level policies are trained with a simple reward that encourages changing the values of the non-proprioceptive dimensions. Furthermore, it is induced to be periodic with the use a ``phase function.'' The high-level policy is trained using a sparse, task-dependent reward, and operates by choosing which of the low-level policies to run at any given time. Using this approach, we solve difficult maze and navigation tasks with sparse rewards using the Mujoco Ant and Humanoid agents and show improvement over recent hierarchical methods. ""","""Strengths The paper presents a method of training two-level hierarchies that is based on relatively intuitive ideas and that performs well. The challenges of hierarchical RL makes this an important problem. The benefits of periodicity and the separation of internal state from external state is a clean principle that can potentially be broadly employed. The method does well in outperforming the alternative baselines. Weaknesses There is no video of the results. There is related work, i.e., [Peng et al. 2016] (rev 4) uses a policy ensemble; phase info is used in DeepLoco/DeepMimic; methods such as ""Virtual Windup Toys for Animation"" exploited periodicity (25y ago); More comparisons with prior work such as Florensa et al. would help. The separation of internal and external state is an assumption that may not hold in many cases. The results are locomotion focussed. There are only two timescales. Decision The reviewers are largely in agreement to accept the paper. There are fairly-simple-but-useful lessons to be found in the paper for those working on HRL problems, particularly those for movement and locomotion. The AC sees the novely with respect to different pieces of related work is the weakest point of the paper. The reviews contain good suggestions for revisions and improvements; the latest version may take care of these (uploaded after the last reviewer comments). Overall, the paper will make a good contribution to ICLR 2019. """
552,"""Differentially Private Federated Learning: A Client Level Perspective""","['Machine Learning', 'Federated Learning', 'Privacy', 'Security', 'Differential Privacy']","""Federated learning is a recent advance in privacy protection. In this context, a trusted curator aggregates parameters optimized in decentralized fashion by multiple clients. The resulting model is then distributed back to all clients, ultimately converging to a joint representative model without explicitly having to share the data. However, the protocol is vulnerable to differential attacks, which could originate from any party contributing during federated optimization. In such an attack, a client's contribution during training and information about their data set is revealed through analyzing the distributed model. We tackle this problem and propose an algorithm for client sided differential privacy preserving federated optimization. The aim is to hide clients' contributions during training, balancing the trade-off between privacy loss and model performance. Empirical studies suggest that given a sufficiently large number of participating clients, our proposed procedure can maintain client-level differential privacy at only a minor cost in model performance. ""","""Following the unanimous vote of the reviewers, this paper is not ready for publication at ICLR. The greatest concern was that the novelty beyond past work has not been sufficiently demonstrated."""
553,"""Regularized Learning for Domain Adaptation under Label Shifts""","['Deep Learning', 'Domain Adaptation', 'Label Shift', 'Importance Weights', 'Generalization']","""We propose Regularized Learning under Label shifts (RLLS), a principled and a practical domain-adaptation algorithm to correct for shifts in the label distribution between a source and a target domain. We first estimate importance weights using labeled source data and unlabeled target data, and then train a classifier on the weighted source samples. We derive a generalization bound for the classifier on the target domain which is independent of the (ambient) data dimensions, and instead only depends on the complexity of the function class. To the best of our knowledge, this is the first generalization bound for the label-shift problem where the labels in the target domain are not available. Based on this bound, we propose a regularized estimator for the small-sample regime which accounts for the uncertainty in the estimated weights. Experiments on the CIFAR-10 and MNIST datasets show that RLLS improves classification accuracy, especially in the low sample and large-shift regimes, compared to previous methods.""","""The paper gives a novel algorithm for transfer learning with label distribution shift with provably guarantees. As the reviewers pointed out, the pros include: 1) a solid and motivated algorithm for a understudied problem 2) the algorithm is implemented empirically and gives good performance. The drawback includes incomplete/unclear comparison with previous work. The authors claimed that the code of the previous work cannot be completed within a reasonable amount of time. The AC decided that the paper could be accepted without such a comparison, but the authors are strongly urged to clarify this point or include the comparison for a smaller dataset in the final revision if possible. """
554,"""Augmented Cyclic Adversarial Learning for Low Resource Domain Adaptation""","['Domain adaptation', 'generative adversarial network', 'cyclic adversarial learning', 'speech']","""Training a model to perform a task typically requires a large amount of data from the domains in which the task will be applied. However, it is often the case that data are abundant in some domains but scarce in others. Domain adaptation deals with the challenge of adapting a model trained from a data-rich source domain to perform well in a data-poor target domain. In general, this requires learning plausible mappings between domains. CycleGAN is a powerful framework that efficiently learns to map inputs from one domain to another using adversarial training and a cycle-consistency constraint. However, the conventional approach of enforcing cycle-consistency via reconstruction may be overly restrictive in cases where one or more domains have limited training data. In this paper, we propose an augmented cyclic adversarial learning model that enforces the cycle-consistency constraint via an external task specific model, which encourages the preservation of task-relevant content as opposed to exact reconstruction. We explore digit classification in a low-resource setting in supervised, semi and unsupervised situation, as well as high resource unsupervised. In low-resource supervised setting, the results show that our approach improves absolute performance by 14% and 4% when adapting SVHN to MNIST and vice versa, respectively, which outperforms unsupervised domain adaptation methods that require high-resource unlabeled target domain. Moreover, using only few unsupervised target data, our approach can still outperforms many high-resource unsupervised models. Our model also outperforms on USPS to MNIST and synthetic digit to SVHN for high resource unsupervised adaptation. In speech domains, we similarly adopt a speech recognition model from each domain as the task specific model. Our approach improves absolute performance of speech recognition by 2% for female speakers in the TIMIT dataset, where the majority of training samples are from male voices.""","""The authors propose a method for low-resource domain adaptation where the number of examples available in the target domain are limited. The proposed method modifies the basic approach in a CycleGAN by augmenting it with a content (task-specific) loss, instead of the standard reconstruction error. The authors also demonstrate experimentally that it is important to enforce the loss in both directions (target source and source --> target). Experiments are conducted on both supervised as well as unsupervised settings. The main concern expressed by the reviewers relates to the novelty of the approach since it is a relatively straightforward extension of CycleGAN/CyCADA, but in the view of a majority of reviewers the work serves a useful contribution as a practical method for developing systems in low-resource conditions where it is feasible to label a few new instances. Although the reviewers were not unanimous in their recommendations, on balance in the view of the AC the work is a useful contribution with clear and detailed experiments in the revised version. """
555,"""Distinguishability of Adversarial Examples""","['Adversarial Examples', 'Machine Learning', 'Neural Networks', 'Distinguishability', 'Defense']","""Machine learning models including traditional models and neural networks can be easily fooled by adversarial examples which are generated from the natural examples with small perturbations. This poses a critical challenge to machine learning security, and impedes the wide application of machine learning in many important domains such as computer vision and malware detection. Unfortunately, even state-of-the-art defense approaches such as adversarial training and defensive distillation still suffer from major limitations and can be circumvented. From a unique angle, we propose to investigate two important research questions in this paper: Are adversarial examples distinguishable from natural examples? Are adversarial examples generated by different methods distinguishable from each other? These two questions concern the distinguishability of adversarial examples. Answering them will potentially lead to a simple yet effective approach, termed as defensive distinction in this paper under the formulation of multi-label classification, for protecting against adversarial examples. We design and perform experiments using the MNIST dataset to investigate these two questions, and obtain highly positive results demonstrating the strong distinguishability of adversarial examples. We recommend that this unique defensive distinction approach should be seriously considered to complement other defense approaches.""","""The paper investigates an interesting question and points at a promising research direction in relation to whether adversarial examples are distinguishable from natural examples. A concern raised in the reviews is that the technical contribution of the paper is weak. A main concern with the paper is that the experiments have been conducted only on one simple data set. The authors proposed to add more experiments and improve other points, but a revision didn't follow. The reviewers consistently rate the paper as ok, but not good enough. I would encourage the authors to conduct the improvements proposed by the reviewers and the authors themselves. """
556,"""Like What You Like: Knowledge Distill via Neuron Selectivity Transfer""",['Knowledge Distill'],"""Despite deep neural networks have demonstrated extraordinary power in various applications, their superior performances are at expense of high storage and computational costs. Consequently, the acceleration and compression of neural networks have attracted much attention recently. Knowledge Transfer (KT), which aims at training a smaller student network by transferring knowledge from a larger teacher model, is one of the popular solutions. In this paper, we propose a novel knowledge transfer method by treating it as a distribution matching problem. Particularly, we match the distributions of neuron selectivity patterns between teacher and student networks. To achieve this goal, we devise a new KT loss function by minimizing the Maximum Mean Discrepancy (MMD) metric between these distributions. Combined with the original loss function, our method can significantly improve the performance of student networks. We validate the effectiveness of our method across several datasets, and further combine it with other KT methods to explore the best possible results. Last but not least, we fine-tune the model to other tasks such as object detection. The results are also encouraging, which confirm the transferability of the learned features.""",""" The paper presents a sensible algorithm for knowledge distillation (KD) from a larger teacher network to a smaller student network by minimizing the Maximum Mean Discrepancy (MMD) between the distributions over students and teachers network activations. As rightly acknowledged by the R3, the benefits of the proposed approach are encouraging in the object detection task, and are less obvious in classification (R1 and R2). The reviewers and AC note the following potential weaknesses: (1) low technical novelty in light of prior works Demystifying Neural Style Transfer by Li et al 2017 and Deep Transfer Learning with Joint Adaptation Networks by Long et al 2017 -- See R2s detailed explanations; (2) lack of empirical evidence that the proposed method is better than the seminal work on KD by Hinton et al, 2014; (3) important practical issues are not justified (e.g. kernel specifications as requested by R3 and R2; accuracy-efficiency trade-off as suggested by R1); (4) presentation clarity. R3 has raised questions regarding deploying the proposed student models on mobile devices without a proper comparison with the MobileNet and ShuffleNet light architectures. This can be seen as a suggestion for future revisions. There is reviewer disagreement on this paper and no author rebuttal. The reviewer with a positive view on the manuscript (R3) was reluctant to champion the paper as the authors did not respond to the concerns of the reviewers. AC suggests in its current state the manuscript is not ready for a publication. We hope the reviews are useful for improving and revising the paper. """
557,"""On the Convergence and Robustness of Batch Normalization""","['Batch normalization', 'Convergence analysis', 'Gradient descent', 'Ordinary least squares', 'Deep neural network']","""Despite its empirical success, the theoretical underpinnings of the stability, convergence and acceleration properties of batch normalization (BN) remain elusive. In this paper, we attack this problem from a modelling approach, where we perform thorough theoretical analysis on BN applied to simplified model: ordinary least squares (OLS). We discover that gradient descent on OLS with BN has interesting properties, including a scaling law, convergence for arbitrary learning rates for the weights, asymptotic acceleration effects, as well as insensitivity to choice of learning rates. We then demonstrate numerically that these findings are not specific to the OLS problem and hold qualitatively for more complex supervised learning problems. This points to a new direction towards uncovering the mathematical principles that underlies batch normalization.""","""The reviewers agree that providing more insights on why batch normalization work is an important topic of investigation, but they all raised several problems with the current submission which need to be addressed before publication. The AC thus proposes ""revise and sesubmit""."""
558,"""On the Spectral Bias of Neural Networks""","['deep learning theory', 'fourier analysis']","""Neural networks are known to be a class of highly expressive functions able to fit even random input-output mappings with 100% accuracy. In this work we present properties of neural networks that complement this aspect of expressivity. By using tools from Fourier analysis, we show that deep ReLU networks are biased towards low frequency functions, meaning that they cannot have local fluctuations without affecting their global behavior. Intuitively, this property is in line with the observation that over-parameterized networks find simple patterns that generalize across data samples. We also investigate how the shape of the data manifold affects expressivity by showing evidence that learning high frequencies gets easier with increasing manifold complexity, and present a theoretical understanding of this behavior. Finally, we study the robustness of the frequency components with respect to parameter perturbation, to develop the intuition that the parameters must be finely tuned to express high frequency functions.""","""This paper considers an interesting hypothesis that ReLU networks are biased towards learning learn low frequency Fourier components, showing a spectral bias towards low frequency functions. The paper backs the hypothesis with theoretical results computing and bounding the Fourier coefficients of ReLU networks and experiments on synthetic datasets. All reviewers find the topic to be interesting and important. However they find the results in the paper to be preliminary and not yet ready for publication. On theoretical front, the paper characterizes the Fourier coefficients for a given piecewise linear region of a ReLU network. However the bounds on Fourier coefficients of the entire network in Theorem 1 seem weak as they depend on number of pieces (N_f) and max Lipschitz constant over all pieces (L_f), quantities that can easily be exponentially big. Authors in their response have said that their bound on Fourier coefficients is tight. If so then the paper needs to discuss/prove why quantities N_f and L_f are expected to be small. Such a discussion will help reviewers in appreciating the theoretical contributions more. On experimental front, the paper does not show spectral bias of networks trained over any real datasets. Reviewers are sympathetic to the challenge of evaluating Fourier coefficients of the network trained on real data sets, but the paper does not outline any potential approach to attack this problem. I strongly suggest authors to address these reviewer concerns before next submission. """
559,"""Open Loop Hyperparameter Optimization and Determinantal Point Processes""","['hyperparameter optimization', 'black box optimization']","""Driven by the need for parallelizable hyperparameter optimization methods, this paper studies open loop search methods: sequences that are predetermined and can be generated before a single configuration is evaluated. Examples include grid search, uniform random search, low discrepancy sequences, and other sampling distributions. In particular, we propose the use of k-determinantal point processes in hyperparameter optimization via random search. Compared to conventional uniform random search where hyperparameter settings are sampled independently, a k-DPP promotes diversity. We describe an approach that transforms hyperparameter search spaces for efficient use with a k-DPP. In addition, we introduce a novel Metropolis-Hastings algorithm which can sample from k-DPPs defined over any space from which uniform samples can be drawn, including spaces with a mixture of discrete and continuous dimensions or tree structure. Our experiments show significant benefits in realistic scenarios with a limited budget for training supervised learners, whether in serial or parallel.""","""This is a very clearly written, well composed paper that does a good job of placing the proposed contribution in the scope of hyperparameter optimization techniques. This paper certainly appears to have been improved over the version submitted to the previous ICLR. In particular, the writing is much clearer and easy to follow and the methodology and experiments have been improved. The ideas are well motivated and it's exciting to see that sampling from a k-DPP can give better low discrepancy sequences than e.g. Sobol. However, the reviewers still seem to have two major concerns, namely novelty of the approach (DPPs have been used for Bayesian optimization before) and the empirical evaluation. Empirical evaluation: As Reviewer1 notes, there are much more recent approaches for Bayesian optimization that have improved significantly over the TPE method, also for conditional parameters. There are also more recent approaches proposing variants of random search such as hyperband. Novelty: There is some work on using determinantal point processes for Bayesian optimization and related work in optimal experimental design. Optimal design has a significant amount of literature dedicated to designing a set of experiments according to the determinant of their covariance matrix - i.e. D-Optimal Design. This work may add some interesting contributions to that literature, including fast sampling from k-DPPs, etc. It would be useful, however, to add some discussion of that literature in the paper. Jegelka and Sra's tutorial at NeurIPS on negative dependence had a nice overview of some of this literature. Unfortunately, two of the three reviewers thought the paper was just below the borderline and none of the reviewers were willing to champion it. There are very promising and interesting ideas in the paper, however, that have a lot of potential. In the opinion of the AC, one of the most powerful aspects of DPPs over e.g. low discrepancy sequences, random search, etc. is the ability to learn a distance over a space under which samples will be diverse. This can make a search *much* more efficient since (as the authors note when discussing random search vs. grid search) the DPP can sample more densely in areas and dimensions that have higher sensitivity. It would be exciting to learn kernels specifically for hyperparameter optimization problems (e.g. a kernel specifically for learning rates that can capture e.g. logarithmic scaling). Taking the objective into account through the quality score, as proposed for future work, also seems very sensible and could significantly improve results as well. """
560,"""Training Hard-Threshold Networks with Combinatorial Search in a Discrete Target Propagation Setting""","['hard-threshold network', 'combinatorial optimization', 'search', 'target propagation']","""Learning deep neural networks with hard-threshold activation has recently become an important problem due to the proliferation of resource-constrained computing devices. In order to circumvent the inability to train with backpropagation in the present of hard-threshold activations, \cite{friesen2017} introduced a discrete target propagation framework for training hard-threshold networks in a layer-by-layer fashion. Rather than using a gradient-based target heuristic, we explore the use of search methods for solving the target setting problem. Building on both traditional combinatorial optimization algorithms and gradient-based techniques, we develop a novel search algorithm Guided Random Local Search (GRLS). We demonstrate the effectiveness of our algorithm in training small networks on several datasets and evaluate our target-setting algorithm compared to simpler search methods and gradient-based techniques. Our results indicate that combinatorial optimization is a viable method for training hard-threshold networks that may have the potential to eventually surpass gradient-based methods in many settings. ""","""The paper proposes a novel local combinatorial search algorithm for the discrete target propagation framework of Friesen & Domingos 2018, and shows a few promising empirical results. Reviewers found the paper well written and clear, and two of them were enthusiastic about the direction of this research. But all reviewers agreed that the paper is too preliminary, particularly in its empirical coverage. More extensive experiments are needed to compare with competitive approaches form the literature, for the task of training hard-threshold networks. Experiments would need to evaluate the algorithms on larger models and data more representative of the field, to measure how the approach can scale, and to convince of the superiority or advantage of the proposed method."""
561,"""N-Ary Quantization for CNN Model Compression and Inference Acceleration""","['low-resource deep neural networks', 'quantized weights', 'weight-clustering', 'resource efficient neural networks']","""The tremendous memory and computational complexity of Convolutional Neural Networks (CNNs) prevents the inference deployment on resource-constrained systems. As a result, recent research focused on CNN optimization techniques, in particular quantization, which allows weights and activations of layers to be represented with just a few bits while achieving impressive prediction performance. However, aggressive quantization techniques still fail to achieve full-precision prediction performance on state-of-the-art CNN architectures on large-scale classification tasks. In this work we propose a method for weight and activation quantization that is scalable in terms of quantization levels (n-ary representations) and easy to compute while maintaining the performance close to full-precision CNNs. Our weight quantization scheme is based on trainable scaling factors and a nested-means clustering strategy which is robust to weight updates and therefore exhibits good convergence properties. The flexibility of nested-means clustering enables exploration of various n-ary weight representations with the potential of high parameter compression. For activations, we propose a linear quantization strategy that takes the statistical properties of batch normalization into account. We demonstrate the effectiveness of our approach using state-of-the-art models on ImageNet.""","""The submission proposes a hierarchical clustering approach (nested-means clustering) to determine good quantization intervals for non-uniform quantization. An empirical validation shows improvement over a very closely related approach (Zhu et al, 2016). There was an overall consensus that the literature review was insufficient in its initial form. The authors have proposed to extend it somewhat. Other concerns are related to the novelty of the technique (R4 was particularly concerned about novelty over Zhu et al, 2016). Two reviewers were against acceptance, and one was positive about the paper. Due to the overall concerns about the novelty of the approach, and that these concerns were confirmed in discussion after the rebuttal, this paper is unlikely to meet the threshold for acceptance to ICLR."""
562,"""PA-GAN: Improving GAN Training by Progressive Augmentation""","['Deep Learning', 'GANs', 'Augmentation', 'Generative Modelling']","""Despite recent progress, Generative Adversarial Networks (GANs) still suffer from training instability, requiring careful consideration of architecture design choices and hyper-parameter tuning. The reason for this fragile training behaviour is partially due to the discriminator performing well very quickly; its loss converges to zero, providing no reliable backpropagation signal to the generator. In this work we introduce a new technique - progressive augmentation of GANs (PA-GAN) - that helps to overcome this fundamental limitation and improve the overall stability of GAN training. The key idea is to gradually increase the task difficulty of the discriminator by progressively augmenting its input space, thus enabling continuous learning of the generator. We show that the proposed progressive augmentation preserves the original GAN objective, does not bias the optimality of the discriminator and encourages the healthy competition between the generator and discriminator, leading to a better-performing generator. We experimentally demonstrate the effectiveness of the proposed approach on multiple benchmarks (MNIST, Fashion-MNIST, CIFAR10, CELEBA) for the image generation task.""","""The submission hypothesizes that in typical GAN training the discriminator is too strong, too fast, and thus suggests a modification by which they gradually increases the task difficulty of the discriminator. This is done by introducing (effectively) a new random variable -- which has an effect on the label -- and which prevents the discriminator from solving its task too quickly. There was a healthy amount of back-and-forth between the authors and the reviewers which allowed for a number of important clarifications to be made (esp. with regards to proofs, comparison with baselines, etc). My judgment of this paper is that it provides a neat way to overcome a particular difficulty of training GANs, but that there is a lot of confusion about the similarities (of lack thereof) with various potentially simpler alternatives such as input dropout, adding noise to the input etc. I was sometimes confused by the author response as well (they at once suggest that the proposed method reduces overfitting of the discriminator but also state that ""We believe our method does not even try to regularize the discriminator""). Because of all this, the significance of this work is unclear and thus I do not recommend acceptance."""
563,"""An Empirical study of Binary Neural Networks' Optimisation""","['binary neural networks', 'quantized neural networks', 'straight-through-estimator']","""Binary neural networks using the Straight-Through-Estimator (STE) have been shown to achieve state-of-the-art results, but their training process is not well-founded. This is due to the discrepancy between the evaluated function in the forward path, and the weight updates in the back-propagation, updates which do not correspond to gradients of the forward path. Efficient convergence and accuracy of binary models often rely on careful fine-tuning and various ad-hoc techniques. In this work, we empirically identify and study the effectiveness of the various ad-hoc techniques commonly used in the literature, providing best-practices for efficient training of binary models. We show that adapting learning rates using second moment methods is crucial for the successful use of the STE, and that other optimisers can easily get stuck in local minima. We also find that many of the commonly employed tricks are only effective towards the end of the training, with these methods making early stages of the training considerably slower. Our analysis disambiguates necessary from unnecessary ad-hoc techniques for training of binary neural networks, paving the way for future development of solid theoretical foundations for these. Our newly-found insights further lead to new procedures which make training of existing binary neural networks notably faster.""","""The paper summarizes existing work on binary neural network optimization and performs an empirical study across a few datasets and neural network architectures. I agree with the reviewers that this is a valuable study and it can establish a benchmark to help practitioners develop better binary neural network optimization techniques. PS: How about ""An empirical study of binary neural network optimization"" as the title? """
564,"""Learning to Decompose Compound Questions with Reinforcement Learning""","['Compound Question Decomposition', 'Reinforcement Learning', 'Knowledge-Based Question Answering', 'Learning-to-decompose']","""As for knowledge-based question answering, a fundamental problem is to relax the assumption of answerable questions from simple questions to compound questions. Traditional approaches firstly detect topic entity mentioned in questions, then traverse the knowledge graph to find relations as a multi-hop path to answers, while we propose a novel approach to leverage simple-question answerers to answer compound questions. Our model consists of two parts: (i) a novel learning-to-decompose agent that learns a policy to decompose a compound question into simple questions and (ii) three independent simple-question answerers that classify the corresponding relations for each simple question. Experiments demonstrate that our model learns complex rules of compositionality as stochastic policy, which benefits simple neural networks to achieve state-of-the-art results on WebQuestions and MetaQA. We analyze the interpretable decomposition process as well as generated partitions.""","""+ an interesting task -- learning to decompose questions without supervision - reviewers are not convinced by evaluation. Initially evaluated on MetaQA only, later relation classification on WebQuestions has been added. It is not really clear that the approach is indeed beneficial on WebQuestion relation classification (no analysis / ablations) and MetaQA is not a very standard dataset. - Reviewers have concerns about comparison to previous work / the lack of state-of-the-art baselines. Some of these issues have been addressed though (e.g., discussion of Iyyer et al. 2016) """
565,"""Adversarially Learned Mixture Model""","['Unsupervised', 'Semi-supervised', 'Generative', 'Adversarial', 'Clustering']","""The Adversarially Learned Mixture Model (AMM) is a generative model for unsupervised or semi-supervised data clustering. The AMM is the first adversarially optimized method to model the conditional dependence between inferred continuous and categorical latent variables. Experiments on the MNIST and SVHN datasets show that the AMM allows for semantic separation of complex data when little or no labeled data is available. The AMM achieves unsupervised clustering error rates of 3.32% and 20.4% on the MNIST and SVHN datasets, respectively. A semi-supervised extension of the AMM achieves a classification error rate of 5.60% on the SVHN dataset.""","""The paper presents a method for unsupervised/semi-supervised clustering, combining adversarial learning and the Mixture of Gaussians model. The authors follow the methodology of ALI, extending the Q and P models with discrete variables, in such a way that the latent space in the P model comprises a mixture-of-Gaussians model. The problem of generative modeling and semi-supervised learning are interesting topics for the ICLR community. The reviewers think that the novelty of the method is unclear. The technique appears to be a mix of various pre-existing techniques, combined with a novel choice of model. The experimental results are somewhat promising, and it is encouraging to see that good generative model results are consistent with improved semi-supervised classification results. The paper seems to rely heavily on empirical results, but they are difficult to verify without published source code. The datasets chosen for experimental validation are also quite limited, making it it difficult to assess the strengths of the proposed method."""
566,"""Optimistic Acceleration for Optimization""","['optimization', 'Adam', 'AMSGrad']","""We consider new variants of optimization algorithms. Our algorithms are based on the observation that mini-batch of stochastic gradients in consecutive iterations do not change drastically and consequently may be predictable. Inspired by the similar setting in online learning literature called Optimistic Online learning, we propose two new optimistic algorithms for AMSGrad and Adam, respectively, by exploiting the predictability of gradients. The new algorithms combine the idea of momentum method, adaptive gradient method, and algorithms in Optimistic Online learning, which leads to speed up in training deep neural nets in practice.""","""The reviewers expressed some interest in this paper, but overall were lukewarm about its contributions. R4 raises a fundamental issue with the presentation of the analysis (see the D_infty assumption). The AC thus goes for a ""revise and resubmit""."""
567,"""Explicit Information Placement on Latent Variables using Auxiliary Generative Modelling Task""","['disentanglement', 'vae', 'clustering', 'prior imposition', 'deep generative models']","""Deep latent variable models, such as variational autoencoders, have been successfully used to disentangle factors of variation in image datasets. The structure of the representations learned by such models is usually observed after training and iteratively refined by tuning the network architecture and loss function. Here we propose a method that can explicitly place information into a specific subset of the latent variables. We demonstrate the use of the method in a task of disentangling global structure from local features in images. One subset of the latent variables is encouraged to represent local features through an auxiliary modelling task. In this auxiliary task, the global structure of an image is destroyed by dividing it into pixel patches which are then randomly shuffled. The full set of latent variables is trained to model the original data, obliging the remainder of the latent representation to model the global structure. We demonstrate that this approach successfully disentangles the latent variables for global structure from local structure by observing the generative samples of SVHN and CIFAR10. We also clustering the disentangled global structure of SVHN and found that the emerging clusters represent meaningful groups of global structures including digit identities and the number of digits presence. Finally, we discuss the problem of evaluating the clustering accuracy when ground truth categories are not expressive enough.""","""While the paper has good quality and clarity and the proposed idea seems interesting, all three reviewers agree that the paper needs more challenging experiments to justify the proposed idea. The authors are not able to include additional experiments (such as these based on different transformations) into their revision to better convince the reviewers. In addition, the AC feels that the technical novelty of the paper is rather minor (some incremental change to VAE). In particular, related to some concerns of Reviewer 3, the AC feels the proposed idea is not too much different than introducing certain kind of side-information for supervision; the main novelty seems to be distorting the data itself somehow to provide these side information (which does not seems to be that novel). """
568,"""Successor Uncertainties: exploration and uncertainty in temporal difference learning""",[],"""We consider the problem of balancing exploration and exploitation in sequential decision making problems. This trade-off naturally lends itself to probabilistic modelling. For a probabilistic approach to be effective, considering uncertainty about all immediate and long-term consequences of agent's actions is vital. An estimate of such uncertainty can be leveraged to guide exploration even in situations where the agent needs to perform a potentially long sequence of actions before reaching an under-explored area of the environment. This observation was made by the authors of the Uncertainty Bellman Equation model (O'Donoghue et al., 2018), which explicitly considers full marginal uncertainty for each decision the agent faces. However, their model still considers a fully factorised posterior over the consequences of each action, meaning that dependencies vital for correlated long-term exploration are ignored. We go a step beyond and develop Successor Uncertainties, a probabilistic model for the state-action value function of a Markov Decision Process with a non-factorised covariance. We demonstrate how this leads to greatly improved performance on classic tabular exploration benchmarks and show strong performance of our method on a subset of Atari baselines. Overall, Successor Uncertainties provides a better probabilistic model for temporal difference learning at a similar computational cost to its predecessors.""","""Pros: - interesting algorithmic idea for using successor features to propagate uncertainty for use in epxloration - clarity Cons: - moderate novelty - initially only simplistic experiments (later complemented with Atari results) - initially missing baseline comparisons - no regret-based analysis - questionable soundness because uncertainty is not guaranteed to go down All the reviewers found the initial submission to be insufficient for acceptance, and the one reviewer who read the rebuttal/revision did not change their mind, despite the addition of some large-scale results (Atari)."""
569,"""CAML: Fast Context Adaptation via Meta-Learning""",[],"""We propose CAML, a meta-learning method for fast adaptation that partitions the model parameters into two parts: context parameters that serve as additional input to the model and are adapted on individual tasks, and shared parameters that are meta-trained and shared across tasks. At test time, the context parameters are updated with one or several gradient steps on a task-specific loss that is backpropagated through the shared part of the network. Compared to approaches that adjust all parameters on a new task (e.g., MAML), our method can be scaled up to larger networks without overfitting on a single task, is easier to implement, and saves memory writes during training and network communication at test time for distributed machine learning systems. We show empirically that this approach outperforms MAML, is less sensitive to the task-specific learning rate, can capture meaningful task embeddings with the context parameters, and outperforms alternative partitionings of the parameter vectors.""","""This paper proposes a meta-learning algorithm that performs gradient-based adaptation (similar to MAML) on a lower dimensional embedding. The paper is generally well-written, and the reviewers generally agree that it has nice conceptual properties. The method also draws similarities to LEO. The main weakness of the paper is with regard to the strength of the experimental results. In a future version of the paper, we encourage the authors to improve the paper by introducing more complex domains or adding experiments that explicitly take advantage of the accessibility of the task embedding. Without such experiments that are more convincing, I do not think the paper meets the bar for acceptance at ICLR."""
570,"""The Singular Values of Convolutional Layers""","['singular values', 'operator norm', 'convolutional layers', 'regularization']","""We characterize the singular values of the linear transformation associated with a standard 2D multi-channel convolutional layer, enabling their efficient computation. This characterization also leads to an algorithm for projecting a convolutional layer onto an operator-norm ball. We show that this is an effective regularizer; for example, it improves the test error of a deep residual network using batch normalization on CIFAR-10 from 6.2% to 5.3%. ""","""This paper proposes an efficient method to compute the singular values of the linear map represented by a convolutional layer. It makes uses of the special block-matrix form of convolutional layers to construct their more efficient method. Furthermore, it shows that this method can be used to devise new regularization schemes for DNNs. The reviewers did note that the diversity of the experiments could be improved, and R2 raised concerns that the wrong singular values were being computed. The authors should add a section clarifying why the singular values of a convolutional linear map are not found directly by performing SVD on the reshaped kernel - indeed the number of singular values would be wrong. A contrast with the singular values obtained by simple reshaping of the kernel would also be helpful."""
571,"""Training Variational Auto Encoders with Discrete Latent Representations using Importance Sampling""","['Variational Auto Encoder', 'Importance Sampling', 'Discrete latent representation']","""The Variational Auto Encoder (VAE) is a popular generative latent variable model that is often applied for representation learning. Standard VAEs assume continuous valued latent variables and are trained by maximization of the evidence lower bound (ELBO). Conventional methods obtain a differentiable estimate of the ELBO with reparametrized sampling and optimize it with Stochastic Gradient Descend (SGD). However, this is not possible if we want to train VAEs with discrete valued latent variables, since reparametrized sampling is not possible. Till now, there exist no simple solutions to circumvent this problem. In this paper, we propose an easy method to train VAEs with binary or categorically valued latent representations. Therefore, we use a differentiable estimator for the ELBO which is based on importance sampling. In experiments, we verify the approach and train two different VAEs architectures with Bernoulli and Categorically distributed latent representations on two different benchmark datasets. ""","""The paper is addressing an important problem, but misses many related references (see Reviewer 2's comments for a long list of highly relevant papers). More importantly, as Reviewer 3 pointed out (which the AC fully agrees): ""The gradient estimator the paper proposes is the REINFORCE estimator [Williams, ML 1992] re-derived through importance sampling."" ""The equivalence would not be exact if the authors chose the importance distribution to be different than the variational approximation q(z|x), so there still may be room for novelty in their proposal, but in the current draft only q(z|x) is considered."" """
572,"""Pix2Scene: Learning Implicit 3D Representations from Images""","['Representation learning', 'generative model', 'adversarial learning', 'implicit 3D generation', 'scene generation']","""Modelling 3D scenes from 2D images is a long-standing problem in computer vision with implications in, e.g., simulation and robotics. We propose pix2scene, a deep generative-based approach that implicitly models the geometric properties of a scene from images. Our method learns the depth and orientation of scene points visible in images. Our model can then predict the structure of a scene from various, previously unseen view points. It relies on a bi-directional adversarial learning mechanism to generate scene representations from a latent code, inferring the 3D representation of the underlying scene geometry. We showcase a novel differentiable renderer to train the 3D model in an end-to-end fashion, using only images. We demonstrate the generative ability of our model qualitatively on both a custom dataset and on ShapeNet. Finally, we evaluate the effectiveness of the learned 3D scene representation in supporting a 3D spatial reasoning.""","""This paper proposes an approach for learning to generate 3D views, using a surfel-based representation, trained entirely from 2D images. After the discussion phase, reviewers rate the paper close to the acceptance threshold. AnonReviewer3, who initially stated ""My second concern is the results are all on synthetic data, and most shapes are very simple"", remains concerned after the rebuttal, stating ""all results are on synthetic, simple scenes. In particular, these synthetic scenes don't have lighting, material, and texture variations, making them considerably easier than any types of real images."" The AC agrees with the concerns raised by AnonReviewer3, and believes that more extensive experimentation, either on more complex synthetic scenes or on real images, is needed to back the claims of the paper. Particularly relevant is the criticism that ""While the paper is called pix2scene, its really about pix2object or pix2shape."" """
573,"""Deep reinforcement learning with relational inductive biases""","['relational reasoning', 'reinforcement learning', 'graph neural networks', 'starcraft', 'generalization', 'inductive bias']","""We introduce an approach for augmenting model-free deep reinforcement learning agents with a mechanism for relational reasoning over structured representations, which improves performance, learning efficiency, generalization, and interpretability. Our architecture encodes an image as a set of vectors, and applies an iterative message-passing procedure to discover and reason about relevant entities and relations in a scene. In six of seven StarCraft II Learning Environment mini-games, our agent achieved state-of-the-art performance, and surpassed human grandmaster-level on four. In a novel navigation and planning task, our agent's performance and learning efficiency far exceeded non-relational baselines, it was able to generalize to more complex scenes than it had experienced during training. Moreover, when we examined its learned internal representations, they reflected important structure about the problem and the agent's intentions. The main contribution of this work is to introduce techniques for representing and reasoning about states in model-free deep reinforcement learning agents via relational inductive biases. Our experiments show this approach can offer advantages in efficiency, generalization, and interpretability, and can scale up to meet some of the most challenging test environments in modern artificial intelligence.""","""The paper presents a family of models for relational reasoning over structured representations. The experiments show good results in learning efficiency and generalization, in Box-World (grid world) and StarCraft 2 mini-games, trained through reinforcement (IMPALA/off-policy A2C). The final version would benefit from more qualitative and/or quantitative details in the experimental section, as noted by all reviewers. The reviewers all agreed that this is worthy of publication at ICLR 2019. E.g. ""The paper clearly demonstrates the utility of relational inductive biases in reinforcement learning."" (R3)"""
574,"""TopicGAN: Unsupervised Text Generation from Explainable Latent Topics""","['unsupervised learning', 'topic model', 'text generation']","""Learning discrete representations of data and then generating data from the discovered representations have been increasingly studied because the obtained discrete representations can benefit unsupervised learning. However, the performance of learning discrete representations of textual data with deep generative models has not been widely explored. In addition, although generative adversarial networks(GAN) have shown impressing results in many areas such as image generation, for text generation, it is notorious for extremely difficult to train. In this work, we propose TopicGAN, a two-step text generative model, which is able to solve those two important problems simultaneously. In the first step, it discovers the latent topics and produced bag-of-words according to the latent topics. In the second step, it generates text from the produced bag-of-words. In our experiments, we show our model can discover meaningful discrete latent topics of texts in an unsupervised fashion and generate high quality natural language from the discovered latent topics.""","""This paper proposes TopicGAN, a generative adversarial approach to topic modeling and text generation. TopicGAN operates in two steps: it first generates latent topics and produces bag-of-words corresponding to those latent topics. In the second step, the model generates text conditioning on those topic words. Pros: It combines the strength of topic models (interpretable topics that are learned unsupervised) with GAN for text generation. Cons: There are three major concerns raised by reviewers: (1) clarity, (2) relatively thin experimental results, and (3) novelty. Of these, the first two were the main concerns. In particular, R1 and R2 raised concerns about insufficient component-wise evaluation (e.g., text classification from topic models) and insufficient GAN-based baselines. Also, the topic model part of TopicGAN seems somewhat underdeveloped in that the model assumes a single topic per document, which is a relatively strong simplifying assumption compared to most other topic models (R1, R3). The technical novelty is not extremely strong in that the proposed model combines existing components together. But this alone would have not been a deal breaker if the empirical results were rigorous and strong. Verdict: Reject. Many technical details require clarification and experiments lack sufficient comparisons against prior art."""
575,"""Co-manifold learning with missing data""","['nonlinear dimensionality reduction', 'missing data', 'manifold learning', 'co-clustering', 'optimization']",""" Representation learning is typically applied to only one mode of a data matrix, either its rows or columns. Yet in many applications, there is an underlying geometry to both the rows and the columns. We propose utilizing this coupled structure to perform co-manifold learning: uncovering the underlying geometry of both the rows and the columns of a given matrix, where we focus on a missing data setting. Our unsupervised approach consists of three components. We first solve a family of optimization problems to estimate a complete matrix at multiple scales of smoothness. We then use this collection of smooth matrix estimates to compute pairwise distances on the rows and columns based on a new multi-scale metric that implicitly introduces a coupling between the rows and the columns. Finally, we construct row and column representations from these multi-scale metrics. We demonstrate that our approach outperforms competing methods in both data visualization and clustering. ""","""This manuscript proposes a technique for co-manifold learning that exploits smoothness jointly over the rows and columns of the data. This is an important topic worth further study in the community. The reviewers and AC opinions were mixed, with reviewers either being unconvinced about the novelty of the proposed work or expressing issues about the clarity of the presentation. Further improvement of the clarity -- particularly clarification of the learning goals, combined with additional convincing experiments would significantly strengthen this submission."""
576,"""Interpreting Adversarial Robustness: A View from Decision Surface in Input Space""","['Adversarial examples', 'Robustness']","""One popular hypothesis of neural network generalization is that the flat local minima of loss surface in parameter space leads to good generalization. However, we demonstrate that loss surface in parameter space has no obvious relationship with generalization, especially under adversarial settings. Through visualizing decision surfaces in both parameter space and input space, we instead show that the geometry property of decision surface in input space correlates well with the adversarial robustness. We then propose an adversarial robustness indicator, which can evaluate a neural network's intrinsic robustness property without testing its accuracy under adversarial attacks. Guided by it, we further propose our robust training method. Without involving adversarial training, our method could enhance network's intrinsic adversarial robustness against various adversarial attacks.""","""This paper studies the relationship between flatness in parameter space and generalization. They show through visualization experiments on MNIST and CIFAR-10 that there is no obvious relationship between the two. However, the reviewers found the motivation for the visualization approach unconvincing and further found significant overlap between the proposed method and that of Ross & Doshi. Thus the paper should improve its framing, experimental insights and relation to prior work before being ready for publication."""
577,"""From Hard to Soft: Understanding Deep Network Nonlinearities via Vector Quantization and Statistical Inference""","['Spline', 'Vector Quantization', 'Inference', 'Nonlinearities', 'Deep Network']","""Nonlinearity is crucial to the performance of a deep (neural) network (DN). To date there has been little progress understanding the menagerie of available nonlinearities, but recently progress has been made on understanding the r\^{o}le played by piecewise affine and convex nonlinearities like the ReLU and absolute value activation functions and max-pooling. In particular, DN layers constructed from these operations can be interpreted as {\em max-affine spline operators} (MASOs) that have an elegant link to vector quantization (VQ) and pseudo-formula -means. While this is good theoretical progress, the entire MASO approach is predicated on the requirement that the nonlinearities be piecewise affine and convex, which precludes important activation functions like the sigmoid, hyperbolic tangent, and softmax. {\em This paper extends the MASO framework to these and an infinitely large class of new nonlinearities by linking deterministic MASOs with probabilistic Gaussian Mixture Models (GMMs).} We show that, under a GMM, piecewise affine, convex nonlinearities like ReLU, absolute value, and max-pooling can be interpreted as solutions to certain natural ``hard'' VQ inference problems, while sigmoid, hyperbolic tangent, and softmax can be interpreted as solutions to corresponding ``soft'' VQ inference problems. We further extend the framework by hybridizing the hard and soft VQ optimizations to create a pseudo-formula -VQ inference that interpolates between hard, soft, and linear VQ inference. A prime example of a pseudo-formula -VQ DN nonlinearity is the {\em swish} nonlinearity, which offers state-of-the-art performance in a range of computer vision tasks but was developed ad hoc by experimentation. Finally, we validate with experiments an important assertion of our theory, namely that DN performance can be significantly improved by enforcing orthogonality in its linear filters. ""","""Dear authors, All reviewers liked your work. However, they also noted that the paper was hard to read, whether because of the notation or the lack of visualization. I strongly encourage you to spend the extra effort making your work more accessible for the final version."""
578,"""Adversarial Reprogramming of Neural Networks""","['Adversarial', 'Neural Networks', 'Machine Learning Security']","""Deep neural networks are susceptible to adversarial attacks. In computer vision, well-crafted perturbations to images can cause neural networks to make mistakes such as confusing a cat with a computer. Previous adversarial attacks have been designed to degrade performance of models or cause machine learning models to produce specific outputs chosen ahead of time by the attacker. We introduce attacks that instead reprogram the target model to perform a task chosen by the attacker without the attacker needing to specify or compute the desired output for each test-time input. This attack finds a single adversarial perturbation, that can be added to all test-time inputs to a machine learning model in order to cause the model to perform a task chosen by the adversaryeven if the model was not trained to do this task. These perturbations can thus be considered a program for the new task. We demonstrate adversarial reprogramming on six ImageNet classification models, repurposing these models to perform a counting task, as well as classification tasks: classification of MNIST and CIFAR-10 examples presented as inputs to the ImageNet model.""","""Reviewers mostly recommended to accept after engaging with the authors. I have decided to reduce the weight of AnonReviewer3 because of the short review. Please take reviewers' comments into consideration to improve your submission for the camera ready. """
579,"""Stability of Stochastic Gradient Method with Momentum for Strongly Convex Loss Functions""","['Generalization Error', 'Stochastic Gradient Descent', 'Uniform Stability']","""While momentum-based methods, in conjunction with the stochastic gradient descent, are widely used when training machine learning models, there is little theoretical understanding on the generalization error of such methods. In practice, the momentum parameter is often chosen in a heuristic fashion with little theoretical guidance. In this work, we use the framework of algorithmic stability to provide an upper-bound on the generalization error for the class of strongly convex loss functions, under mild technical assumptions. Our bound decays to zero inversely with the size of the training set, and increases as the momentum parameter is increased. We also develop an upper-bound on the expected true risk, in terms of the number of training steps, the size of the training set, and the momentum parameter.""","""The paper according to Reviewers needs more work for publication and significantly more clarifications. The Reviewers are not convinced on publishing even after intensive discussion that the AC read in full. The AC recommends further improvements on the paper to address better Reviewer's concerns."""
580,"""TabNN: A Universal Neural Network Solution for Tabular Data""","['neural network', 'machine learning', 'tabular data']","""Neural Network (NN) has achieved state-of-the-art performances in many tasks within image, speech, and text domains. Such great success is mainly due to special structure design to fit the particular data patterns, such as CNN capturing spatial locality and RNN modeling sequential dependency. Essentially, these specific NNs achieve good performance by leveraging the prior knowledge over corresponding domain data. Nevertheless, there are many applications with all kinds of tabular data in other domains. Since there are no shared patterns among these diverse tabular data, it is hard to design specific structures to fit them all. Without careful architecture design based on domain knowledge, it is quite challenging for NN to reach satisfactory performance in these tabular data domains. To fill the gap of NN in tabular data learning, we propose a universal neural network solution, called TabNN, to derive effective NN architectures for tabular data in all kinds of tasks automatically. Specifically, the design of TabNN follows two principles: \emph{to explicitly leverages expressive feature combinations} and \emph{to reduce model complexity}. Since GBDT has empirically proven its strength in modeling tabular data, we use GBDT to power the implementation of TabNN. Comprehensive experimental analysis on a variety of tabular datasets demonstrate that TabNN can achieve much better performance than many baseline solutions.""","""All reviewers agree in their assessment that this paper has merits but is not yet ready for acceptance into ICLR. The area chair commends the authors for their responses to the reviews."""
581,"""Few-Shot Intent Inference via Meta-Inverse Reinforcement Learning""","['Inverse Reinforcement Learning', 'Meta-Learning', 'Deep Learning']","""A significant challenge for the practical application of reinforcement learning toreal world problems is the need to specify an oracle reward function that correctly defines a task. Inverse reinforcement learning (IRL) seeks to avoid this challenge by instead inferring a reward function from expert behavior. While appealing, it can be impractically expensive to collect datasets of demonstrations that cover the variation common in the real world (e.g. opening any type of door). Thus in practice, IRL must commonly be performed with only a limited set of demonstrations where it can be exceedingly difficult to unambiguously recover a reward function. In this work, we exploit the insight that demonstrations from other tasks can be used to constrain the set of possible reward functions by learning a ""prior"" that is specifically optimized for the ability to infer expressive reward functions from limited numbers of demonstrations. We demonstrate that our method can efficiently recover rewards from images for novel tasks and provide intuition as to how our approach is analogous to learning a prior.""","""This work proposes to use the MAML meta-learning approach in order to tackle the typical problem of insufficient demonstrations in IRL. All reviewers found this work to contain a novel and well-motivated idea and the manuscript to be well-written. The combination of MAML and MaxEnt IRL is straightforward, as R2 points out, however the AC does not consider this to be a flaw given that the main novelty here is the high-level idea rather than the technical details. However, all reviewers agree that for this paper to meet the ICLR standards, there has to be an increase in rigorousness through (a) a more close examination of assumptions, sensitivity of parameters and connections to imitation learning (b) expanding the experimental section."""
582,"""Spectral Inference Networks: Unifying Deep and Spectral Learning""","['spectral learning', 'unsupervised learning', 'manifold learning', 'dimensionality reduction']","""We present Spectral Inference Networks, a framework for learning eigenfunctions of linear operators by stochastic optimization. Spectral Inference Networks generalize Slow Feature Analysis to generic symmetric operators, and are closely related to Variational Monte Carlo methods from computational physics. As such, they can be a powerful tool for unsupervised representation learning from video or graph-structured data. We cast training Spectral Inference Networks as a bilevel optimization problem, which allows for online learning of multiple eigenfunctions. We show results of training Spectral Inference Networks on problems in quantum mechanics and feature learning for videos on synthetic datasets. Our results demonstrate that Spectral Inference Networks accurately recover eigenfunctions of linear operators and can discover interpretable representations from video in a fully unsupervised manner.""","""The paper proposes a deep learning framework to solve large-scale spectral decomposition. The reviewers and AC note that the paper is quite weak from presentation. However, technically, the proposed ideas make sense, as Reviewer 1 and Reviewer 2 mentioned. In particular, as Reviewer 1 pointed out, the paper has high practical value as it aims for solving the problem at a scale larger than any existing method. Reviewer 3 pointed out no comparison with existing algorithms, but this is understandable due to the new goal. In overall, AC thinks this is quite a boarderline paper. But, AC tends to suggest acceptance since the paper can be interested for a broad range of readers if presentation is improved."""
583,"""Deep Reinforcement Learning of Universal Policies with Diverse Environment Summaries""","['Domain Randomization', 'Diverse Summaries', 'Reinforcement learning']","""Deep reinforcement learning has enabled robots to complete complex tasks in simulation. However, the resulting policies do not transfer to real robots due to model errors in the simulator. One solution is to randomize the simulation environment, so that the resulting, trained policy achieves high performance in expectation over a variety of configurations that could represent the real-world. However, the distribution over simulator configurations must be carefully selected to represent the relevant dynamic modes of the system, as otherwise it can be unlikely to sample challenging configurations frequently enough. Moreover, the ideal distribution to improve the policy changes as the policy (un)learns to solve tasks in certain configurations. In this paper, we propose to use an inexpensive, kernel-based summarization method method that identifies configurations that lead to diverse behaviors. Since failure modes for the given task are naturally diverse, the policy trains on a mixture of representative and challenging configurations, which leads to more robust policies. In experiments, we show that the proposed method achieves the same performance as domain randomization in simple cases, but performs better when domain randomization does not lead to diverse dynamic modes.""","""The paper proposes an approach to learn policies that can effectively transfer to new environments. The perspective on this problem from the perspective of streaming submodular optimization is nice; the paper introduces new ideas that are likely of interest to the ICLR community. Unfortunately, there are significant concerns about how convincing the results are. Multiple reviewers were concerned about there only being two experiments, and the lack of comparison to ep-opt on the half-cheetah experiment. Without a more solid empirical validation of the ideas, the paper does not meet the bar for publication at ICLR."""
584,"""Generating Images from Sounds Using Multimodal Features and GANs""","['deep learning', 'machine learning', 'multimodal', 'generative adversarial networks']","""Although generative adversarial networks (GANs) have enabled us to convert images from one domain to another similar one, converting between different sensory modalities, such as images and sounds, has been difficult. This study aims to propose a network that reconstructs images from sounds. First, video data with both images and sounds are labeled with pre-trained classifiers. Second, image and sound features are extracted from the data using pre-trained classifiers. Third, multimodal layers are introduced to extract features that are common to both the images and sounds. These layers are trained to extract similar features regardless of the input modality, such as images only, sounds only, and both images and sounds. Once the multimodal layers have been trained, features are extracted from input sounds and converted into image features using a feature-to-feature GAN. Finally, the generated image features are used to reconstruct images. Experimental results show that this method can successfully convert from the sound domain into the image domain. When we applied a pre-trained classifier to both the generated and original images, 31.9% of the examples had at least one of their top 10 labels in common, suggesting reasonably good image generation. Our results suggest that common representations can be learned for different modalities, and that proposed method can be applied not only to sound-to-image conversion but also to other conversions, such as from images to sounds.""","""The work presents a new way to generate images from sounds. The reviewers found the problem ill-defined, the method not well-motivated and the results not compelling. There are a number of missing references and things to compare to, which the authors should change in a follow-up."""
585,"""Neural separation of observed and unobserved distributions""","['source separation', 'non-adversarial training', 'source unmixing', 'iterative neural training', 'generative modeling']","""Separating mixed distributions is a long standing challenge for machine learning and signal processing. Applications include: single-channel multi-speaker separation (cocktail party problem), singing voice separation and separating reflections from images. Most current methods either rely on making strong assumptions on the source distributions (e.g. sparsity, low rank, repetitiveness) or rely on having training samples of each source in the mixture. In this work, we tackle the scenario of extracting an unobserved distribution additively mixed with a signal from an observed (arbitrary) distribution. We introduce a new method: Neural Egg Separation - an iterative method that learns to separate the known distribution from progressively finer estimates of the unknown distribution. In some settings, Neural Egg Separation is initialization sensitive, we therefore introduce GLO Masking which ensures a good initialization. Extensive experiments show that our method outperforms current methods that use the same level of supervision and often achieves similar performance to full supervision. ""","""This paper presents a novel technique for separating signals in a given mixture, a common problem encountered in audio and vision tasks. The algorithm assumes that training samples from only one of the sources and the mixture distributions are available, which is a realistic assumption in a lot of cases. It then iteratively learns a model that can separate the mixture by using the available samples in a clever fashion. Strengths: - The novelty lies in how the authors formulate the problem, and the iterative approach used to learn the unknown distribution and thereby improve source separation. - The use of existing GLO masking techniques for initialization to improve performance is also novel and interesting. Weaknesses - There are some concerns around guarantees of convergence. Empirically, the algorithm works well, but it is unclear when the algorithm will fail. Some analysis here would have greatly improved the quality of the paper. - The reviewers also raised concerns around clarity of presentation and consistency of notation. While the presentation improved after revision, there are parts which remain unclear (e.g., those raised by R3) that may hinder readability and reproducibility. - The mixing model assumed by the authors is additive, which may not always be the case, e.g. when noise is convolutive (room reverberation, for instance). - (Minor) Experiments can also be improved. The vision tasks are not very realistic. For the speech separation task, relatively clean speech is easy to obtain. Therefore, it would be worth considering speech as observed, and noise as unobserved. The authors cite separating animal sounds from background, but the task chosen does not quite match that setup. Overall, the reviewers agree that the paper presents an interesting approach to separation. But given the issues with presentation and evaluations, the recommendation is to reject the paper. We strongly encourage the authors to address these concerns and resubmit in the future."""
586,"""Mapping the hyponymy relation of wordnet onto vector Spaces""","['fasttext', 'hyponymy', 'wordnet']",""" In this paper, we investigate mapping the hyponymy relation of wordnet to feature vectors. We aim to model lexical knowledge in such a way that it can be used as input in generic machine-learning models, such as phrase entailment predictors. We propose two models. The first one leverages an existing mapping of words to feature vectors (fasttext), and attempts to classify such vectors as within or outside of each class. The second model is fully supervised, using solely wordnet as a ground truth. It maps each concept to an interval or a disjunction thereof. On the first model, we approach, but not quite attain state of the art performance. The second model can achieve near-perfect accuracy. ""","""All three reviewers found this to be an interesting exploration of a reasonable topicthe use of ontologies in word representationsbut all three also expressed serious concerns about clarity and none could identify a concrete, sound result that the paper contributes to the field."""
587,"""Multi-turn Dialogue Response Generation in an Adversarial Learning Framework""","['dialogue models', 'adversarial networks', 'dialogue generation']","""We propose an adversarial learning approach to the generation of multi-turn dialogue responses. Our proposed framework, hredGAN, is based on conditional generative adversarial networks (GANs). The GAN's generator is a modified hierarchical recurrent encoder-decoder network (HRED) and the discriminator is a word-level bidirectional RNN that shares context and word embedding with the generator. During inference, noise samples conditioned on the dialogue history are used to perturb the generator's latent space to generate several possible responses. The final response is the one ranked best by the discriminator. The hredGAN shows major advantages over existing methods: (1) it generalizes better than networks trained using only the log-likelihood criterion, and (2) it generates longer, more informative and more diverse responses with high utterance and topic relevance even with limited training data. This superiority is demonstrated on the Movie triples and Ubuntu dialogue datasets with both the automatic and human evaluations.""",""" This paper proposes an adversarial learning framework for dialogue generation. The generator is based on previously proposed hierarchical recurrent encoder-decoder network (HRED) by Serban et al., and the discriminator is a bidirectional RNN. Noise is introduced in generator for response generation. The approach is evaluated on two commonly used corpora, movie data and ubuntu corpus. In the original version of the paper, human evaluation was missing, an issue raised by all reviewers, however, this has been added in the revisions. These supplement the previous automated measures in demonstrating the benefits and significant gains from the proposed approach. All reviewers raise the issue of the work being incremental and not novel enough given the previous work in HRED/VHRED and use of hierarchical approaches to model dialogue context. Furthermore, noise generation seems new, but is not well motivated, justified and analyzed. """
588,"""Incremental training of multi-generative adversarial networks""","['GAN', 'Incremental training', 'Information projection', 'Mixture distribution']","""Generative neural networks map a standard, possibly distribution to a complex high-dimensional distribution, which represents the real world data set. However, a determinate input distribution as well as a specific architecture of neural networks may impose limitations on capturing the diversity in the high dimensional target space. To resolve this difficulty, we propose a training framework that greedily produce a series of generative adversarial networks that incrementally capture the diversity of the target space. We show theoretically and empirically that our training algorithm converges to the theoretically optimal distribution, the projection of the real distribution onto the convex hull of the network's distribution space.""","""The reviewers and the AC acknowledge the paper contains interesting ideas on using an incremental sequence of multiple generators to capture the diversity of the examples. However, the reviewers and the AC also note that the potential drawback of the paper is the lack of evaluation with other metrics such as inception score, FID score, etc. Therefore the paper is not quite ready for acceptance right now, but the AC encourages the authors to submit to other top venues with more thorough experiments. """
589,"""Large Scale GAN Training for High Fidelity Natural Image Synthesis""","['GANs', 'Generative Models', 'Large Scale Training', 'Deep Learning']","""Despite recent progress in generative image modeling, successfully generating high-resolution, diverse samples from complex datasets such as ImageNet remains an elusive goal. To this end, we train Generative Adversarial Networks at the largest scale yet attempted, and study the instabilities specific to such scale. We find that applying orthogonal regularization to the generator renders it amenable to a simple ""truncation trick"", allowing fine control over the trade-off between sample fidelity and variety by reducing the variance of the Generator's input. Our modifications lead to models which set the new state of the art in class-conditional image synthesis. When trained on ImageNet at 128x128 resolution, our models (BigGANs) achieve an Inception Score (IS) of 166.3 and Frechet Inception Distance (FID) of 9.6, improving over the previous best IS of 52.52 and FID of 18.65.""","""The paper proposes a set of tricks leading to a new SOTA for sampling high resolution images. It is clearly written and the presented contribution will be of high interest for practitioners."""
590,"""A Solution to China Competitive Poker Using Deep Learning""","['artificial intelligence', 'China competitive poker', 'Dou dizhu', 'CNN', 'imperfect information game']","""Recently, deep neural networks have achieved superhuman performance in various games such as Go, chess and Shogi. Compared to Go, China Competitive Poker, also known as Dou dizhu, is a type of imperfect information game, including hidden information, randomness, multi-agent cooperation and competition. It has become widespread and is now a national game in China. We introduce an approach to play China Competitive Poker using Convolutional Neural Network (CNN) to predict actions. This network is trained by supervised learning from human game records. Without any search, the network already beats the best AI program by a large margin, and also beats the best human amateur players in duplicate mode.""","""The paper presents a CNN that is trained from human games to predict which actions to take for China Competitive Poker (Dou dizhu). The paper is poorly written, not because of the English, but because it is hard to understand the details of the proposed solution: it is not straight-forward to reimplement a solution from the presentation in the paper. It lacks explanations for several design decisions. This is unfortunate, as the authors point out in the rebuttal that they actually did way more experiments that are presented in the paper. Moreover, the experimental results lack comparisons to baselines, ablations, so that the proposed solution could be evaluated fairly. In its current state, this paper can not be accepted for presentation at ICLR 2019."""
591,"""Learning to Learn with Conditional Class Dependencies""","['meta-learning', 'learning to learn', 'few-shot learning']","""Neural networks can learn to extract statistical properties from data, but they seldom make use of structured information from the label space to help representation learning. Although some label structure can implicitly be obtained when training on huge amounts of data, in a few-shot learning context where little data is available, making explicit use of the label structure can inform the model to reshape the representation space to reflect a global sense of class dependencies. We propose a meta-learning framework, Conditional class-Aware Meta-Learning (CAML), that conditionally transforms feature representations based on a metric space that is trained to capture inter-class dependencies. This enables a conditional modulation of the feature representations of the base-learner to impose regularities informed by the label space. Experiments show that the conditional transformation in CAML leads to more disentangled representations and achieves competitive results on the miniImageNet benchmark.""","""The reviewers think that incorporating class conditional dependencies into the metric space of a few-shot learner is a sufficiently good idea to merit acceptance. The performance isnt necessarily better than the state-of-the-art approaches like LEO, but it is nonetheless competitive. One reviewer suggests incorporating a pre-training strategy to strengthen your results. In terms of experimental details, one reviewer pointed out that the embedding network architecture is quite a bit more powerful than the base learner and would like some additional justification for this. They would also like more detail on the computing the MAML gradients in the context of this method. Beyond this, please ensure that you have incorporated all of the clarifications that were required during the discussion phase."""
592,"""Optimal Completion Distillation for Sequence Learning""","['Sequence Learning', 'Edit Distance', 'Speech Recognition', 'Deep Reinforcement Learning']","""We present Optimal Completion Distillation (OCD), a training procedure for optimizing sequence to sequence models based on edit distance. OCD is efficient, has no hyper-parameters of its own, and does not require pre-training or joint optimization with conditional log-likelihood. Given a partial sequence generated by the model, we first identify the set of optimal suffixes that minimize the total edit distance, using an efficient dynamic programming algorithm. Then, for each position of the generated sequence, we use a target distribution which puts equal probability on the first token of all the optimal suffixes. OCD achieves the state-of-the-art performance on end-to-end speech recognition, on both Wall Street Journal and Librispeech datasets, achieving pseudo-formula WER and pseudo-formula WER, respectively.""","""This paper proposes an algorithm for training sequence-to-sequence models from scratch to optimize edit distance. The algorithm, called optimal completion distillation (OCD), avoids the exposure bias problem inherent in maximum likelihood estimation training, is efficient and easily implemented, and does not have any tunable hyperparameters. Experiments on Librispeech and Wall Street Journal show that OCD improves test performance over both maximum likelihood and scheduled sampling, yielding state-of-the-art results. The primary concerns expressed by the reviewers pertained to the relationship of OCD to methods such as SEARN, DAgger, AggreVaTe, LOLS, and several other papers. The revision addresses the problem with a substantially larger number of references and discussion relating OCD to the previous work. Some issues of clarity were also well addressed by the revision."""
593,"""Lagging Inference Networks and Posterior Collapse in Variational Autoencoders""","['variational autoencoders', 'posterior collapse', 'generative models']","""The variational autoencoder (VAE) is a popular combination of deep latent variable model and accompanying variational learning technique. By using a neural inference network to approximate the model's posterior on latent variables, VAEs efficiently parameterize a lower bound on marginal data likelihood that can be optimized directly via gradient methods. In practice, however, VAE training often results in a degenerate local optimum known as ""posterior collapse"" where the model learns to ignore the latent variable and the approximate posterior mimics the prior. In this paper, we investigate posterior collapse from the perspective of training dynamics. We find that during the initial stages of training the inference network fails to approximate the model's true posterior, which is a moving target. As a result, the model is encouraged to ignore the latent encoding and posterior collapse occurs. Based on this observation, we propose an extremely simple modification to VAE training to reduce inference lag: depending on the model's current mutual information between latent variable and observation, we aggressively optimize the inference network before performing each model update. Despite introducing neither new model components nor significant complexity over basic VAE, our approach is able to avoid the problem of collapse that has plagued a large amount of previous work. Empirically, our approach outperforms strong autoregressive baselines on text and image benchmarks in terms of held-out likelihood, and is competitive with more complex techniques for avoiding collapse while being substantially faster.""","""This paper introduces a method that aims to solve the problem of 'posterior collapse' in variational autoencoders (VAEs). The problem of posterior collapse is well-documented in the VAE literature, and various solutions have been proposed. Existing proposed solutions, however, aim to solve the problem by either changing the objective function (e.g. beta-VAE) or by changing the prior and/or approximate posterior models. The proposed method, in contrast, aims to solve the problem by bringing the VAE optimization procedure closer to the EM optimization procedure. Every iteration in optimization consists of SGD updates to the inference model (E-step), performed until the approximate posterior converges. This is followed by a single SGD update of the generative model. The multi-update E-step makes sure that the M-step optimizes something closer to the marginal log-likelihood, compared to what we would normaly do in VAEs (joint optimization of both inference model and generative model). The experiments are relatively small-scale, but convincing. The reviewers agree that the method is clearly described, and that the proposed technique is well supported by the experiments. We think that this work will probably be of high interest to the ICLR community."""
594,"""Prior Convictions: Black-box Adversarial Attacks with Bandits and Priors""","['adversarial examples', 'gradient estimation', 'black-box attacks', 'model-based optimization', 'bandit optimization']","""We study the problem of generating adversarial examples in a black-box setting in which only loss-oracle access to a model is available. We introduce a framework that conceptually unifies much of the existing work on black-box attacks, and demonstrate that the current state-of-the-art methods are optimal in a natural sense. Despite this optimality, we show how to improve black-box attacks by bringing a new element into the problem: gradient priors. We give a bandit optimization-based algorithm that allows us to seamlessly integrate any such priors, and we explicitly identify and incorporate two examples. The resulting methods use two to four times fewer queries and fail two to five times less than the current state-of-the-art. The code for reproducing our work is available at pseudo-url.""","""This paper is on the problem of adversarial example generation in the setting where the predictor is only accessible via function evaluations with no gradients available. The associated problem can be cast as a blackbox optimization problem wherein finite difference and related gradient estimation techniques can be used. This setting appears to be pervasive. The reviewers agree that the paper is well written and the proposed bandit optimization-based algorithm provides a nice framework in which to integrate priors, resulting in impressive empirical improvements. """
595,"""The Deep Weight Prior""","['deep learning', 'variational inference', 'prior distributions']","""Bayesian inference is known to provide a general framework for incorporating prior knowledge or specific properties into machine learning models via carefully choosing a prior distribution. In this work, we propose a new type of prior distributions for convolutional neural networks, deep weight prior (DWP), that exploit generative models to encourage a specific structure of trained convolutional filters e.g., spatial correlations of weights. We define DWP in the form of an implicit distribution and propose a method for variational inference with such type of implicit priors. In experiments, we show that DWP improves the performance of Bayesian neural networks when training data are limited, and initialization of weights with samples from DWP accelerates training of conventional convolutional neural networks. ""","""This paper proposes factorized prior distributions for CNN weights by using explicit and implicit parameterization for the prior. The paper suggest a few tractable methods to learn the prior and the model jointly. The paper, overall, is interesting. The reviewers have had some disagreement regarding the effectiveness of the method. The factorized prior may not be the most informative prior and using extra machinery to estimate it might deteriorates the performance. On the other hand, estimating a more informative prior might be difficult. It is extremely important to discuss this trade-off in the paper. I strongly recommend for the authors to discuss the pros and cons of using priors that are weakly informative vs strongly informative. The idea of using a hierarchical model has been around, e.g., see the paper on ""Hierarchical variational models"" and more recently ""semi-implicit Variational Inference"". Please include a related work on such existing work. Please discuss why your proposed method is better than these existing methods. Conditioned on the two discussions added to the paper, we can accept it. """
596,"""Kernel RNN Learning (KeRNL)""","['RNNs', 'Biologically plausible learning rules', 'Algorithm', 'Neural Networks', 'Supervised Learning']","""We describe Kernel RNN Learning (KeRNL), a reduced-rank, temporal eligibility trace-based approximation to backpropagation through time (BPTT) for training recurrent neural networks (RNNs) that gives competitive performance to BPTT on long time-dependence tasks. The approximation replaces a rank-4 gradient learning tensor, which describes how past hidden unit activations affect the current state, by a simple reduced-rank product of a sensitivity weight and a temporal eligibility trace. In this structured approximation motivated by node perturbation, the sensitivity weights and eligibility kernel time scales are themselves learned by applying perturbations. The rule represents another step toward biologically plausible or neurally inspired ML, with lower complexity in terms of relaxed architectural requirements (no symmetric return weights), a smaller memory demand (no unfolding and storage of states over time), and a shorter feedback time. ""","""this submission follows on a line of work on online learning of a recurrent net, which is an important problem both in theory and in practice. it would have been better to see even more realistic experiments, but already with the set of experiments the authors have conducted the merit of the proposed approach shines. """
597,"""Beyond Greedy Ranking: Slate Optimization via List-CVAE""","['CVAE', 'VAE', 'recommendation system', 'slate optimization', 'whole page optimization']","""The conventional approach to solving the recommendation problem greedily ranks individual document candidates by prediction scores. However, this method fails to optimize the slate as a whole, and hence, often struggles to capture biases caused by the page layout and document interdepedencies. The slate recommendation problem aims to directly find the optimally ordered subset of documents (i.e. slates) that best serve users interests. Solving this problem is hard due to the combinatorial explosion of document candidates and their display positions on the page. Therefore we propose a paradigm shift from the traditional viewpoint of solving a ranking problem to a direct slate generation framework. In this paper, we introduce List Conditional Variational Auto-Encoders (ListCVAE), which learn the joint distribution of documents on the slate conditioned on user responses, and directly generate full slates. Experiments on simulated and real-world data show that List-CVAE outperforms greedy ranking methods consistently on various scales of documents corpora.""","""The paper presents a novel perspective on optimizing lists of documents (""slates"") in a recommendation setting. The proposed approach builds on progress in variational auto-encoders, and proposes an approach that generates slates of the desired quality, conditioned on user responses. The paper presents an interesting and promising novel idea that is expected to motivate follow-up work. Conceptually, the proposed model can learn complex relationships between documents and account for these when generating slates. The paper is clearly written. The empirical results show clear improvements over competitive baselines in synthetic and semi-synthetic experiments (real users and clicks, learned user model). The reviewers and AC also note several potential shortcomings. The reviewers asked for additional baselines that reflect current state of the art approaches, and for comparisons in terms of prediction times. There are also concerns about the model's ability to generalize to (responses on) slates unseen during training, as well as concerns about the realism of the simulated user model in the evaluation. There were questions regarding the presentation, including model details / formalism. In the rebuttal phase, the authors addressed the above as follows. They added new baselines that reflect sequential document selection (auto-regressive MLP and LSTM) and demonstrate that these perform on par with greedy approaches. They provide details on an experiment to test generalization, showing both when the model succeeds and where it fails - which is valuable for understanding the advantages and limitations of the proposed approach. The authors clarified modeling and evaluation choices. Through the rebuttal and discussion phase, the reviewers reached consensus on a borderline / lean to accept decision. The AC suggests accepting the paper, based on the innovative approach and potential directions for follow up work. """
598,"""Model-Agnostic Meta-Learning for Multimodal Task Distributions""","['Meta-learning', 'gradient-based meta-learning', 'model-based meta-learning']","""Gradient-based meta-learners such as MAML (Finn et al., 2017) are able to learn a meta-prior from similar tasks to adapt to novel tasks from the same distribution with few gradient updates. One important limitation of such frameworks is that they seek a common initialization shared across the entire task distribution, substantially limiting the diversity of the task distributions that they are able to learn from. In this paper, we augment MAML with the capability to identify tasks sampled from a multimodal task distribution and adapt quickly through gradient updates. Specifically, we propose a multimodal MAML algorithm that is able to modulate its meta-learned prior according to the identified task, allowing faster adaptation. We evaluate the proposed model on a diverse set of problems including regression, few-shot image classification, and reinforcement learning. The results demonstrate the effectiveness of our model in modulating the meta-learned prior in response to the characteristics of tasks sampled from a multimodal distribution.""","""This paper proposes a meta-learning algorithm that extends MAML, particularly focusing on multimodal task distributions. The paper is generally well-written, especially with the latest revisions, and the qualitative experiments show some interesting structure recovered. The primary weakness of the paper is that the experiments are largely on relatively simple benchmarks, such as Omniglot and low-dimensional regression problems. Meta-learning papers with convincing results have shown results on MiniImagenet, CIFAR, CelebA, and/or other natural image datasets. Hence, the paper would be more compelling with more difficult experimental settings. In the paper's current form, the reviewers and the AC agree that it does not meet the bar for ICLR."""
599,"""Diversity-Sensitive Conditional Generative Adversarial Networks""","['Conditional Generative Adversarial Network', 'mode-collapse', 'multi-modal generation', 'image-to-image translation', 'image in-painting', 'video prediction']","""We propose a simple yet highly effective method that addresses the mode-collapse problem in the Conditional Generative Adversarial Network (cGAN). Although conditional distributions are multi-modal (i.e., having many modes) in practice, most cGAN approaches tend to learn an overly simplified distribution where an input is always mapped to a single output regardless of variations in latent code. To address such issue, we propose to explicitly regularize the generator to produce diverse outputs depending on latent codes. The proposed regularization is simple, general, and can be easily integrated into most conditional GAN objectives. Additionally, explicit regularization on generator allows our method to control a balance between visual quality and diversity. We demonstrate the effectiveness of our method on three conditional generation tasks: image-to-image translation, image inpainting, and future video prediction. We show that simple addition of our regularization to existing models leads to surprisingly diverse generations, substantially outperforming the previous approaches for multi-modal conditional generation specifically designed in each individual task.""","""The paper proposes a regularization term on the generator's gradient that increases sensitivity of the generator to the input noise variable in conditional and unconditional Generative Adversarial networks, and results in multimodal predictions. All reviewers agree that this is a simple and useful addition to current GANs. Experiments that demonstrate the trade off between diversity and generation quality would be important to include, as well as the experiment on using the proposed method on unconditional GANs, which was conducted during the discussion period. """
600,"""Coupled Recurrent Models for Polyphonic Music Composition""","['music composition', 'music generation', 'polyphonic music modeling']","""This work describes a novel recurrent model for music composition, which accounts for the rich statistical structure of polyphonic music. There are many ways to factor the probability distribution over musical scores; we consider the merits of various approaches and propose a new factorization that decomposes a score into a collection of concurrent, coupled time series: ""parts."" The model we propose borrows ideas from both convolutional neural models and recurrent neural models; we argue that these ideas are natural for capturing music's pitch invariances, temporal structure, and polyphony. We train generative models for homophonic and polyphonic composition on the KernScores dataset (Sapp, 2005), a collection of 2,300 musical scores comprised of around 2.8 million notes spanning time from the Renaissance to the early 20th century. While evaluation of generative models is know to be hard (Theis et al., 2016), we present careful quantitative results using a unit-adjusted cross entropy metric that is independent of how we factor the distribution over scores. We also present qualitative results using a blind discrimination test. ""","""This paper proposes novel recurrent models for polyphonic music composition and demonstrates the approach with qualitative and quantitative evaluations as well as samples. The technical parts in the original write-up were not very clear, as noted by multiple reviewers. During the review period, the presentation was improved. Unfortunately the reviewer scores are mixed, and are on the lower side, mainly because of the lack of clarity and quality of the results."""
601,"""Emerging Disentanglement in Auto-Encoder Based Unsupervised Image Content Transfer""","['Image-to-image Translation', 'Disentanglement', 'Autoencoders', 'Faces']","""We study the problem of learning to map, in an unsupervised way, between domains pseudo-formula and pseudo-formula , such that the samples \in B contain all the information that exists in samples A and some additional information. For example, ignoring occlusions, pseudo-formula can be people with glasses, pseudo-formula people without, and the glasses, would be the added information. When mapping a sample pseudo-formula from the first domain to the other domain, the missing information is replicated from an independent reference sample B Thus, in the above example, we can create, for every person without glasses a version with the glasses observed in any face image. Our solution employs a single two-pathway encoder and a single decoder for both domains. The common part of the two domains and the separate part are encoded as two vectors, and the separate part is fixed at zero for domain pseudo-formula . The loss terms are minimal and involve reconstruction losses for the two domains and a domain confusion term. Our analysis shows that under mild assumptions, this architecture, which is much simpler than the literature guided-translation methods, is enough to ensure disentanglement between the two domains. We present convincing results in a few visual domains, such as no-glasses to glasses, adding facial hair based on a reference image, etc.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. The proposed method performed well on 3 visual content transfer problems. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The paper is hard to follow at times - The problem being addressed is technically interesting but not well-motivated. That is, the question ""why is this of interest to the ICLR community"" was not well-answered. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. There were no major points of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be accepted. """
602,"""Adversarial Vulnerability of Neural Networks Increases with Input Dimension""","['adversarial vulnerability', 'neural networks', 'gradients', 'FGSM', 'adversarial data-augmentation', 'gradient regularization', 'robust optimization']","""Over the past four years, neural networks have been proven vulnerable to adversarial images: targeted but imperceptible image perturbations lead to drastically different predictions. We show that adversarial vulnerability increases with the gradients of the training objective when viewed as a function of the inputs. For most current network architectures, we prove that the L1-norm of these gradients grows as the square root of the input size. These nets therefore become increasingly vulnerable with growing image size. Our proofs rely on the networks weight distribution at initialization, but extensive experiments confirm that our conclusions still hold after usual training.""","""This paper suggests that adversarial vulnerability scales with the dimension of the input of neural networks, and support this hypothesis theoretically and experimentally. The work is well-written, and all of the reviewers appreciated the easy-to-read and clear nature of the theoretical results, including the assumptions and limitations. (The AC did not consider the criticisms raised by Reviewer 3 justified. The norm-bound perturbations considered here are a sufficiently interesting unsolved problem in the community and a clear prerequisite to solving the broader network robustness problem.) However, many of the reviewers also agreed that the theoretical assumptions - and, in particular, the random initialization of the weights - greatly oversimplify the problem. Reviewers point out that the lack of data dependence and only considering the norm of the gradient considerably limit the significance of the corresponding theoretical results, and also does not properly address the issue of gradient masking. """
603,"""A Variational Inequality Perspective on Generative Adversarial Networks""","['optimization', 'variational inequality', 'games', 'saddle point', 'extrapolation', 'averaging', 'extragradient', 'generative modeling', 'generative adversarial network']","""Generative adversarial networks (GANs) form a generative modeling approach known for producing appealing samples, but they are notably difficult to train. One common way to tackle this issue has been to propose new formulations of the GAN objective. Yet, surprisingly few studies have looked at optimization methods designed for this adversarial training. In this work, we cast GAN optimization problems in the general variational inequality framework. Tapping into the mathematical programming literature, we counter some common misconceptions about the difficulties of saddle point optimization and propose to extend methods designed for variational inequalities to the training of GANs. We apply averaging, extrapolation and a computationally cheaper variant that we call extrapolation from the past to the stochastic gradient method (SGD) and Adam.""","""The paper presents a variational inequality perspective on the optimization problem arising in GANs. Convergence of stochastic gradient descent methods (averaging and extragradient variants) is given under monotonicity (or convex) assumptions. In particular, binlinear saddle point problem is carefully studied with batch and stochastic algorithms. Experiments on CIFAR10 with WGAN etc. show that the proposed averaging and extrapolation techniques improve the GAN training in such a nonconvex optimization practices. General convergence results in the context of general non-monotone VIPs is still an open problem for future exploration. The questions raised by the reviewers are well answered. The reviewers unanimously accept the paper for ICLR publication."""
604,"""Computing committor functions for the study of rare events using deep learning with importance sampling""","['committor function', 'rare event', 'deep learning', 'importance sampling']","""The committor function is a central object of study in understanding transitions between metastable states in complex systems. However, computing the committor function for realistic systems at low temperatures is a challenging task, due to the curse of dimensionality and the scarcity of transition data. In this paper, we introduce a computational approach that overcomes these issues and achieves good performance on complex benchmark problems with rough energy landscapes. The new approach combines deep learning, importance sampling and feature engineering techniques. This establishes an alternative practical method for studying rare transition events among metastable states of complex, high dimensional systems.""","""This paper proposes a neural network based method for computing committor functions, which are used to understand transitions between stable states in complex systems. The authors improve over the techniques of Khoo et al. with a method to approximately satisfy boundary conditions and an importance sampling method to deal with rare events. This is a good application paper, introducing a new application to the ML audience, but the technical novelty is a bit limited. The reviewers see value in the paper, however scaling w.r.t. dimensionality appears to be an issue with this approach."""
605,""" Reasoning About Physical Interactions with Object-Oriented Prediction and Planning""","['structured scene representation', 'predictive models', 'intuitive physics', 'self-supervised learning']","""Object-based factorizations provide a useful level of abstraction for interacting with the world. Building explicit object representations, however, often requires supervisory signals that are difficult to obtain in practice. We present a paradigm for learning object-centric representations for physical scene understanding without direct supervision of object properties. Our model, Object-Oriented Prediction and Planning (O2P2), jointly learns a perception function to map from image observations to object representations, a pairwise physics interaction function to predict the time evolution of a collection of objects, and a rendering function to map objects back to pixels. For evaluation, we consider not only the accuracy of the physical predictions of the model, but also its utility for downstream tasks that require an actionable representation of intuitive physics. After training our model on an image prediction task, we can use its learned representations to build block towers more complicated than those observed during training.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The problem is interesting and challenging - The proposed approach is novel and performs well. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The clarity could be improved 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. Many concerns were clarified during the discussion period. One major concern had been the experimental evaluation. In particular, some reviewers felt that experiments on real images (rather than in simulation) was needed. To strengthen this aspect, the authors added new qualitative and quantitative results on a real-world experiment with a robot arm, under 10 different scenarios, showing good performance on this challenging task. Still, one reviewer was left unconvinced that the experimental evaluation was sufficient. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. Consensus was not reached. The final decision is aligned with the positive reviews as the AC believes that the evaluation was adequate. """
606,"""A quantifiable testing of global translational invariance in Convolutional and Capsule Networks""","['Translational invariance', 'CNN', 'Capsule Network']",""" We design simple and quantifiable testing of global translation-invariance in deep learning models trained on the MNIST dataset. Experiments on convolutional and capsules neural networks show that both models have poor performance in dealing with global translation-invariance; however, the performance improved by using data augmentation. Although the capsule network is better on the MNIST testing dataset, the convolutional neural network generally has better performance on the translation-invariance.""","""The paper presents an empirical comparison of translation invariance property in CNN and capsule networks. As the reviewers point out, the paper is not acceptable quality at ICLR due to low novelty and significance. """
607,"""STCN: Stochastic Temporal Convolutional Networks""","['latent variables', 'variational inference', 'temporal convolutional networks', 'sequence modeling', 'auto-regressive modeling']","""Convolutional architectures have recently been shown to be competitive on many sequence modelling tasks when compared to the de-facto standard of recurrent neural networks (RNNs) while providing computational and modelling advantages due to inherent parallelism. However, currently, there remains a performance gap to more expressive stochastic RNN variants, especially those with several layers of dependent random variables. In this work, we propose stochastic temporal convolutional networks (STCNs), a novel architecture that combines the computational advantages of temporal convolutional networks (TCN) with the representational power and robustness of stochastic latent spaces. In particular, we propose a hierarchy of stochastic latent variables that captures temporal dependencies at different time-scales. The architecture is modular and flexible due to the decoupling of the deterministic and stochastic layers. We show that the proposed architecture achieves state of the art log-likelihoods across several tasks. Finally, the model is capable of predicting high-quality synthetic samples over a long-range temporal horizon in modelling of handwritten text.""","""The paper presents a generative model of sequences based on the VAE framework, where the generative model is given by CNN with causal and dilated connections. Novelty of the method is limited; it mainly consists of bringing together the idea of causal and dilated convolutions and the VAE framework. However, knowing how well this performs is valuable the community. The proposed method appears to have significant benefits, as shown in experiments. The result on MNIST is, however, so strong that it seems incorrect; more digging into this result, or sourcecode, would have been better."""
608,"""HC-Net: Memory-based Incremental Dual-Network System for Continual learning""","['continual learning', 'lifelong learning', 'catastrophic forgetting']","""Training a neural network for a classification task typically assumes that the data to train are given from the beginning. However, in the real world, additional data accumulate gradually and the model requires additional training without accessing the old training data. This usually leads to the catastrophic forgetting problem which is inevitable for the traditional training methodology of neural networks. In this paper, we propose a memory-based continual learning method that is able to learn additional tasks while retaining the performance of previously learned tasks. Composed of two complementary networks, the Hippocampus-Net (H-Net) and the Cortex-Net (C-Net), our model estimates the index of the corresponding task for an input sample and utilizes a particular portion of itself with the estimated index. The C-Net guarantees no degradation in the performance of the previously learned tasks and the H-Net shows high confidence in finding the origin of an input sample.""","""This work is effectively an extension of progressive nets, where the task ID is not given at test time. There were several concerns about novelty of this work and the evaluation being insufficient. There was a reasonable back and forth between the reviewers and authors, and the reviewers are all aligned with the idea that this work would need a substantial rewrite in order to be accepted at ICLR."""
609,"""AdaShift: Decorrelation and Convergence of Adaptive Learning Rate Methods""","['optimizer', 'Adam', 'convergence', 'decorrelation']","""Adam is shown not being able to converge to the optimal solution in certain cases. Researchers recently propose several algorithms to avoid the issue of non-convergence of Adam, but their efficiency turns out to be unsatisfactory in practice. In this paper, we provide a new insight into the non-convergence issue of Adam as well as other adaptive learning rate methods. We argue that there exists an inappropriate correlation between gradient pseudo-formula and the second moment term pseudo-formula in Adam ( pseudo-formula is the timestep), which results in that a large gradient is likely to have small step size while a small gradient may have a large step size. We demonstrate that such unbalanced step sizes are the fundamental cause of non-convergence of Adam, and we further prove that decorrelating pseudo-formula and pseudo-formula will lead to unbiased step size for each gradient, thus solving the non-convergence problem of Adam. Finally, we propose AdaShift, a novel adaptive learning rate method that decorrelates pseudo-formula and pseudo-formula by temporal shifting, i.e., using temporally shifted gradient pseudo-formula to calculate pseudo-formula . The experiment results demonstrate that AdaShift is able to address the non-convergence issue of Adam, while still maintaining a competitive performance with Adam in terms of both training speed and generalization. ""","""This paper proposes a new stochastic optimization scheme similar to Adam. The authors claim that Adam can be improved upon by decorrelating the second-moment estimate v_t from gradient estimates g_t. This is done through the temporal decorrelation scheme, as well as block-wise sharing of estimates v_t. The reviewers agree that the paper is sufficiently well-written, original and significant to be accepted for ICLR, although some unclarity remains after the reviews. A disadvantage of the method is mainly an increased computational cost (linear in 'n', however this might be negligible when sharing v_t across blocks)."""
610,"""Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Network""",[],"""We present a new algorithm to train a robust neural network against adversarial attacks. Our algorithm is motivated by the following two ideas. First, although recent work has demonstrated that fusing randomness can improve the robustness of neural networks (Liu 2017), we noticed that adding noise blindly to all the layers is not the optimal way to incorporate randomness. Instead, we model randomness under the framework of Bayesian Neural Network (BNN) to formally learn the posterior distribution of models in a scalable way. Second, we formulate the mini-max problem in BNN to learn the best model distribution under adversarial attacks, leading to an adversarial-trained Bayesian neural net. Experiment results demonstrate that the proposed algorithm achieves state-of-the-art performance under strong attacks. On CIFAR-10 with VGG network, our model leads to 14% accuracy improvement compared with adversarial training (Madry 2017) and random self-ensemble (Liu, 2017) under PGD attack with 0.035 distortion, and the gap becomes even larger on a subset of ImageNet.""","""Reviewers are in a consensus and recommended to accept after engaging with the authors. Please take reviewers' comments into consideration to improve your submission for the camera ready. """
611,"""Area Attention""","['Deep Learning', 'attentional mechanisms', 'neural machine translation', 'image captioning']","""Existing attention mechanisms, are mostly item-based in that a model is trained to attend to individual items in a collection (the memory) where each item has a predefined, fixed granularity, e.g., a character or a word. Intuitively, an area in the memory consisting of multiple items can be worth attending to as a whole. We propose area attention: a way to attend to an area of the memory, where each area contains a group of items that are either spatially adjacent when the memory has a 2-dimensional structure, such as images, or temporally adjacent for 1-dimensional memory, such as natural language sentences. Importantly, the size of an area, i.e., the number of items in an area or the level of aggregation, is dynamically determined via learning, which can vary depending on the learned coherence of the adjacent items. By giving the model the option to attend to an area of items, instead of only individual items, a model can attend to information with varying granularity. Area attention can work along multi-head attention for attending to multiple areas in the memory. We evaluate area attention on two tasks: neural machine translation (both character and token-level) and image captioning, and improve upon strong (state-of-the-art) baselines in all the cases. These improvements are obtainable with a basic form of area attention that is parameter free. In addition to proposing the novel concept of area attention, we contribute an efficient way for computing it by leveraging the technique of summed area tables.""","""although the idea is a straightforward extension of the usual (flat) attention mechanism (which is positive), it does show some improvement in a series of experiments done in this submission. the reviewers however found the experimental results to be rather weak and believe that there may be other problems in which the proposed attention mechanism could be better utilized, despite the authors' effort at improving the result further during the rebuttal period. this may be due to a less-than-desirable form the initial submission was in, and when the new version with perhaps a new set of more convincing experiments is reviewed elsewhere, it may be received with a more positive attitude from the reviewers."""
612,"""Outlier Detection from Image Data""","['Image outlier', 'CNN', 'Deep Neural Forest']","""Modern applications from Autonomous Vehicles to Video Surveillance generate massive amounts of image data. In this work we propose a novel image outlier detection approach (IOD for short) that leverages the cutting-edge image classifier to discover outliers without using any labeled outlier. We observe that although intuitively the confidence that a convolutional neural network (CNN) has that an image belongs to a particular class could serve as outlierness measure to each image, directly applying this confidence to detect outlier does not work well. This is because CNN often has high confidence on an outlier image that does not belong to any target class due to its generalization ability that ensures the high accuracy in classification. To solve this issue, we propose a Deep Neural Forest-based approach that harmonizes the contradictory requirements of accurately classifying images and correctly detecting the outlier images. Our experiments using several benchmark image datasets including MNIST, CIFAR-10, CIFAR-100, and SVHN demonstrate the effectiveness of our IOD approach for outlier detection, capturing more than 90% of outliers generated by injecting one image dataset into another, while still preserving the classification accuracy of the multi-class classification problem.""","""The paper proposes a decision forest based method for outlier detection. The reviewers and AC note the improvement over the existing method is incremental. Although the problem is of significant practical importance, AC decided that the authors should do more works to attract the attention of a broader range of ICLR audience."""
613,"""Learning-Based Frequency Estimation Algorithms""","['streaming algorithms', 'heavy-hitters', 'Count-Min', 'Count-Sketch']","""Estimating the frequencies of elements in a data stream is a fundamental task in data analysis and machine learning. The problem is typically addressed using streaming algorithms which can process very large data using limited storage. Today's streaming algorithms, however, cannot exploit patterns in their input to improve performance. We propose a new class of algorithms that automatically learn relevant patterns in the input data and use them to improve its frequency estimates. The proposed algorithms combine the benefits of machine learning with the formal guarantees available through algorithm theory. We prove that our learning-based algorithms have lower estimation errors than their non-learning counterparts. We also evaluate our algorithms on two real-world datasets and demonstrate empirically their performance gains.""","""The paper conveys interesting ideas but reviewers are concern about an incremental nature of results, choice of comparators, and in general empirical and analytical novelty."""
614,"""Adaptive Convolutional Neural Networks""","['Adaptive kernels', 'Dynamic kernels', 'Pattern recognition', 'low memory CNNs']","""The quest for increased visual recognition performance has led to the development of highly complex neural networks with very deep topologies. To avoid high computing resource requirements of such complex networks and to enable operation on devices with limited resources, this paper introduces adaptive kernels for convolutional layers. Motivated by the non-linear perception response in human visual cells, the input image is used to define the weights of a dynamic kernel called Adaptive kernel. This new adaptive kernel is used to perform a second convolution of the input image generating the output pixel. Adaptive kernels enable accurate recognition with lower memory requirements; This is accomplished through reducing the number of kernels and the number of layers needed in the typical CNN configuration, in addition to reducing the memory used, increasing 2X the training speed and the number of activation function evaluations. Our experiments show a reduction of 70X in the memory used for MNIST, maintaining 99% accuracy and 16X memory reduction for CIFAR10 with 92.5% accuracy.""","""The paper presents a modification of the convolution layer, where the convolution weights are generated by another convolution operation. While this is an interesting idea, all reviewers felt that the evaluation and results are not particularly convincing, and the paper is not ready for acceptance."""
615,"""Featurized Bidirectional GAN: Adversarial Defense via Adversarially Learned Semantic Inference""",[],"""Deep neural networks have been demonstrated to be vulnerable to adversarial attacks, where small perturbations intentionally added to the original inputs can fool the classifier. In this paper, we propose a defense method, Featurized Bidirectional Generative Adversarial Networks (FBGAN), to extract the semantic features of the input and filter the non-semantic perturbation. FBGAN is pre-trained on the clean dataset in an unsupervised manner, adversarially learning a bidirectional mapping between a high-dimensional data space and a low-dimensional semantic space; also mutual information is applied to disentangle the semantically meaningful features. After the bidirectional mapping, the adversarial data can be reconstructed to denoised data, which could be fed into any pre-trained classifier. We empirically show the quality of reconstruction images and the effectiveness of defense.""","""The reviewers agree the paper is not ready for publication. """
616,"""Clinical Risk: wavelet reconstruction networks for marked point processes""","['point processes', 'wavelets', 'temporal neural networks', 'Hawkes processes']","""Timestamped sequences of events, pervasive in domains with data logs, e.g., health records, are often modeled as point processes with rate functions over time. Leading classical methods for risk scores such as Cox and Hawkes processes use such data but make strong assumptions about the shape and form of multivariate influences, resulting in time-to-event distributions irreflective of many real world processes. Recent methods in point processes and recurrent neural networks capably model rate functions but may be complex and difficult to interrogate. Our work develops a high-performing, interrogable model. We introduce wavelet reconstruction networks, a multivariate point process with a sparse wavelet reconstruction kernel to model rate functions from marked, timestamped data. We show they achieve improved performance and interrogability over baselines in forecasting complications and scheduled care visits in patients with diabetes.""","""There was discussion of this paper, and the accept reviewer was not willing to argue for acceptance of this paper, while the reject reviewers, specifically pointing to the clarity of the work, argued for rejection. There appear to be many good ideas related to wavelets, and hopefully the authors can work on polishing the paper and resubmitting."""
617,"""Multiple-Attribute Text Rewriting""","['controllable text generation', 'generative models', 'conditional generative models', 'style transfer']","""The dominant approach to unsupervised ""style transfer'' in text is based on the idea of learning a latent representation, which is independent of the attributes specifying its ""style''. In this paper, we show that this condition is not necessary and is not always met in practice, even with domain adversarial training that explicitly aims at learning such disentangled representations. We thus propose a new model that controls several factors of variation in textual data where this condition on disentanglement is replaced with a simpler mechanism based on back-translation. Our method allows control over multiple attributes, like gender, sentiment, product type, etc., and a more fine-grained control on the trade-off between content preservation and change of style with a pooling operator in the latent space. Our experiments demonstrate that the fully entangled model produces better generations, even when tested on new and more challenging benchmarks comprising reviews with multiple sentences and multiple attributes.""","""The paper shows how techniques introduced in the context of unsupervised machine translation can be used to build a style transfer methods. Pros: - The approach is simple and questions assumptions made by previous style transfer methods (specifically, they show that we do not need to specifically enforce disentanglement). - The evaluation is thorough and shows benefits of the proposed method - Multi-attribute style transfer is introduced and benchmarks are created - Given the success of unsupervised NMT, it makes a lot of sense to see if it can be applied to the style transfer problem Cons: - Technical novelty is limited - Some findings may be somewhat trivial (e.g., we already know that offline classifiers are stronger than the adversarials, e.g., see Elazar and Goldberg, EMNLP 2018). """
618,"""Multi-task Learning with Gradient Communication""","['Pretend to share', 'Gradient Communication']",""" In this paper, we describe a general framework to systematically analyze current neural models for multi-task learning, in which we find that existing models expect to disentangle features into different spaces while features learned in practice are still entangled in shared space, leaving potential hazards for other training or unseen tasks. We propose to alleviate this problem by incorporating a new inductive bias into the process of multi-task learning, that different tasks can communicate with each other not only by passing hidden variables but gradients explicitly. Experimentally, we evaluate proposed methods on three groups of tasks and two types of settings (\textsc{in-task} and \textsc{out-of-task}). Quantitative and qualitative results show their effectiveness.""","""This paper presents a novel idea of transferring gradients between tasks to improve multi-task learning in neural network models. The write-up includes experiments with multi-task experiments with text classification and sequence labeling, as well as multi-domain experiments. After the reviews, there are still some open questions in the reviewer comments, hence the reviewer decisions were not updated. For example, the impact of sequential update in pairwise task communication on performance can be analyzed. Two reviewers question task relatedness and the impact of how and when it is computed could be good to include in the work. Baselines could be improved to reflect reviewer suggestions."""
619,"""Plan Online, Learn Offline: Efficient Learning and Exploration via Model-Based Control""","['deep reinforcement learning', 'exploration', 'model-based']","""We propose a ""plan online and learn offline"" framework for the setting where an agent, with an internal model, needs to continually act and learn in the world. Our work builds on the synergistic relationship between local model-based control, global value function learning, and exploration. We study how local trajectory optimization can cope with approximation errors in the value function, and can stabilize and accelerate value function learning. Conversely, we also study how approximate value functions can help reduce the planning horizon and allow for better policies beyond local solutions. Finally, we also demonstrate how trajectory optimization can be used to perform temporally coordinated exploration in conjunction with estimating uncertainty in value function approximation. This exploration is critical for fast and stable learning of the value function. Combining these components enable solutions to complex control tasks, like humanoid locomotion and dexterous in-hand manipulation, in the equivalent of a few minutes of experience in the real world.""","""The paper makes novel explorations into how MPC and approximate-DP / value-function approaches, with value-fn ensembles to model value-fn uncertainty, can be effectively combined. The novelty lies in exploring their combination. The experiments are solid. The paper is clearly written. Open issues include overall novelty, and delineating the setting in which this method is appropriate. The reviewers and AC are in agreement on what is in the paper. The open question is whether the combination of the ideas is interesting. After further reviewing the paper and results. the AC believes that the overall combination of ideas and related evaluations that make a useful and promising contribution. As evidenced in some of the reviewer discussion, there is often a considerable schism in the community regarding what is considered fair to introduce in terms of prior knowledge, and blurred definitions regarding planning and control. The AC discounted some of the concerns of R2 that related more to discrete action settings and theoretical considerations; these often fail to translate to difficult problems in continuous action settings. The AC believes that R3 nicely articulates the issues of the paper that can be (and should be) addressed in the writing, i.e., to describe and motivate the settings that the proposed framework targets, as articulated in the reviews and ensuing discussion. """
620,"""Learning to Navigate the Web""","['navigating web pages', 'reinforcement learning', 'q learning', 'curriculum learning', 'meta training']","""Learning in environments with large state and action spaces, and sparse rewards, can hinder a Reinforcement Learning (RL) agents learning through trial-and-error. For instance, following natural language instructions on the Web (such as booking a flight ticket) leads to RL settings where input vocabulary and number of actionable elements on a page can grow very large. Even though recent approaches improve the success rate on relatively simple environments with the help of human demonstrations to guide the exploration, they still fail in environments where the set of possible instructions can reach millions. We approach the aforementioned problems from a different perspective and propose guided RL approaches that can generate unbounded amount of experience for an agent to learn from. Instead of learning from a complicated instruction with a large vocabulary, we decompose it into multiple sub-instructions and schedule a curriculum in which an agent is tasked with a gradually increasing subset of these relatively easier sub-instructions. In addition, when the expert demonstrations are not available, we propose a novel meta-learning framework that generates new instruction following tasks and trains the agent more effectively. We train DQN, deep reinforcement learning agent, with Q-value function approximated with a novel QWeb neural network architecture on these smaller, synthetic instructions. We evaluate the ability of our agent to generalize to new instructions onWorld of Bits benchmark, on forms with up to 100 elements, supporting 14 million possible instructions. The QWeb agent outperforms the baseline without using any human demonstration achieving 100% success rate on several difficult environments.""","""All reviewers (including those with substantial expertise in RL) were solid in their praise for this paper that is also tackling an interesting application that is much less well studied but deserves attention. """
621,"""Neural Program Repair by Jointly Learning to Localize and Repair""","['neural program repair', 'neural program embeddings', 'pointer networks']","""Due to its potential to improve programmer productivity and software quality, automated program repair has been an active topic of research. Newer techniques harness neural networks to learn directly from examples of buggy programs and their fixes. In this work, we consider a recently identified class of bugs called variable-misuse bugs. The state-of-the-art solution for variable misuse enumerates potential fixes for all possible bug locations in a program, before selecting the best prediction. We show that it is beneficial to train a model that jointly and directly localizes and repairs variable-misuse bugs. We present multi-headed pointer networks for this purpose, with one head each for localization and repair. The experimental results show that the joint model significantly outperforms an enumerative solution that uses a pointer based model for repair alone.""","""This paper provides an approach to jointly localize and repair VarMisuse bugs, where a wrong variable from the context has been used. The proposed work provides an end-to-end training pipeline for jointly localizing and repairing, as opposed to independent predictions in existing work. The reviewers felt that the manuscript was very well-written and clear, with fairly strong results on a number of datasets. The reviewers and AC note the following potential weaknesses: (1) reviewer 4 brings up related approaches from automated program repair (APR), that are much more general than the VarMisuse bugs, and the paper lacks citation and comparison to them, (2) the baselines that were compared against are fairly weak, and some recent approaches like DeepBugs and Sk_p are ignored, (3) the approach is trained and evaluated only on synthetic bugs, which look very different from the realistic ones, and (4) the contributions were found to be restricted in novelty, just uses a pointer-based LSTM for locating and fixing bugs. The authors provided detailed comments and a revision to address and clarify these concerns. They added an evaluation on realistic bugs, along with differences from DeepBugs and Sk_p, and differences between neural and automated program repair. They also added more detail comparisons, including separating the localization vs repair aspects by comparing against enumeration. During the discussion, the reviewers disagree on the ""weakness"" of the baseline, as reviewers 1 and 4 feel it is a reasonable baseline as it builds upon the Allamanis paper. They found, to different degrees, that the results on realistic bugs are much more convincing than the synthetic bug evaluation. Finally, all reviewers agree that the novelty of this work is limited. Although the reviewers disagree on the strength of the baselines (a recent paper) and the evaluation benchmarks, they agreed that the results are quite strong. The paper, however, addressed many of the concerns in the response/revision, and thus, the reviewers agree that it meets the bar for acceptance."""
622,"""Dual Learning: Theoretical Study and Algorithmic Extensions""","['machine translation', 'dual learning']","""Dual learning has been successfully applied in many machine learning applications, including machine translation, image-to-image transformation, etc. The high-level idea of dual learning is very intuitive: if we map an x from one domain to another and then map it back, we should recover the original x. Although its effectiveness has been empirically verified, theoretical understanding of dual learning is still missing. In this paper, we conduct a theoretical study to understand why and when dual learning can improve a mapping function. Based on the theoretical discoveries, we extend dual learning by introducing more related mappings and propose highly symmetric frameworks, cycle dual learning and multipath dual learning, in both of which we can leverage the feedback signals from additional domains to improve the qualities of the mappings. We prove that both cycle dual learning and multipath dual learning can boost the performance of standard dual learning under mild conditions. Experiments on WMT 14 EnglishGerman and MultiUN EnglishFrench translations verify our theoretical findings on dual learning, and the results on the translations among English, French, and Spanish of MultiUN demonstrate the efficacy of cycle dual learning and multipath dual learning.""","""The reviewers vary in their scores but overall there is agreement that this paper is not ready for acceptance. """
623,"""Biologically-Plausible Learning Algorithms Can Scale to Large Datasets""","['biologically plausible learning algorithm', 'ImageNet', 'sign-symmetry', 'feedback alignment']","""The backpropagation (BP) algorithm is often thought to be biologically implausible in the brain. One of the main reasons is that BP requires symmetric weight matrices in the feedforward and feedback pathways. To address this weight transport problem (Grossberg, 1987), two biologically-plausible algorithms, proposed by Liao et al. (2016) and Lillicrap et al. (2016), relax BPs weight symmetry requirements and demonstrate comparable learning capabilities to that of BP on small datasets. However, a recent study by Bartunov et al. (2018) finds that although feedback alignment (FA) and some variants of target-propagation (TP) perform well on MNIST and CIFAR, they perform significantly worse than BP on ImageNet. Here, we additionally evaluate the sign-symmetry (SS) algorithm (Liao et al., 2016), which differs from both BP and FA in that the feedback and feedforward weights do not share magnitudes but share signs. We examined the performance of sign-symmetry and feedback alignment on ImageNet and MS COCO datasets using different network architectures (ResNet-18 and AlexNet for ImageNet; RetinaNet for MS COCO). Surprisingly, networks trained with sign-symmetry can attain classification performance approaching that of BP-trained networks. These results complement the study by Bartunov et al. (2018) and establish a new benchmark for future biologically-plausible learning algorithms on more difficult datasets and more complex architectures.""","""This heavily disputed paper discusses a biologically motivated alternative to back-propagation learning. In particular, methods focussing on sign-symmetry rather than weight-symmetry are investigated and, importantly, scaled to large problems. The paper demonstrates the viability of the approach. If nothing else, it instigates a wonderful platform for debate. The results are convincing and the paper is well-presented. But the biological plausibility of the methods needed for these algorithms can be disputed. In my opinion, these are best tackled in a poster session, following the good practice at neuroscience meetings. On an aside note, the use of the approach to ResNet should be questioned. The skip-connections in ResNet may be all but biologically relevant."""
624,"""Real-time Neural-based Input Method""","['input method', 'language model', 'neural network', 'softmax']","""The input method is an essential service on every mobile and desktop devices that provides text suggestions. It converts sequential keyboard inputs to the characters in its target language, which is indispensable for Japanese and Chinese users. Due to critical resource constraints and limited network bandwidth of the target devices, applying neural models to input method is not well explored. In this work, we apply a LSTM-based language model to input method and evaluate its performance for both prediction and conversion tasks with Japanese BCCWJ corpus. We articulate the bottleneck to be the slow softmax computation during conversion. To solve the issue, we propose incremental softmax approximation approach, which computes softmax with a selected subset vocabulary and fix the stale probabilities when the vocabulary is updated in future steps. We refer to this method as incremental selective softmax. The results show a two order speedup for the softmax computation when converting Japanese input sequences with a large vocabulary, reaching real-time speed on commodity CPU. We also exploit the model compressing potential to achieve a 92% model size reduction without losing accuracy.""","""All reviewers agree in their assessment that this paper is not ready for acceptance into ICLR."""
625,"""Structured Prediction using cGANs with Fusion Discriminator""","['Generative Adversarial Networks', 'GANs', 'conditional GANs', 'Discriminator', 'Fusion']","""We propose a novel method for incorporating conditional information into a generative adversarial network (GAN) for structured prediction tasks. This method is based on fusing features from the generated and conditional information in feature space and allows the discriminator to better capture higher-order statistics from the data. This method also increases the strength of the signals passed through the network where the real or generated data and the conditional data agree. The proposed method is conceptually simpler than the joint convolutional neural network - conditional Markov random field (CNN-CRF) models and enforces higher-order consistency without being limited to a very specific class of high-order potentials. Experimental results demonstrate that this method leads to improvement on a variety of different structured prediction tasks including image synthesis, semantic segmentation, and depth estimation.""","""All three reviewers argue for rejection on the basis that this paper does not make a sufficiently novel and substantial contribution to warrant publication. The AC follows their recommendation."""
626,"""HIGHLY EFFICIENT 8-BIT LOW PRECISION INFERENCE OF CONVOLUTIONAL NEURAL NETWORKS""","['8-bit low precision inference', 'convolutional neural networks', 'statistical accuracy', '8-bit Winograd convolution']","""High throughput and low latency inference of deep neural networks are critical for the deployment of deep learning applications. This paper presents a general technique toward 8-bit low precision inference of convolutional neural networks, including 1) channel-wise scale factors of weights, especially for depthwise convolution, 2) Winograd convolution, and 3) topology-wise 8-bit support. We experiment the techniques on top of a widely-used deep learning framework. The 8-bit optimized model is automatically generated with a calibration process from FP32 model without the need of fine-tuning or retraining. We perform a systematical and comprehensive study on 18 widely-used convolutional neural networks and demonstrate the effectiveness of 8-bit low precision inference across a wide range of applications and use cases, including image classification, object detection, image segmentation, and super resolution. We show that the inference throughput and latency are improved by 1.6X and 1.5X respectively with minimal within 0.6%1to no loss in accuracy from FP32 baseline. We believe the methodology can provide the guidance and reference design of 8-bit low precision inference for other frameworks. All the code and models will be publicly available soon.""","""The paper proposes to combine three methods of quantization and apply them to neural network compression. The methods are known in the literature. There is a lack of theoretical contribution, and experimental results show variable speedups that may not be competitive with the current state-of-the-art in neural network compression. The majority of reviewers recommend that this paper be rejected. The authors have not provided a response."""
627,"""Theoretical and Empirical Study of Adversarial Examples""","['Adversarial examples', 'Feature smoothing', 'Data augmentation', 'Decision boundary']","""Many techniques are developed to defend against adversarial examples at scale. So far, the most successful defenses generate adversarial examples during each training step and add them to the training data. Yet, this brings significant computational overhead. In this paper, we investigate defenses against adversarial attacks. First, we propose feature smoothing, a simple data augmentation method with little computational overhead. Essentially, feature smoothing trains a neural network on virtual training data as an interpolation of features from a pair of samples, with the new label remaining the same as the dominant data point. The intuition behind feature smoothing is to generate virtual data points as close as adversarial examples, and to avoid the computational burden of generating data during training. Our experiments on MNIST and CIFAR10 datasets explore different combinations of known regularization and data augmentation methods and show that feature smoothing with logit squeezing performs best for both adversarial and clean accuracy. Second, we propose an unified framework to understand the connections and differences among different efficient methods by analyzing the biases and variances of decision boundary. We show that under some symmetrical assumptions, label smoothing, logit squeezing, weight decay, mix up and feature smoothing all produce an unbiased estimation of the decision boundary with smaller estimated variance. All of those methods except weight decay are also stable when the assumptions no longer hold.""","""The paper proposes a feature smoothing technique as a new and ""cheaper"" technique for training adversarially robust models. Pros: * the paper is generally well written and the claimed results seem quite promising * the theory contribution are interesting Cons: * the main technique is fairly incremental * there were concerns regarding the comprehensiveness of evaluations and baselines used"""
628,"""A More Globally Accurate Dimensionality Reduction Method Using Triplets""","['Dimensionality Reduction', 'Visualization', 'Triplets', 't-SNE', 'LargeVis']","""We first show that the commonly used dimensionality reduction (DR) methods such as t-SNE and LargeVis poorly capture the global structure of the data in the low dimensional embedding. We show this via a number of tests for the DR methods that can be easily applied by any practitioner to the dataset at hand. Surprisingly enough, t-SNE performs the best w.r.t. the commonly used measures that reward the local neighborhood accuracy such as precision-recall while having the worst performance in our tests for global structure. We then contrast the performance of these two DR method against our new method called TriMap. The main idea behind TriMap is to capture higher orders of structure with triplet information (instead of pairwise information used by t-SNE and LargeVis), and to minimize a robust loss function for satisfying the chosen triplets. We provide compelling experimental evidence on large natural datasets for the clear advantage of the TriMap DR results. As LargeVis, TriMap is fast and and provides comparable runtime on large datasets.""","""Dear authors, The reviewers all appreciated your goal of improving dimensionality reduction techniques. This is a field which does not enjoy the popularity it once did but remains nonetheless important. They also appreciated the novel loss and the use of triplets.to get the global structure. However, the paper lacks some guidance. In particular, it oscillates between showing qualitative results (robustness to outliers, ""nice"" visualizations) and quantitative ones (running time, classification performance). I agree with the reviewers that the quantitative ones should have used the same preprocessing for t-SNE and TriMap (either PCA or no PCA), regardless of the current implementation in software tools. Given that the quantitative results are not that impressive, may I suggest focusing on the qualitative ones for a resubmission? The robustness of the emeddings to the addition or removal of a few points is definitely interesting and worth further investigation, optionally with a corresponding metric."""
629,"""Pseudosaccades: A simple ensemble scheme for improving classification performance of deep nets""","['Ensemble classification', 'random subspace', 'data sketching']","""We describe a simple ensemble approach that, unlike conventional ensembles, uses multiple random data sketches (pseudosaccades) rather than multiple classifiers to improve classification performance. Using this simple, but novel, approach we obtain statistically significant improvements in classification performance on AlexNet, GoogLeNet, ResNet-50 and ResNet-152 baselines on Imagenet data e.g. of the order of 0.3% to 0.6% in Top-1 accuracy and similar improvements in Top-k accuracy essentially nearly for free.""","""The paper proposes a data augmentation technique to ensemble classifiers. Reviewers pointed to a few concerns, including a lack of novelty, a lack of proper comparison with state-of-the-art models or other data augmentation approaches. Overall, all reviewers recommended to reject the paper, and I concur with them."""
630,"""Improving Generalization and Stability of Generative Adversarial Networks""","['GAN', 'generalization', 'gradient penalty', 'zero centered', 'convergence']","""Generative Adversarial Networks (GANs) are one of the most popular tools for learning complex high dimensional distributions. However, generalization properties of GANs have not been well understood. In this paper, we analyze the generalization of GANs in practical settings. We show that discriminators trained on discrete datasets with the original GAN loss have poor generalization capability and do not approximate the theoretically optimal discriminator. We propose a zero-centered gradient penalty for improving the generalization of the discriminator by pushing it toward the optimal discriminator. The penalty guarantees the generalization and convergence of GANs. Experiments on synthetic and large scale datasets verify our theoretical analysis. ""","""The paper received unanimous accept over reviewers (7,7,6), hence proposed as definite accept. """
631,"""Analyzing Federated Learning through an Adversarial Lens""","['federated learning', 'model poisoning']","""Federated learning distributes model training among a multitude of agents, who, guided by privacy concerns, perform training using their local data but share only model parameter updates, for iterative aggregation at the server. In this work, we explore the threat of model poisoning attacks on federated learning initiated by a single, non-colluding malicious agent where the adversarial objective is to cause the model to misclassify a set of chosen inputs with high confidence. We explore a number of strategies to carry out this attack, starting with simple boosting of the malicious agent's update to overcome the effects of other agents' updates. To increase attack stealth, we propose an alternating minimization strategy, which alternately optimizes for the training loss and the adversarial objective. We follow up by using parameter estimation for the benign agents' updates to improve on attack success. Finally, we use a suite of interpretability techniques to generate visual explanations of model decisions for both benign and malicious models and show that the explanations are nearly visually indistinguishable. Our results indicate that even a highly constrained adversary can carry out model poisoning attacks while simultaneously maintaining stealth, thus highlighting the vulnerability of the federated learning setting and the need to develop effective defense strategies.""","""This paper proposes model poisoning (poisoned parameter updates in a federated setting) in contrast to data poisoning (poisoned training data). It proposes an attack method and compares to baselines that are also proposed in the paper (there are no external baselines). While model poisoning is indeed an interesting direction to consider, I agree with reviewer concerns that the relation to data poisoning is not clearly addressed. In particular, any data poisoning attack could be used as a model poisoning attack (just provide whatever updates would be induced by the poisoned data), so there is no good excuse to not compare to the existing strong data poisoning attacks. One reviewer raised concerns about lack of theoretical guarantees but I do not agree with these concerns (the authors correctly point out in the rebuttal that this is not necessary for an attack-focused paper). I do feel there is room to improve the overall clarity/motivation (for instance, equation (1) is presented without any explanation and it is still not clear to me why this is the right formulation)."""
632,"""Universal discriminative quantum neural networks""","['quantum machine learning', 'quantum data classification']","""Quantum mechanics fundamentally forbids deterministic discrimination of quantum states and processes. However, the ability to optimally distinguish various classes of quantum data is an important primitive in quantum information science. In this work, we trained near-term quantum circuits to classify data represented by quantum states using the Adam stochastic optimization algorithm. This is achieved by iterative interactions of a classical device with a quantum processor to discover the parameters of an unknown non-unitary quantum circuit. This circuit learns to simulate the unknown structure of a generalized quantum measurement, or positive-operator valued measure (POVM), that is required to optimally distinguish possible distributions of quantum inputs. Notably we used universal circuit topologies, with a theoretically motivated circuit design which guaranteed that our circuits can perform arbitrary input-output mappings. Our numerical simulations showed that quantum circuits could be trained to discriminate among various pure and mixed quantum states, exhibiting a trade-off between minimizing erroneous and inconclusive outcomes with comparable performance to theoretically optimal POVMs. We trained the circuit on different classes of quantum data and evaluated the generalization error on unseen quantum data. This generalization power hence distinguishes our work from standard circuit optimization and provides an example of quantum machine learning for a task that has inherently no classical analogue. ""","""The paper needs work to improve clarity and strengthen the technical message. Also, the authors broke the policy of anonymous submission which disqualifies the paper."""
633,"""Unsupervised Emergence of Spatial Structure from Sensorimotor Prediction""","['spatial perception', 'grounding', 'sensorimotor prediction', 'unsupervised learning', 'representation learning']","""Despite its omnipresence in robotics application, the nature of spatial knowledge and the mechanisms that underlie its emergence in autonomous agents are still poorly understood. Recent theoretical work suggests that the concept of space can be grounded by capturing invariants induced by the structure of space in an agent's raw sensorimotor experience. Moreover, it is hypothesized that capturing these invariants is beneficial for a naive agent trying to predict its sensorimotor experience. Under certain exploratory conditions, spatial representations should thus emerge as a byproduct of learning to predict. We propose a simple sensorimotor predictive scheme, apply it to different agents and types of exploration, and evaluate the pertinence of this hypothesis. We show that a naive agent can capture the topology and metric regularity of its spatial configuration without any a priori knowledge, nor extraneous supervision.""","""This paper is borderline for publication for the following reasons: 1) the title is misleading. The majority of the ICLR audience understands by ""spatial structure"" the structure of the external 3D world, as opposed to the position of the sensors in the internal coordinate system of the agent. Though the authors argue that knowing the positions of the sensors eventually leads to learning the 3D world structure, this appears like a leap in the argument. 2) The equation s=\phi(m) described a mapping from robot postures to sensory states. This means the agent should remain within the same scene. The description of this equation in the manuscript as ""The mapping  can be seen as describing how the world transforms changes in motor states into changes in sensory states ..."" makes this equation appear more general than what it is. s'=\psi(s,m) would be better described by such sentence. """
634,"""G-SGD: Optimizing ReLU Neural Networks in its Positively Scale-Invariant Space""","['optimization', 'neural network', 'irreducible positively scale-invariant space', 'deep learning']","""It is well known that neural networks with rectified linear units (ReLU) activation functions are positively scale-invariant. Conventional algorithms like stochastic gradient descent optimize the neural networks in the vector space of weights, which is, however, not positively scale-invariant. This mismatch may lead to problems during the optimization process. Then, a natural question is: \emph{can we construct a new vector space that is positively scale-invariant and sufficient to represent ReLU neural networks so as to better facilitate the optimization process }? In this paper, we provide our positive answer to this question. First, we conduct a formal study on the positive scaling operators which forms a transformation group, denoted as pseudo-formula . We prove that the value of a path (i.e. the product of the weights along the path) in the neural network is invariant to positive scaling and the value vector of all the paths is sufficient to represent the neural networks under mild conditions. Second, we show that one can identify some basis paths out of all the paths and prove that the linear span of their value vectors (denoted as pseudo-formula -space) is an invariant space with lower dimension under the positive scaling group. Finally, we design stochastic gradient descent algorithm in pseudo-formula -space (abbreviated as pseudo-formula -SGD) to optimize the value vector of the basis paths of neural networks with little extra cost by leveraging back-propagation. Our experiments show that pseudo-formula -SGD significantly outperforms the conventional SGD algorithm in optimizing ReLU networks on benchmark datasets. ""","""This paper proposes a new optimization method for ReLU networks that optimizes in a scale-invariant vector space in the hopes of facilitating learning. The proposed method is novel and is validated by some experiments on CIFAR-10 and CIFAR-100. The reviewers find the analysis of the invariance group informative but have raised questions about the computational cost of the method. These concerns were addressed by the authors in the revision. The method could be of practical interest to the community and so acceptance is recommended."""
635,"""ACTRCE: Augmenting Experience via Teachers Advice""","['language goals', 'task generalization', 'hindsight experience replays', 'language grounding']","""Sparse reward is one of the most challenging problems in reinforcement learning (RL). Hindsight Experience Replay (HER) attempts to address this issue by converting a failure experience to a successful one by relabeling the goals. Despite its effectiveness, HER has limited applicability because it lacks a compact and universal goal representation. We present Augmenting experienCe via TeacheR's adviCE (ACTRCE), an efficient reinforcement learning technique that extends the HER framework using natural language as the goal representation. We first analyze the differences among goal representation, and show that ACTRCE can efficiently solve difficult reinforcement learning problems in challenging 3D navigation tasks, whereas HER with non-language goal representation failed to learn. We also show that with language goal representations, the agent can generalize to unseen instructions, and even generalize to instructions with unseen lexicons. We further demonstrate it is crucial to use hindsight advice to solve challenging tasks, but we also found that little amount of hindsight advice is sufficient for the learning to take off, showing the practical aspect of the method.""","""This paper was reviewed by three experts (I assure the authors R3 is indeed familiar with RL and this area). Initially, the reviews were mixed with several concerns raised. After the author response, R2 and R3 recommend rejecting the paper, and R1 is unwilling to defend/champion/support it (not visible to the authors). The AC agrees with the concerns raised (in particular by R2) and finds no basis for overruling this recommendation. We encourage the authors to incorporate reviewer feedback and submit a stronger manuscript at a future venue. """
636,"""Phase-Aware Speech Enhancement with Deep Complex U-Net""","['speech enhancement', 'deep learning', 'complex neural networks', 'phase estimation']","""Most deep learning-based models for speech enhancement have mainly focused on estimating the magnitude of spectrogram while reusing the phase from noisy speech for reconstruction. This is due to the difficulty of estimating the phase of clean speech. To improve speech enhancement performance, we tackle the phase estimation problem in three ways. First, we propose Deep Complex U-Net, an advanced U-Net structured model incorporating well-defined complex-valued building blocks to deal with complex-valued spectrograms. Second, we propose a polar coordinate-wise complex-valued masking method to reflect the distribution of complex ideal ratio masks. Third, we define a novel loss function, weighted source-to-distortion ratio (wSDR) loss, which is designed to directly correlate with a quantitative evaluation measure. Our model was evaluated on a mixture of the Voice Bank corpus and DEMAND database, which has been widely used by many deep learning models for speech enhancement. Ablation experiments were conducted on the mixed dataset showing that all three proposed approaches are empirically valid. Experimental results show that the proposed method achieves state-of-the-art performance in all metrics, outperforming previous approaches by a large margin.""","""The authors propose an algorithm for enhancing noisy speech by also accounting for the phase information. This is done by adapting UNets to handle features defined in the complex space, and by adapting the loss function to improve an appropriate evaluation metric. Strengths - Modifies existing techniques well to better suit the domain for which the algorithm is being proposed. Modifications like extending UNet to complex Unet to deal with phase, redefining the mask and loss are all interesting improvements. - Extensive results and analysis. Weaknesses - The work is centered around speech enhancement, and hence has limited focus. Even though the paper is limited to speech enhancement, the reviewers agreed that the contributions made by the paper are significant and can help improve related applications like ASR. The paper is well written with interesting results and analysis. Therefore, it is recommended that the paper be accepted. """
637,"""Hyper-Regularization: An Adaptive Choice for the Learning Rate in Gradient Descent""","['Adaptive learning rate', 'novel framework']","""We present a novel approach for adaptively selecting the learning rate in gradient descent methods. Specifically, we impose a regularization term on the learning rate via a generalized distance, and cast the joint updating process of the parameter and the learning rate into a maxmin problem. Some existing schemes such as AdaGrad (diagonal version) and WNGrad can be rederived from our approach. Based on our approach, the updating rules for the learning rate do not rely on the smoothness constant of optimization problems and are robust to the initial learning rate. We theoretically analyze our approach in full batch and online learning settings, which achieves comparable performances with other first-order gradient-based algorithms in terms of accuracy as well as convergence rate.""","""All three reviewers found that the motivation for the proposed method was lacking and recommend rejection. The AC thus recommends the authors to take these comments in consideration when revising their manuscript."""
638,"""GEOMETRIC AUGMENTATION FOR ROBUST NEURAL NETWORK CLASSIFIERS""","['Bayesian nonparametric', 'robust', 'deep neural network', 'classifier', 'unsupervised learning', 'geometric']","""We introduce a novel geometric perspective and unsupervised model augmentation framework for transforming traditional deep (convolutional) neural networks into adversarially robust classifiers. Class-conditional probability densities based on Bayesian nonparametric mixtures of factor analyzers (BNP-MFA) over the input space are used to design soft decision labels for feature to label isometry. Classconditional distributions over features are also learned using BNP-MFA to develop plug-in maximum a posterior (MAP) classifiers to replace the traditional multinomial logistic softmax classification layers. This novel unsupervised augmented framework, which we call geometrically robust networks (GRN), is applied to CIFAR-10, CIFAR-100, and to Radio-ML (a time series dataset for radio modulation recognition). We demonstrate the robustness of GRN models to adversarial attacks from fast gradient sign method, Carlini-Wagner, and projected gradient descent.""","""All three reviewers feel that the paper needs to provide more convincing results to support their robustness claim, in addition to a number of other issues that need to be clarified/improved. The authors did not provide any response. """
639,"""Learning Finite State Representations of Recurrent Policy Networks""","['recurrent neural networks', 'finite state machine', 'quantization', 'interpretability', 'autoencoder', 'moore machine', 'reinforcement learning', 'imitation learning', 'representation', 'Atari', 'Tomita']","""Recurrent neural networks (RNNs) are an effective representation of control policies for a wide range of reinforcement and imitation learning problems. RNN policies, however, are particularly difficult to explain, understand, and analyze due to their use of continuous-valued memory vectors and observation features. In this paper, we introduce a new technique, Quantized Bottleneck Insertion, to learn finite representations of these vectors and features. The result is a quantized representation of the RNN that can be analyzed to improve our understanding of memory use and general behavior. We present results of this approach on synthetic environments and six Atari games. The resulting finite representations are surprisingly small in some cases, using as few as 3 discrete memory states and 10 observations for a perfect Pong policy. We also show that these finite policy representations lead to improved interpretability. ""","""The paper addresses the problem of interpreting recurrent neural networks by quantizing their states an mapping them onto a Moore Machine. The paper presents some interesting results on reinforcement learning and other tasks. I believe the experiments could have been more informative if the proposed technique was compared against a simple quantization baseline (e.g. based on k-means) so that one can get a better understanding of the difficulty of these task. This paper is clearly above the acceptance threshold at ICLR. """
640,"""Meta-Learning for Contextual Bandit Exploration""","['meta-learning', 'bandits', 'exploration', 'imitation learning']","""We describe MLE, a meta-learning algorithm for learning a good exploration policy in the interactive contextual bandit setting. Here, an algorithm must take actions based on contexts, and learn based only on a reward signal from the action taken, thereby generating an exploration/exploitation trade-off. MLE addresses this trade-off by learning a good exploration strategy based on offline synthetic tasks, on which it can simulate the contextual bandit setting. Based on these simulations, MLE uses an imitation learning strategy to learn a good exploration policy that can then be applied to true contextual bandit tasks at test time. We compare MLE to seven strong baseline contextual bandit algorithms on a set of three hundred real-world datasets, on which it outperforms alternatives in most settings, especially when differences in rewards are large. Finally, we demonstrate the importance of having a rich feature representation for learning how to explore. ""","""This paper provides an interesting strategy for learning to explore, by first training on fully supervised data before deploying that policy to an online setting. There are some concerns, however, on the realism and utility of this setting that should be further discussed. If the offline data is not related to the contextual bandit problem, it would be surprising for this to have much benefit, and this should be better motivated and discussed. Because there are no theoretical guarantees for exploration, a discussion is needed and as suggested by a reviewer the learned exploration policies could be qualitatively examined. For example, the paper says ""While these approaches are effective if the distribution of tasks is very similar and the state space is shared among different tasks, they fail to generalize when the tasks are different. Our approach targets an easier problem than exploration in full reinforcement learning environments, and can generalize well across a wide range of different tasks with completely unrelated features spaces."" This is a pretty surprising statement, that your idea would not work well in an RL setting, but does work well in a contextual bandit setting. There should also be a bit more discussion comparing to previous approach to learn how to explore, including in active learning. It is true that active learning is a different setting, but in both a goal is to become optimal as quickly as possible. Similarly, the ideas used for RL could be used here as well, essentially by setting gamma to 0. Overall, the ideas here are interesting and well-written, but need a bit more development on previous work, and motivation for why this approach will be effective. """
641,"""Sorting out Lipschitz function approximation""","['deep learning', 'lipschitz neural networks', 'generalization', 'universal approximation', 'adversarial examples', 'generative models', 'optimal transport', 'adversarial robustness']","""Training neural networks subject to a Lipschitz constraint is useful for generalization bounds, provable adversarial robustness, interpretable gradients, and Wasserstein distance estimation. By the composition property of Lipschitz functions, it suffices to ensure that each individual affine transformation or nonlinear activation function is 1-Lipschitz. The challenge is to do this while maintaining the expressive power. We identify a necessary property for such an architecture: each of the layers must preserve the gradient norm during backpropagation. Based on this, we propose to combine a gradient norm preserving activation function, GroupSort, with norm-constrained weight matrices. We show that norm-constrained GroupSort architectures are universal Lipschitz function approximators. Empirically, we show that norm-constrained GroupSort networks achieve tighter estimates of Wasserstein distance than their ReLU counterparts and can achieve provable adversarial robustness guarantees with little cost to accuracy.""","""This paper presents an interesting and theoretically motivated approach to imposing Lipschitz constraints on functions learned by neural networks. R2 and R3 found the idea interesting, but R1 and R2 both point out several issues with the submitted version, including some problems with the proof--probably fixable--as well as a number of writing issues. The authors submitted a cleaned-up revised version, but upon checking revisions it appears the paper was almost completely re-written after the deadline. I do not think reviewers should be expected to comment a second time on such large changes, so I am okay with R1's decision to not review the updated version. Future reviewers of a more polished version of the paper will be in a better position to assess its merits in detail."""
642,"""SNAS: stochastic neural architecture search""",['Neural Architecture Search'],"""We propose Stochastic Neural Architecture Search (SNAS), an economical end-to-end solution to Neural Architecture Search (NAS) that trains neural operation parameters and architecture distribution parameters in same round of back-propagation, while maintaining the completeness and differentiability of the NAS pipeline. In this work, NAS is reformulated as an optimization problem on parameters of a joint distribution for the search space in a cell. To leverage the gradient information in generic differentiable loss for architecture search, a novel search gradient is proposed. We prove that this search gradient optimizes the same objective as reinforcement-learning-based NAS, but assigns credits to structural decisions more efficiently. This credit assignment is further augmented with locally decomposable reward to enforce a resource-efficient constraint. In experiments on CIFAR-10, SNAS takes less epochs to find a cell architecture with state-of-the-art accuracy than non-differentiable evolution-based and reinforcement-learning-based NAS, which is also transferable to ImageNet. It is also shown that child networks of SNAS can maintain the validation accuracy in searching, with which attention-based NAS requires parameter retraining to compete, exhibiting potentials to stride towards efficient NAS on big datasets.""","""This paper provides an alternative way to enable differentiable optimization to the neural architecture search problem. Different from DARTS, SNAS reformulates the problem and employs Gumbel random variables to directly optimize the NAS objective. In addition, the resource-constrained regularization is interesting. The major cons of the paper is that the empirical results are not quite impressive, especially when compared to DARTS, in terms of both accuracy and convergence. I think this is a borderline paper but maybe good enough for acceptance. """
643,"""Unsupervised classification into unknown number of classes""",['unsupervised learning'],"""We propose a novel unsupervised classification method based on graph Laplacian. Unlike the widely used classification method, this architecture does not require the labels of data and the number of classes. Our key idea is to introduce a approximate linear map and a spectral clustering theory on the dimension reduced spaces into generative adversarial networks. Inspired by the human visual recognition system, the proposed framework can classify and also generate images as the human brains do. We build an approximate linear connector network pseudo-formula analogous to the cerebral cortex, between the discriminator pseudo-formula and the generator pseudo-formula . The connector network allows us to estimate the unknown number of classes. Estimating the number of classes is one of the challenging researches in the unsupervised learning, especially in spectral clustering. The proposed method can also classify the images by using the estimated number of classes. Therefore, we define our method as an unsupervised classification method.""","""Following the unanimous vote of the submitted reviews, this paper is not ready for publication at ICLR. Among other concerns raised, the experiments need significant work, and the exposition needs clarification."""
644,"""Attentive Neural Processes""","['Neural Processes', 'Conditional Neural Processes', 'Stochastic Processes', 'Regression', 'Attention']","""Neural Processes (NPs) (Garnelo et al., 2018) approach regression by learning to map a context set of observed input-output pairs to a distribution over regression functions. Each function models the distribution of the output given an input, conditioned on the context. NPs have the benefit of fitting observed data efficiently with linear complexity in the number of context input-output pairs, and can learn a wide family of conditional distributions; they learn predictive distributions conditioned on context sets of arbitrary size. Nonetheless, we show that NPs suffer a fundamental drawback of underfitting, giving inaccurate predictions at the inputs of the observed data they condition on. We address this issue by incorporating attention into NPs, allowing each input location to attend to the relevant context points for the prediction. We show that this greatly improves the accuracy of predictions, results in noticeably faster training, and expands the range of functions that can be modelled. ""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The paper is clear and well-motivated. - The experimental results indicate that the proposed method outperforms the SOTA 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The novelty is somewhat minor. - An interesting (but not essential) ablation study is missing (but the authors promised to include it in the final version). 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. There were no major points of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be accepted. """
645,"""Multi-Grained Entity Proposal Network for Named Entity Recognition""",[],"""In this paper, we focus on a new Named Entity Recognition (NER) task, i.e., the Multi-grained NER task. This task aims to simultaneously detect both fine-grained and coarse-grained entities in sentences. Correspondingly, we develop a novel Multi-grained Entity Proposal Network (MGEPN). Different from traditional NER models which regard NER as a sequential labeling task, MGEPN provides a new method that proposes entity candidates in the Proposal Network and classifies entities into different categories in the Classification Network. All possible entity candidates including fine-grained ones and coarse-grained ones are proposed in the Proposal Network, which enables the MGEPN model to identify multi-grained entities. In order to better identify named entities and determine their categories, context information is utilized and transferred from the Proposal Network to the Classification Network during the learning process. A novel Entity-Context attention mechanism is also introduced to help the model focus on entity-related context information. Experiments show that our model can obtain state-of-the-art performance on two real-world datasets for both the Multi-grained NER task and the traditional NER task.""","""The authors present a method for fine grained entity tagging, which could be useful in certain practical scenarios. I found the labeling of the CoNLL data with the fine grained entities a bit confusing. The authors did not talk about the details of how the coarse grained labels were changed to fine grained ones. This detail is important and is missing from the paper. Moreover, there are concerns about the novelty of the work, both in terms of the task definition and the model (see the review of Reviewer 1, e.g.). There is consensus amongst the reviewers, in that, their feedback is lukewarm about the paper. """
646,"""MLPrune: Multi-Layer Pruning for Automated Neural Network Compression""","['Automated Model Compression', 'Neural Network Pruning']","""Model compression can significantly reduce the computation and memory footprint of large neural networks. To achieve a good trade-off between model size and accuracy, popular compression techniques usually rely on hand-crafted heuristics and require manually setting the compression ratio of each layer. This process is typically costly and suboptimal. In this paper, we propose a Multi-Layer Pruning method (MLPrune), which is theoretically sound, and can automatically decide appropriate compression ratios for all layers. Towards this goal, we use an efficient approximation of the Hessian as our pruning criterion, based on a Kronecker-factored Approximate Curvature method. We demonstrate the effectiveness of our method on several datasets and architectures, outperforming previous state-of-the-art by a large margin. Our experiments show that we can compress AlexNet and VGG16 by 25x without loss in accuracy on ImageNet. Furthermore, our method has much fewer hyper-parameters and requires no expert knowledge.""","""The authors propose a technique for pruning networks by using second-order information through the Hessian. The Hessian is approximated using the Fisher Information Matrix, which is itself approximated using KFAC. The paper is clearly written and easy to follow, and is evaluated on a number of systems where the authors find that the proposed method achieves good compression ratios without requiring extensive hyperparameter tuning. The reviewers raised concerns about 1) the novelty of the work (which builds on the KFAC work of Martens and Grosse), 2) whether zeroing out individual connections as opposed to neurons will have practical runtime benefits, 3) the lack of comparisons against baselines on overall training time/complexity, 4) comparisons to work which directly prune as part of training (instead of the train-prune-finetune scheme adopted by the authors). In the view of the AC, 4) would be an interesting comparison but was not critical to the decision. Ultimately, the decision came down to the concern of lack of novelty and whether the proposed techniques would have an impact on runtime in practice. """
647,"""InstaGAN: Instance-aware Image-to-Image Translation""","['Image-to-Image Translation', 'Generative Adversarial Networks']","""Unsupervised image-to-image translation has gained considerable attention due to the recent impressive progress based on generative adversarial networks (GANs). However, previous methods often fail in challenging cases, in particular, when an image has multiple target instances and a translation task involves significant changes in shape, e.g., translating pants to skirts in fashion images. To tackle the issues, we propose a novel method, coined instance-aware GAN (InstaGAN), that incorporates the instance information (e.g., object segmentation masks) and improves multi-instance transfiguration. The proposed method translates both an image and the corresponding set of instance attributes while maintaining the permutation invariance property of the instances. To this end, we introduce a context preserving loss that encourages the network to learn the identity function outside of target instances. We also propose a sequential mini-batch inference/training technique that handles multiple instances with a limited GPU memory and enhances the network to generalize better for multiple instances. Our comparative evaluation demonstrates the effectiveness of the proposed method on different image datasets, in particular, in the aforementioned challenging cases. Code and results are available in pseudo-url""","""This paper addresses a promising method for unpaired cross-domain image-to-image translation that can accommodate multi-instance images. It extends the previously proposed CycleGAN model by taking into account per-instance segmentation masks. All three reviewers and AC agree that performing such transformation in general is a hard problem when significant changes in shape or appearance of the object have to be made, and that the proposed approach is sound and shows promising results. As rightly acknowledged by R1 The formulation is intuitive and well done! There are several potential weaknesses and suggestions to further strengthen this work: (1) R1 and R2 raised important concerns about the absence of baselines such as crop & attach simple baseline and CycleGAN+Seg. Pleased to report that the authors showed and discussed in their response some preliminary qualitative results regarding these baselines. In considering the author response and reviewer comments, the AC decided that the paper could be accepted given the comparison in the revised version, but the authors are strongly urged to include more results and evaluations on crop & attach baseline in the final revision if possible. (2) more quantitative results are needed for assessing the benefits of this approach (R3). The authors discussed in their response to R3 that more quantitative results such as the segmentation accuracy of the synthesized images are not possible since no ground-truth segmentation labels are available. This is true in general for unpaired image-to-image translation, however collecting annotations and performing such quantitative evaluation could have a substantial impact for assessing the significance of this work and can be seen as a recommendation for further improvement. (3) the proposed model performs translation for a pair of domains; extending the work to multi-domain translation like StarGAN by Choi et al 2018 or GANimation by Pumarola 2018 would strengthen the significance of the work. The authors discussed in their response to R3 that this is indeed possible. """
648,"""Detecting Topological Defects in 2D Active Nematics Using Convolutional Neural Networks""",[],"""Active matter consists of active agents which transform energy extracted from surroundings into momentum, producing a variety of collective phenomena. A model, synthetic active system composed of microtubule polymers driven by protein motors spontaneously forms a liquid-crystalline nematic phase. Extensile stress created by the protein motors precipitates continuous buckling and folding of the microtubules creating motile topological defects and turbulent fluid flows. Defect motion is determined by the rheological properties of the material; however, these remain largely unquantified. Measuring defects dynamics can yield fundamental insights into active nematics, a class of materials that include bacterial films and animal cells. Current methods for defect detection lack robustness and precision, and require fine-tuning for datasets with different visual quality. In this study, we applied Deep Learning to train a defect detector to automatically analyze microscopy videos of the microtubule active nematic. Experimental results indicate that our method is robust and accurate. It is expected to significantly increase the amount of video data that can be processed.""","""The reviewers raised a number of major concerns including the incremental novelty of the proposed (if any) and insufficient and unconvincing experimental evaluation presented. The authors did not provide any rebuttal. Hence, I cannot suggest this paper for presentation at ICLR."""
649,"""DialogWAE: Multimodal Response Generation with Conditional Wasserstein Auto-Encoder""","['dialogue', 'GAN', 'VAE', 'WAE', 'chatbot']","""Variational autoencoders (VAEs) have shown a promise in data-driven conversation modeling. However, most VAE conversation models match the approximate posterior distribution over the latent variables to a simple prior such as standard normal distribution, thereby restricting the generated responses to a relatively simple (e.g., single-modal) scope. In this paper, we propose DialogWAE, a conditional Wasserstein autoencoder (WAE) specially designed for dialogue modeling. Unlike VAEs that impose a simple distribution over the latent variables, DialogWAE models the distribution of data by training a GAN within the latent variable space. Specifically, our model samples from the prior and posterior distributions over the latent variables by transforming context-dependent random noise using neural networks and minimizes the Wasserstein distance between the two distributions. We further develop a Gaussian mixture prior network to enrich the latent space. Experiments on two popular datasets show that DialogWAE outperforms the state-of-the-art approaches in generating more coherent, informative and diverse responses.""","""This paper tackles the task of end-to-end systems for dialogue generation and proposes a novel, improved GAN for dialogue modeling, which adopts conditional Wasserstein Auto-Encoder to learn high-level representations of responses. In experiments, the proposed approach is compared to several state-of-the-art baselines on two dialog datasets, and improvements are shown both in terms of objective measures and human evaluation, making a strong support for the proposed approach. Two reviewers suggest similarities with a recent ICML paper on ARAE and request including reference to it and also request examples demonstrating differences, which are included in the latest version of the paper."""
650,"""k-Nearest Neighbors by Means of Sequence to Sequence Deep Neural Networks and Memory Networks""",[],"""k-Nearest Neighbors is one of the most fundamental but effective classification models. In this paper, we propose two families of models built on a sequence to sequence model and a memory network model to mimic the k-Nearest Neighbors model, which generate a sequence of labels, a sequence of out-of-sample feature vectors and a final label for classification, and thus they could also function as oversamplers. We also propose `out-of-core' versions of our models which assume that only a small portion of data can be loaded into memory. Computational experiments show that our models outperform k-Nearest Neighbors, a feed-forward neural network and a memory network, due to the fact that our models must produce additional output and not just the label. As an oversampler on imbalanced datasets, the sequence to sequence kNN model often outperforms Synthetic Minority Over-sampling Technique and Adaptive Synthetic Sampling. ""","""the proposed approach of predicting k nearest neighbouring examples as an auxiliary task is an interesting idea. however, the submission should have studied further on how those examples are predicted (e.g., sequence prediction is one, but you could try set prediction, or so on) rather than how sequential prediction of nearest neighbours is done together with different types of classifiers (many of which are arguably not necessarily suitable for classification,) which was a sentiment shared by all the reviewers. more careful investigation of different ways in which nearest neighbour prediction could be incorporated and more careful/thorough analysis on how the incorporation of this auxiliary task changes the behaviours or properties of the representation would make it a much better paper (also with clearer writing.)"""
651,"""Knowledge Distillation from Few Samples""","['knowledge distillation', 'few-sample learning', 'network compression']","""Current knowledge distillation methods require full training data to distill knowledge from a large ""teacher"" network to a compact ""student"" network by matching certain statistics between ""teacher"" and ""student"" such as softmax outputs and feature responses. This is not only time-consuming but also inconsistent with human cognition in which children can learn knowledge from adults with few examples. This paper proposes a novel and simple method for knowledge distillation from few samples. Taking the assumption that both ""teacher"" and ""student"" have the same feature map sizes at each corresponding block, we add a 1$ conv-layer at the end of each block in the student-net, and align the block-level outputs between ""teacher"" and ""student"" by estimating the parameters of the added layer with limited samples. We prove that the added layer can be absorbed/merged into the previous conv-layer \hl{to formulate a new conv-layer with the same size of parameters and computation cost as previous one. Experiments verifies that the proposed method is very efficient and effective to distill knowledge from teacher-net to student-net constructing in different ways on various datasets.""","""The paper considers the problem of knowledge distillation from a few samples. The proposed solution is to align feature representations of the student network with the teacher by adding 1x1 convolutions to each student block, and learning only the parameters of those layers. As noted by Reviewers 1 and 2, the performance of the proposed method is rather poor in absolute terms, and the use case considered (distillation from a few samples) is not motivated well enough. Reviewers also note the method is quite simplistic and incremental."""
652,"""Predicting the Generalization Gap in Deep Networks with Margin Distributions""","['Deep learning', 'large margin', 'generalization bounds', 'generalization gap.']","""As shown in recent research, deep neural networks can perfectly fit randomly labeled data, but with very poor accuracy on held out data. This phenomenon indicates that loss functions such as cross-entropy are not a reliable indicator of generalization. This leads to the crucial question of how generalization gap should be predicted from the training data and network parameters. In this paper, we propose such a measure, and conduct extensive empirical studies on how well it can predict the generalization gap. Our measure is based on the concept of margin distribution, which are the distances of training points to the decision boundary. We find that it is necessary to use margin distributions at multiple layers of a deep network. On the CIFAR-10 and the CIFAR-100 datasets, our proposed measure correlates very strongly with the generalization gap. In addition, we find the following other factors to be of importance: normalizing margin values for scale independence, using characterizations of margin distribution rather than just the margin (closest distance to decision boundary), and working in log space instead of linear space (effectively using a product of margins rather than a sum). Our measure can be easily applied to feedforward deep networks with any architecture and may point towards new training loss functions that could enable better generalization.""","""The paper suggests a new measurement of layer-wise margin distributions for generalization ability. Extensive experiments are conducted. Though there lacks a solid theory to explain the phenomenon. The majority of reviewers suggest acceptance (9,6,5). Therefore, it is proposed as probable accept."""
653,"""Function Space Particle Optimization for Bayesian Neural Networks""","['Bayesian neural networks', 'uncertainty estimation', 'variational inference']","""While Bayesian neural networks (BNNs) have drawn increasing attention, their posterior inference remains challenging, due to the high-dimensional and over-parameterized nature. To address this issue, several highly flexible and scalable variational inference procedures based on the idea of particle optimization have been proposed. These methods directly optimize a set of particles to approximate the target posterior. However, their application to BNNs often yields sub-optimal performance, as such methods have a particular failure mode on over-parameterized models. In this paper, we propose to solve this issue by performing particle optimization directly in the space of regression functions. We demonstrate through extensive experiments that our method successfully overcomes this issue, and outperforms strong baselines in a variety of tasks including prediction, defense against adversarial examples, and reinforcement learning.""","""Reviewers are in a consensus and recommended to accept after engaging with the authors. Please take reviewers' comments into consideration to improve your submission for the camera ready."""
654,"""A Unified Theory of Early Visual Representations from Retina to Cortex through Anatomically Constrained Deep CNNs""","['visual system', 'convolutional neural networks', 'efficient coding', 'retina']","""The vertebrate visual system is hierarchically organized to process visual information in successive stages. Neural representations vary drastically across the first stages of visual processing: at the output of the retina, ganglion cell receptive fields (RFs) exhibit a clear antagonistic center-surround structure, whereas in the primary visual cortex (V1), typical RFs are sharply tuned to a precise orientation. There is currently no unified theory explaining these differences in representations across layers. Here, using a deep convolutional neural network trained on image recognition as a model of the visual system, we show that such differences in representation can emerge as a direct consequence of different neural resource constraints on the retinal and cortical networks, and for the first time we find a single model from which both geometries spontaneously emerge at the appropriate stages of visual processing. The key constraint is a reduced number of neurons at the retinal output, consistent with the anatomy of the optic nerve as a stringent bottleneck. Second, we find that, for simple downstream cortical networks, visual representations at the retinal output emerge as nonlinear and lossy feature detectors, whereas they emerge as linear and faithful encoders of the visual scene for more complex cortical networks. This result predicts that the retinas of small vertebrates (e.g. salamander, frog) should perform sophisticated nonlinear computations, extracting features directly relevant to behavior, whereas retinas of large animals such as primates should mostly encode the visual scene linearly and respond to a much broader range of stimuli. These predictions could reconcile the two seemingly incompatible views of the retina as either performing feature extraction or efficient coding of natural scenes, by suggesting that all vertebrates lie on a spectrum between these two objectives, depending on the degree of neural resources allocated to their visual system.""","""The paper advocates neuroscience-based V1 models to adapt CNNs. The results of the simulations are convincing from a neuroscience-perspective. The reviewers equivocally recommend publication."""
655,"""Visual Imitation Learning with Recurrent Siamese Networks""","['Reinforcement Learning', 'Imitation Learning', 'Deep Learning']","""People are incredibly skilled at imitating others by simply observing them. They achieve this even in the presence of significant morphological differences and capabilities. Further, people are able to do this from raw perceptions of the actions of others, without direct access to the abstracted demonstration actions and with only partial state information. People therefore solve a difficult problem of understanding the salient features of both observations of others and the relationship to their own state when learning to imitate specific tasks. However, we can attempt to reproduce a similar demonstration via trail and error and through this gain more understanding of the task space. To reproduce this ability an agent would need to both learn how to recognize the differences between itself and some demonstration and at the same time learn to minimize the distance between its own performance and that of the demonstration. In this paper we propose an approach using only visual information to learn a distance metric between agent behaviour and a given video demonstration. We train an RNN-based siamese model to compute distances in space and time between motion clips while training an RL policy to minimize this distance. Furthermore, we examine a particularly challenging form of this problem where the agent must learn an imitation based task given a single demonstration. We demonstrate our approach in the setting of deep learning based control for physical simulation of humanoid walking in both 2D with pseudo-formula degrees of freedom (DoF) and 3D with pseudo-formula DoF.""","""This paper proposes an approach for imitation learning from video data. The problem is important and the contribution is timely. The reviewers brought up several concerns regarding the clarity of the paper and the lack of sufficient comparisons. The authors have improved the paper significantly, adding several new comparisons and improving the presentation. However, concerns still remain regarding the description of the method and the presentation of the results. Hence, the reviewers agree that the paper does not meet the bar for publication. """
656,"""Adaptive Input Representations for Neural Language Modeling""",['Neural language modeling'],"""We introduce adaptive input representations for neural language modeling which extend the adaptive softmax of Grave et al. (2017) to input representations of variable capacity. There are several choices on how to factorize the input and output layers, and whether to model words, characters or sub-word units. We perform a systematic comparison of popular choices for a self-attentional architecture. Our experiments show that models equipped with adaptive embeddings are more than twice as fast to train than the popular character input CNN while having a lower number of parameters. On the WikiText-103 benchmark we achieve 18.7 perplexity, an improvement of 10.5 perplexity compared to the previously best published result and on the Billion Word benchmark, we achieve 23.02 perplexity.""","""There is a clear consensus among the reviews to accept this submission thus I am recommending acceptance. The paper makes a clear, if modest, contribution to language modeling that is likely to be valuable to many other researchers."""
657,"""Cross-Task Knowledge Transfer for Visually-Grounded Navigation""",[],"""Recent efforts on training visual navigation agents conditioned on language using deep reinforcement learning have been successful in learning policies for two different tasks: learning to follow navigational instructions and embodied question answering. In this paper, we aim to learn a multitask model capable of jointly learning both tasks, and transferring knowledge of words and their grounding in visual objects across tasks. The proposed model uses a novel Dual-Attention unit to disentangle the knowledge of words in the textual representations and visual objects in the visual representations, and align them with each other. This disentangled task-invariant alignment of representations facilitates grounding and knowledge transfer across both tasks. We show that the proposed model outperforms a range of baselines on both tasks in simulated 3D environments. We also show that this disentanglement of representations makes our model modular, interpretable, and allows for zero-shot transfer to instructions containing new words by leveraging object detectors.""","""The authors have proposed a language+vision 'dual' attention architecture, trained in a multitask setting across SGN and EQA in vizDoom, to allow for knowledge grounding. The paper is interesting to read. The complex architecture is very clearly described and motivated, and the knowledge grounding problem is ambitious and relevant. However, the actual proposed solution does not make a novel contribution and the reviewers were unconvinced that the approach would be at all scalable to natural language or more complex tasks. In addition, the question was raised as to whether the 'knowledge grounding' claims by the authors are actually much more shallow associations of color and shape that are beneficial in cluttered environments. This is a borderline case, but the AC agrees that the paper falls a bit short of its goals."""
658,"""Large-Scale Answerer in Questioner's Mind for Visual Dialog Question Generation""",[],"""Answerer in Questioner's Mind (AQM) is an information-theoretic framework that has been recently proposed for task-oriented dialog systems. AQM benefits from asking a question that would maximize the information gain when it is asked. However, due to its intrinsic nature of explicitly calculating the information gain, AQM has a limitation when the solution space is very large. To address this, we propose AQM+ that can deal with a large-scale problem and ask a question that is more coherent to the current context of the dialog. We evaluate our method on GuessWhich, a challenging task-oriented visual dialog problem, where the number of candidate classes is near 10K. Our experimental results and ablation studies show that AQM+ outperforms the state-of-the-art models by a remarkable margin with a reasonable approximation. In particular, the proposed AQM+ reduces more than 60% of error as the dialog proceeds, while the comparative algorithms diminish the error by less than 6%. Based on our results, we argue that AQM+ is a general task-oriented dialog algorithm that can be applied for non-yes-or-no responses. ""","""Important problem (visually grounded dialog); incremental (but not in a negative sense of the word) extension of prior work to an important new setting (GuessWhich); well-executed. Paper was reviewed by three experts. Initially there were some concerns but after the author response and reviewer discussion, all three unanimously recommend acceptance. """
659,""" pseudo-formula sampling with probability matching""",[],"""Probabilistic methods often need to draw samples from a nontrivial distribution. pseudo-formula sampling is a nice algorithm by building upon a top-down construction of a Gumbel process, where a large state space is divided into subsets and at each round pseudo-formula sampling selects a subset to process. However, the selection rule depends on a bound function, which can be intractable. Moreover, we show that such a selection criterion can be inefficient. This paper aims to improve pseudo-formula sampling by addressing these issues. To design a suitable selection rule, we apply \emph{Probability Matching}, a widely used method for decision making, to pseudo-formula sampling. We provide insights into the relationship between pseudo-formula sampling and probability matching by analyzing a nontrivial special case in which the state space is partitioned into two subsets. We show that in this case probability matching is optimal within a constant gap. Furthermore, as directly applying probability matching to pseudo-formula sampling is time consuming, we design an approximate version based on Monte-Carlo estimators. We also present an efficient implementation by leveraging special properties of Gumbel distributions and well-designed balanced trees. Empirical results show that our method saves a significantly amount of computational resources on suboptimal regions compared with pseudo-formula sampling.""","""This paper applied probability matching to A* sampling in order to provide an approximate variant without a bound function. It is a novel idea and a good contribution to the A* sampling family. The authors also provided regret analysis for the adoption of PM. However, as pointed out by R1 and R3, the authors failed to clarify the approximation introduced by the PM and its implication in the output samples. The empirical comparison should also take into account this difference. Further analysis of the bias in the sample distribution would also help clarify the pros and cons of the proposed method. R3 also raised the concern that the description of the preliminary section and the main contribution in section 4 was not clear."""
660,"""Meta-Learning Neural Bloom Filters""","['meta-learning', 'memory', 'one-shot learning', 'bloom filter', 'set membership', 'familiarity', 'compression']","""There has been a recent trend in training neural networks to replace data structures that have been crafted by hand, with an aim for faster execution, better accuracy, or greater compression. In this setting, a neural data structure is instantiated by training a network over many epochs of its inputs until convergence. In many applications this expensive initialization is not practical, for example streaming algorithms --- where inputs are ephemeral and can only be inspected a small number of times. In this paper we explore the learning of approximate set membership over a stream of data in one-shot via meta-learning. We propose a novel memory architecture, the Neural Bloom Filter, which we show to be more compressive than Bloom Filters and several existing memory-augmented neural networks in scenarios of skewed data or structured sets.""","""This work proposes and interesting approach to learn approximate set membership. While the proposed architecture is rather closely related to existing work, it is still interesting, as recognized by reviewers. Authors's substantial rewrites has also helped make the paper clearer. However, the empirical merits of the approach are still a bit limited; when combined with the narrow novelty compared to existing work, this makes the overall contribution a bit too thin for ICLR. Authors are encouraged to strengthen their work by showing more convincing practical benefit of their approach."""
661,"""Online Hyperparameter Adaptation via Amortized Proximal Optimization""","['hyperparameters', 'optimization', 'learning rate adaptation']","""Effective performance of neural networks depends critically on effective tuning of optimization hyperparameters, especially learning rates (and schedules thereof). We present Amortized Proximal Optimization (APO), which takes the perspective that each optimization step should approximately minimize a proximal objective (similar to the ones used to motivate natural gradient and trust region policy optimization). Optimization hyperparameters are adapted to best minimize the proximal objective after one weight update. We show that an idealized version of APO (where an oracle minimizes the proximal objective exactly) achieves global convergence to stationary point and locally second-order convergence to global optimum for neural networks. APO incurs minimal computational overhead. We experiment with using APO to adapt a variety of optimization hyperparameters online during training, including (possibly layer-specific) learning rates, damping coefficients, and gradient variance exponents. For a variety of network architectures and optimization algorithms (including SGD, RMSprop, and K-FAC), we show that with minimal tuning, APO performs competitively with carefully tuned optimizers.""","""This paper proposes an amortized proximal optimization method to adapt optimization hyperparameters. Empirical results on many problems are performed. Reviewers overall find the ideas interesting, however there still are some questions whether strong baselines are used in the experimental comparisons. The reviewers also point that the theoretical results are not useful ones since the assumptions are not satisfied in practice. One of the reviewer increased their score, but the other has maintained that the paper requires more work. The presentation of the result is also a bit problematic; the font sizes in the figure are too small to read. The paper contains interesting ideas, but it does not make the bar for acceptance in ICLR. Therefore I recommend a reject. I encourage the authors to resubmit this work after improving the presentation and experiments. """
662,"""Accelerating first order optimization algorithms""","['Optimization', 'Optimizer', 'Adam', 'Gradient Descent']","""There exist several stochastic optimization algorithms. However in most cases, it is difficult to tell for a particular problem which will be the best optimizer to choose as each of them are good. Thus, we present a simple and intuitive technique, when applied to first order optimization algorithms, is able to improve the speed of convergence and reaches a better minimum for the loss function compared to the original algorithms. The proposed solution modifies the update rule, based on the variation of the direction of the gradient during training. We conducted several tests with Adam and AMSGrad on two different datasets. The preliminary results show that the proposed technique improves the performance of existing optimization algorithms and works well in practice.""","""Dear authors, All reviewers commented that the paper had issues with the presentations and the results, making it unsuitable for publication to ICLR. Please address these comments should you decide to resubmit this work."""
663,"""Perception-Aware Point-Based Value Iteration for Partially Observable Markov Decision Processes""","['partially observable Markov decision processes', 'active perception', 'submodular optimization', 'point-based value iteration', 'reinforcement learning']","""Partially observable Markov decision processes (POMDPs) are a widely-used framework to model decision-making with uncertainty about the environment and under stochastic outcome. In conventional POMDP models, the observations that the agent receives originate from fixed known distribution. However, in a variety of real-world scenarios the agent has an active role in its perception by selecting which observations to receive. Due to combinatorial nature of such selection process, it is computationally intractable to integrate the perception decision with the planning decision. To prevent such expansion of the action space, we propose a greedy strategy for observation selection that aims to minimize the uncertainty in state. We develop a novel point-based value iteration algorithm that incorporates the greedy strategy to achieve near-optimal uncertainty reduction for sampled belief points. This in turn enables the solver to efficiently approximate the reachable subspace of belief simplex by essentially separating computations related to perception from planning. Lastly, we implement the proposed solver and demonstrate its performance and computational advantage in a range of robotic scenarios where the robot simultaneously performs active perception and planning.""","""This was a borderline paper and a very difficult decision to make. The paper addresses a potentially interesting problem in approximate POMDP planning, based on simplifying assumptions that perception can be decoupled from action and that a set of sensors exhibits certain conditional independence structure. As a result, a simple approach can be devised that incorporates a simple greedy perception method within a point-based value iteration scheme. Unfortunately, the assumptions the paper makes are so strong and seemingly artificial to the extent that they appear reverse engineered to the use of a simple perception heuristic. In principle, such a simplification might not be a problem if the resulting formulation captured practically important scenarios, but that was not convincingly achieved in this paper---indeed, another major limitation of the paper is its weak motivation. In more detail, the proposed approach relies on decoupling of perception and action, which is a restrictive assumption that bypasses the core issue of exploration versus exploitation in POMDPS. As model of active perception, the proposal is simplistic and somewhat artificial; the motivation for the particular cost model (cardinality of the sensor set) is particularly weak---a point that was not convincingly defended in the discussion. Perhaps the biggest underlying weakness is the experimental evaluation, which is inadequate to support a claim that the proposed methods show meaningful advantages over state-of-the-art approaches in important scenarios. A reviewer also raised legitimate questions about the strength of the theoretical analysis. In the end, the reviewers did not disagree on any substantive technical matter, but nevertheless did disagree in their assessments of the significance of the contribution. This is clearly a borderline paper, which on the positive side, was competently executed, but on the negative side, is pursuing an artificial scenario that enables a particularly simple algorithmic approach. Despite the lack of consensus, a difficult decision has to be made nonetheless. In the end, my judgement is that the paper is not yet strong enough for publication. I would recommend the authors significantly strengthen the experimental evaluation to cover off at least two of the major shortcomings of the current paper: (1) The true utility of the proposed method needs to be better established against stronger baselines in more realistic scenarios. (2) The relevance of the restrictive assumptions needs to be more convincingly established by providing concrete, realistic and more challenging case studies where the proposed techniques are still applicable. The paper would also be improved if the theoretical analysis could be strengthened to better address the criticisms of Reviewer 4."""
664,"""Value Propagation Networks""","['Reinforcement Learning', 'Value Iteration', 'Navigation', 'Convolutional Neural Networks', 'Learning to plan']","""We present Value Propagation (VProp), a set of parameter-efficient differentiable planning modules built on Value Iteration which can successfully be trained using reinforcement learning to solve unseen tasks, has the capability to generalize to larger map sizes, and can learn to navigate in dynamic environments. We show that the modules enable learning to plan when the environment also includes stochastic elements, providing a cost-efficient learning system to build low-level size-invariant planners for a variety of interactive navigation problems. We evaluate on static and dynamic configurations of MazeBase grid-worlds, with randomly generated environments of several different sizes, and on a StarCraft navigation scenario, with more complex dynamics, and pixels as input.""",""" Interesting idea, reviewers were positive and indicated presentation should be improved. """
665,"""Geometry aware convolutional filters for omnidirectional images representation""","['omnidirectional images', 'classification', 'deep learning', 'graph signal processing']","""Due to their wide field of view, omnidirectional cameras are frequently used by autonomous vehicles, drones and robots for navigation and other computer vision tasks. The images captured by such cameras, are often analysed and classified with techniques designed for planar images that unfortunately fail to properly handle the native geometry of such images. That results in suboptimal performance, and lack of truly meaningful visual features. In this paper we aim at improving popular deep convolutional neural networks so that they can properly take into account the specific properties of omnidirectional data. In particular we propose an algorithm that adapts convolutional layers, which often serve as a core building block of a CNN, to the properties of omnidirectional images. Thus, our filters have a shape and size that adapts with the location on the omnidirectional image. We show that our method is not limited to spherical surfaces and is able to incorporate the knowledge about any kind of omnidirectional geometry inside the deep learning network. As depicted by our experiments, our method outperforms the existing deep neural network techniques for omnidirectional image classification and compression tasks.""","""Strengths: This paper proposed to use graph-based deep learning methods to apply deep learning techniques to images coming from omnidirectional cameras. Weaknesses: The projected MNIST dataset looks very localized on the sphere and therefore does not seem to leverage that much of the global connectivity of the graph All reviewers pointed out limitations in the experimental results. There were significant concerns about the relation of the model to the existing literature. It was pointed out that both the comparison to other methodology, and empirical comparisons were lacking. The paper received three reject recommendations. There was some discussion with reviewers, which emphasized open issues in the comparison to and references to existing literature as highlighted by contributed comment from Michael Bronstein. Work is clearly not mature enough at this point for ICLR, insufficient comparisons / illustrations"""
666,"""An Efficient and Margin-Approaching Zero-Confidence Adversarial Attack""","['adversarial attack', 'zero-confidence attack']","""There are two major paradigms of white-box adversarial attacks that attempt to impose input perturbations. The first paradigm, called the fix-perturbation attack, crafts adversarial samples within a given perturbation level. The second paradigm, called the zero-confidence attack, finds the smallest perturbation needed to cause misclassification, also known as the margin of an input feature. While the former paradigm is well-resolved, the latter is not. Existing zero-confidence attacks either introduce significant approximation errors, or are too time-consuming. We therefore propose MarginAttack, a zero-confidence attack framework that is able to compute the margin with improved accuracy and efficiency. Our experiments show that MarginAttack is able to compute a smaller margin than the state-of-the-art zero-confidence attacks, and matches the state-of-the-art fix-perturbation attacks. In addition, it runs significantly faster than the Carlini-Wagner attack, currently the most accurate zero-confidence attack algorithm.""","""The paper proposes a new method for adversarial attacks, MarginAttack, which finds adversarial examples with small distortion and runs faster than the CW baseline, but slower than other methods. The authors provide theoretical guarantees and a broad set of experiments. In the discussion, a consistent concern has been that, experimentally, the method does not perform noticeably better than previous approaches. The authors mention that the lines are too thick to reveal the difference. It has been pointed out that this might be related to the way the experiments are conducted, but the proposed method still does better than other methods. AnonReviewer1 mentions that the assumptions needed for the theoretical part might be too strong, meaning that the main contribution of the paper is in the experimental side. The comparisons with other methods and the assumptions made in the theorems seem to have caused quite some confusion and there was a fair amount of discussion. Following the discussion session, AnonReviewer1 updated his rating from 5 to 6 with high confidence. The referees all rate the paper as not very strong, with one marginally above acceptance threshold and two marginally below the acceptance threshold. Although the paper seems to propose valuable ideas, and it appears that the discussion has clarified many questions from the initial submission, the paper has not provided a clear, convincing, selling point at this time. """
667,"""MixFeat: Mix Feature in Latent Space Learns Discriminative Space""","['regularization', 'generalization', 'image classification', 'latent space', 'feature learning']","""Deep learning methods perform well in various tasks. However, the over-fitting problem, which causes the performance to decrease for unknown data, remains. We hence propose a method named MixFeat that directly creates latent spaces in a network that can distinguish classes. MixFeat mixes two feature maps in each latent space in the network and uses unmixed labels for learning. We discuss the difference between a method that mixes only features (MixFeat) and a method that mixes both features and labels (mixup and its family). Mixing features repeatedly is effective in expanding feature diversity, but mixing labels repeatedly makes learning difficult. MixFeat makes it possible to obtain the advantages of repeated mixing by mixing only features. We report improved results obtained using existing network models with MixFeat on CIFAR-10/100 datasets. In addition, we show that MixFeat effectively reduces the over-fitting problem even when the training dataset is small or contains errors. MixFeat is easy to implement and can be added to various network models without additional computational cost in the inference phase.""","""The paper describes a method to improve generalization by mixing examples in the hidden space. Experiments on CIFAR-10 and CIFAR-100 showed that the proposed method improves the generalization of the networks. The reviewers found these results promising, but argue that the experimental section was too weak in its current form - notable lacking experiments on larger scale datasets such as Imagenet. Notably the paper should compare more with the relevant baselines to better understand its significance."""
668,"""Overcoming Multi-model Forgetting""","['multi-model forgetting', 'deep learning', 'machine learning', 'multi-model training', 'neural architecture search']","""We identify a phenomenon, which we refer to as *multi-model forgetting*, that occurs when sequentially training multiple deep networks with partially-shared parameters; the performance of previously-trained models degrades as one optimizes a subsequent one, due to the overwriting of shared parameters. To overcome this, we introduce a statistically-justified weight plasticity loss that regularizes the learning of a model's shared parameters according to their importance for the previous models, and demonstrate its effectiveness when training two models sequentially and for neural architecture search. Adding weight plasticity in neural architecture search preserves the best models to the end of the search and yields improved results in both natural language processing and computer vision tasks.""",""" pros: - nicely written paper - clear and precise with a derivation of the loss function cons: novelty/impact: I think all the reviewers acknowledge that you are doing something different in the neural brainwashing (NB) problem than is done in the typical catastropic forgetting (CF) setting. You have one dataset and a set of models with shared weights; the CF setting has one model and trains on different datasets/tasks. But whereas solving the CF problem would solve a major problem of continual machine learning, the value of solving the NB problem is harder to assess from this paper... The main application seems to be improving neural architecture search. At the meta-level, the techniques used to derive the main loss are already well known and the result similar to EWC, so they don't add a lot from the analysis perspective. I think it would be very helpful to revise the paper to show a range of applications that could benefit from solving the NB problem and that the technique you propose applies more broadly."""
669,"""Collapse of deep and narrow neural nets""","['neural networks', 'deep and narrow', 'ReLU', 'collapse']","""Recent theoretical work has demonstrated that deep neural networks have superior performance over shallow networks, but their training is more difficult, e.g., they suffer from the vanishing gradient problem. This problem can be typically resolved by the rectified linear unit (ReLU) activation. However, here we show that even for such activation, deep and narrow neural networks (NNs) will converge to erroneous mean or median states of the target function depending on the loss with high probability. Deep and narrow NNs are encountered in solving partial differential equations with high-order derivatives. We demonstrate this collapse of such NNs both numerically and theoretically, and provide estimates of the probability of collapse. We also construct a diagram of a safe region for designing NNs that avoid the collapse to erroneous states. Finally, we examine different ways of initialization and normalization that may avoid the collapse problem. Asymmetric initializations may reduce the probability of collapse but do not totally eliminate it.""","""The paper studies difficulties in training deep and narrow networks. It shows that there is high probability that deep and narrow ReLU networks will converge to an erroneous state, depending on the type of training that is employed. The results add to our current understanding of the limitations of these architectures. The main criticism is that the analysis might be very limited, being restricted to very narrow networks (of width about 10 or less) which are not very common in practice, and that the observed collapse phenomenon can be easily addressed by non symmetric initialization. There were some issues with the proofs that were covered in the discussed between authors and reviewers. The revision is relatively extensive. This is a borderline case. The paper receives one good rating, one negative rating, and a borderline accept rating. Although the paper contributes interesting insights to a relevant problem that clearly needs contributions in this direction, the analysis presented in the paper and its applicability in practice seems to be very restrictive at this point. """
670,"""Variational Smoothing in Recurrent Neural Network Language Models""",[],"""We present a new theoretical perspective of data noising in recurrent neural network language models (Xie et al., 2017). We show that each variant of data noising is an instance of Bayesian recurrent neural networks with a particular variational distribution (i.e., a mixture of Gaussians whose weights depend on statistics derived from the corpus such as the unigram distribution). We use this insight to propose a more principled method to apply at prediction time and propose natural extensions to data noising under the variational framework. In particular, we propose variational smoothing with tied input and output embedding matrices and an element-wise variational smoothing method. We empirically verify our analysis on two benchmark language modeling datasets and demonstrate performance improvements over existing data noising methods.""","""as r1 and r2 have pointed out, this work presents an interesting and potentially more generalizable extension of the earlier work on introducing noise as regularization in autoregressive language modelling. although it would have been better with more extensive evaluation that goes beyond unsupervised language modelling and toward conditional language modelling, but i believe this is all fine for this further work to be left as follow-up. r3's concern is definitely valid, but i believe the existing evaluation set as well as exposition merit presentation and discussion at the conference, which was shared by the other reviewers as well as a programme chair."""
671,"""Transferring Knowledge across Learning Processes""","['meta-learning', 'transfer learning']","""In complex transfer learning scenarios new tasks might not be tightly linked to previous tasks. Approaches that transfer information contained only in the final parameters of a source model will therefore struggle. Instead, transfer learning at at higher level of abstraction is needed. We propose Leap, a framework that achieves this by transferring knowledge across learning processes. We associate each task with a manifold on which the training process travels from initialization to final parameters and construct a meta-learning objective that minimizes the expected length of this path. Our framework leverages only information obtained during training and can be computed on the fly at negligible cost. We demonstrate that our framework outperforms competing methods, both in meta-learning and transfer learning, on a set of computer vision tasks. Finally, we demonstrate that Leap can transfer knowledge across learning processes in demanding reinforcement learning environments (Atari) that involve millions of gradient steps.""","""This paper proposes an approach for learning to transfer knowledge across multiple tasks. It develops a principled approach for an important problem in meta-learning (short horizon bias). Nearly all of the reviewer's concerns were addressed throughout the discussion phase. The main weakness is that the experimental settings are somewhat non-standard (i.e. the Omniglot protocol in the paper is not at all standard). I would encourage the authors to mention the discrepancies from more standard protocols in the paper, to inform the reader. The results are strong nonetheless, evaluating in settings where typical meta-learning algorithms would struggle. The reviewers and I all agree that the paper should be accepted, and I think it should be considered for an oral presentation."""
672,"""ACCELERATING NONCONVEX LEARNING VIA REPLICA EXCHANGE LANGEVIN DIFFUSION""",[],"""Langevin diffusion is a powerful method for nonconvex optimization, which enables the escape from local minima by injecting noise into the gradient. In particular, the temperature parameter controlling the noise level gives rise to a tradeoff between ``global exploration'' and ``local exploitation'', which correspond to high and low temperatures. To attain the advantages of both regimes, we propose to use replica exchange, which swaps between two Langevin diffusions with different temperatures. We theoretically analyze the acceleration effect of replica exchange from two perspectives: (i) the convergence in pseudo-formula -divergence, and (ii) the large deviation principle. Such an acceleration effect allows us to faster approach the global minima. Furthermore, by discretizing the replica exchange Langevin diffusion, we obtain a discrete-time algorithm. For such an algorithm, we quantify its discretization error in theory and demonstrate its acceleration effect in practice. ""","""The main criticisms were around novelty: that the analysis is rather standard. Given that all the reviewers agreed the paper is well written, I'm inclined to think the paper will be a useful contribution to the literature. The authors also highlight the analysis of the discretization, which seems to be missed by the most critical reviewer. I would suggest to the reviewers that they use the criticisms to rework the paper's introduction, to better explain which parts of the work are novel and which parts are standard. I would also suggest that standard background be moved to the appendix so that it is there for the nonexpert, while making the body of the work more focused on the novel aspects."""
673,"""Unsupervised Exploration with Deep Model-Based Reinforcement Learning""","['exploration', 'model based reinforcement learning']","""Reinforcement learning (RL) often requires large numbers of trials to solve a single specific task. This is in sharp contrast to human and animal learning: humans and animals can use past experience to acquire an understanding about the world, which they can then use to perform new tasks with minimal additional learning. In this work, we study how an unsupervised exploration phase can be used to build up such prior knowledge, which can then be utilized in a second phase to perform new tasks, either directly without any additional exploration, or through minimal fine-tuning. A critical question with this approach is: what kind of knowledge should be transferred from the unsupervised phase to the goal-directed phase? We argue that model-based RL offers an appealing solution. By transferring models, which are task-agnostic, we can perform new tasks without any additional learning at all. However, this relies on having a suitable exploration method during unsupervised training, and a model-based RL method that can effectively utilize modern high-capacity parametric function classes, such as deep neural networks. We show that both challenges can be addressed by representing model-uncertainty, which can both guide exploration in the unsupervised phase and ensure that the errors in the model are not exploited by the planner in the goal-directed phase. We illustrate, on simple simulated benchmark tasks, that our method can perform various goal-directed skills on the first attempt, and can improve further with fine-tuning, exceeding the performance of alternative exploration methods.""","""Strengths The paper proposes to include exploration for the PETS (probabilistic ensembles with trajectory sampling) approach to learning the state transition function. The paper is clearly written. Weaknesses All reviewers are in agreement regarding a number of key weaknesses: limited novelty, limited evaluation, and aspects of the paper are difficult to follow or are sparse on details. No revisions have been posted. Summary All reviewers are in agreement that the paper requires significant work and that it is not ready for ICLR publication. """
674,"""Shrinkage-based Bias-Variance Trade-off for Deep Reinforcement Learning""","['bias-variance trade-off', 'James-stein estimator', 'reinforcement learning']","""Deep reinforcement learning has achieved remarkable successes in solving various challenging artificial intelligence tasks. A variety of different algorithms have been introduced and improved towards human-level performance. Although technical advances have been developed for each individual algorithms, there has been strong evidence showing that further substantial improvements can be achieved by properly combining multiple approaches with difference biases and variances. In this work, we propose to use the James-Stein (JS) shrinkage estimator to combine on-policy policy gradient estimators which have low bias but high variance, with low-variance high-bias gradient estimates such as those constructed based on model-based methods or temporally smoothed averaging of historical gradients. Empirical results show that our simple shrinkage approach is very effective in practice and substantially improve the sample efficiency of the state-of-the-art on-policy methods on various continuous control tasks. ""","""The paper introduces the use of J-S shrinkage estimator in policy optimization, which is new and promising. The results also show the potential. That said, reviewers are not fully convinced that in its current stage the paper is ready for publication. The approach taken here is essentially a combination existing techniques. While it is useful, more work is probably needed to strengthen the contribution. A few directions have been suggested by reviewers, including theoretical guarantees and stronger empirical support."""
675,"""Diffusion Scattering Transforms on Graphs""","['graph neural networks', 'deep learning', 'stability', 'scattering transforms', 'convolutional neural networks']","""Stability is a key aspect of data analysis. In many applications, the natural notion of stability is geometric, as illustrated for example in computer vision. Scattering transforms construct deep convolutional representations which are certified stable to input deformations. This stability to deformations can be interpreted as stability with respect to changes in the metric structure of the domain. In this work, we show that scattering transforms can be generalized to non-Euclidean domains using diffusion wavelets, while preserving a notion of stability with respect to metric changes in the domain, measured with diffusion maps. The resulting representation is stable to metric perturbations of the domain while being able to capture ''high-frequency'' information, akin to the Euclidean Scattering. ""","""The paper gives an extension of scattering transform to non-Euclidean domains by introducing scattering transforms on graphs using diffusion wavelet representations, and presents a stability analysis of such a representation under deformation of the underlying graph metric defined in terms of graph diffusion. Concerns of the reviewers are primarily with what type of graphs is the primary consideration (small world social networks or point cloud submanifold samples) and experimental studies. Technical development like deformation in the proposed graph metric is motivated by sub-manifold scenarios in computer vision, and whether the development is well suitable to social networks in experiments still needs further investigations. The authors make satisfied answers to the reviewers questions. The reviewers unanimously accept the paper for ICLR publication."""
676,"""Energy-Constrained Compression for Deep Neural Networks via Weighted Sparse Projection and Layer Input Masking""","['model compression', 'inference energy saving', 'deep neural network pruning']","""Deep Neural Networks (DNNs) are increasingly deployed in highly energy-constrained environments such as autonomous drones and wearable devices while at the same time must operate in real-time. Therefore, reducing the energy consumption has become a major design consideration in DNN training. This paper proposes the first end-to-end DNN training framework that provides quantitative energy consumption guarantees via weighted sparse projection and input masking. The key idea is to formulate the DNN training as an optimization problem in which the energy budget imposes a previously unconsidered optimization constraint. We integrate the quantitative DNN energy estimation into the DNN training process to assist the constrained optimization. We prove that an approximate algorithm can be used to efficiently solve the optimization problem. Compared to the best prior energy-saving techniques, our framework trains DNNs that provide higher accuracies under same or lower energy budgets.""","""All of the reviewers agree that this is a well-written paper with the novel perspective of minimizing energy consumption in neural networks, as opposed to maximizing sparsity, which does not always correlate with energy cost. There are a number of promised clarifications and additional results that have emerged from the discussion that should be put into the final draft. Namely, describing the overhead of converting from sparse to dense representations, adding the Imagenet sparsity results, and adding the time taken to run the projection step."""
677,"""Data-Dependent Coresets for Compressing Neural Networks with Applications to Generalization Bounds""","['coresets', 'neural network compression', 'generalization bounds', 'matrix sparsification']","""We present an efficient coresets-based neural network compression algorithm that sparsifies the parameters of a trained fully-connected neural network in a manner that provably approximates the network's output. Our approach is based on an importance sampling scheme that judiciously defines a sampling distribution over the neural network parameters, and as a result, retains parameters of high importance while discarding redundant ones. We leverage a novel, empirical notion of sensitivity and extend traditional coreset constructions to the application of compressing parameters. Our theoretical analysis establishes guarantees on the size and accuracy of the resulting compressed network and gives rise to generalization bounds that may provide new insights into the generalization properties of neural networks. We demonstrate the practical effectiveness of our algorithm on a variety of neural network configurations and real-world data sets.""","""The reviewers and AC note that the strength of the paper includes a) an interesting compression algorithm of neural networks with provable guarantees (under some assumptions), b) solid experimental comparison with the existing *matrix sparsification* algorithms. The AC's main concern of the experimental part of the paper is that it doesn't outperform or match the performance of the ""vanilla"" neural network compression algorithms such as Han et al'15. The AC decided to suggest acceptance for the paper but also strongly encourage the paper to clarify the algorithms in comparison don't include state-of-the-art compression algorithms. """
678,"""Unified recurrent network for many feature types""","['sparse', 'recurrent', 'asynchronous', 'time', 'series']","""There are time series that are amenable to recurrent neural network (RNN) solutions when treated as sequences, but some series, e.g. asynchronous time series, provide a richer variation of feature types than current RNN cells take into account. In order to address such situations, we introduce a unified RNN that handles five different feature types, each in a different manner. Our RNN framework separates sequential features into two groups dependent on their frequency, which we call sparse and dense features, and which affect cell updates differently. Further, we also incorporate time features at the sequential level that relate to the time between specified events in the sequence and are used to modify the cell's memory state. We also include two types of static (whole sequence level) features, one related to time and one not, which are combined with the encoder output. The experiments show that the proposed modeling framework does increase performance compared to standard cells.""","""This paper presents an algorithm for combining various feature types when training recurrent networks. The features are handled by modifying the update rules and cell states based on the features' type -- dense, sparse, static, w/ decay, etc. Strengths - The model handles each feature according to its type and handles cell state and transitions appropriately. - Extends earlier work to handle more feature types, like sparse features. Weaknesses - Limited novelty. Models similar to various aspects of the proposed system have been presented in prior works. For example: TLSTM, which the authors use as a baseline. Although some components are novel, like the treatment of sparse features, contributions, in my opinion, are not sufficient to be accepted at ICLR. - Presentation: Confusing and not enough information for reproducing results; multiple reviewers raised concerns about presentation of the feature types and experimental results. There were suggestions to improve, which the authors did consider during revision, but some concerns still remain. In the end, the reviewers agreed about the limited novelty of this work, given existing literature. The recommendation, therefore, is to reject the paper. """
679,"""Safe Policy Learning from Observations""","['learning from observations', 'safe reinforcement learning', 'deep reinforcement learning']","""In this paper, we consider the problem of learning a policy by observing numerous non-expert agents. Our goal is to extract a policy that, with high-confidence, acts better than the agents' average performance. Such a setting is important for real-world problems where expert data is scarce but non-expert data can easily be obtained, e.g. by crowdsourcing. Our approach is to pose this problem as safe policy improvement in reinforcement learning. First, we evaluate an average behavior policy and approximate its value function. Then, we develop a stochastic policy improvement algorithm that safely improves the average behavior. The primary advantages of our approach, termed Rerouted Behavior Improvement (RBI), over other safe learning methods are its stability in the presence of value estimation errors and the elimination of a policy search process. We demonstrate these advantages in the Taxi grid-world domain and in four games from the Atari learning environment.""","""The paper studies safer policy improvement based on non-expert demonstrations. The paper contains some interesting ideas, and is supported by reasonable empirical evidence. Overall, the work has a good potential. The author response was also helpful. That said, after considering the paper and rebuttal, the reviewers were not convinced the paper is ready for publication, as the significance of this work is limited by a rather strong assumption (see reviews for details). Furthermore, the presentation of the paper also requires some work to improve (see reviews for detailed comments)."""
680,"""Stochastic Gradient Descent Learns State Equations with Nonlinear Activations""","['recurrent neural network', 'state equation', 'gradient descent', 'sample complexity']","""We study discrete time dynamical systems governed by the state equation pseudo-formula . Here A,B are weight matrices, is an activation function, and pseudo-formula is the input data. This relation is the backbone of recurrent neural networks (e.g. LSTMs) which have broad applications in sequential learning tasks. We utilize stochastic gradient descent to learn the weight matrices from a finite input/state trajectory pseudo-formula . We prove that SGD estimate linearly converges to the ground truth weights while using near-optimal sample size. Our results apply to increasing activations whose derivatives are bounded away from zero. The analysis is based on i) an SGD convergence result with nonlinear activations and ii) careful statistical characterization of the state vector. Numerical experiments verify the fast convergence of SGD on ReLU and leaky ReLU in consistence with our theory.""","""This paper shows convergence of stochastic gradient descent for the problem of learning weight matrices for a linear dynamical system with non-linear activation. Reviewers agree that the problem considered is both interesting and challenging. However the paper makes many simplifying assumptions - 1) both input and hidden state are observed, a very non standard assumption, 2) analysis requires increasing activation functions, cannot handle ReLU functions. I agree with R2 and think these assumptions make the results significantly weaker. R1 and R3 are more optimistic, but authors response does not give an insight into how one might extend this analysis to the setting where hidden state is not observed. Relaxing these assumptions will make the paper more interesting. """
681,"""Accidental exploration through value predictors""","['reinforcement learning', 'value predictors', 'exploration']","""Infinite length of trajectories is an almost universal assumption in the theoretical foundations of reinforcement learning. In practice learning occurs on finite trajectories. In this paper we examine a specific result of this disparity, namely a strong bias of the time-bounded Every-visit Monte Carlo value estimator. This manifests as a vastly different learning dynamic for algorithms that use value predictors, including encouraging or discouraging exploration. We investigate these claims theoretically for a one dimensional random walk, and empirically on a number of simple environments. We use GAE as an algorithm involving a value predictor and evolution strategies as a reference point.""","""The paper studies the mismatch between value estimation in RL from finite vs. infinite trajectories. This is an interesting problem, but the reviewers raised concerns regarding (1) the consistency and coherence of the story (2) the significance of theoretical analysis and (3) significance of the results. I appreciate that the authors made significant changes to the paper to address the comments. However, given the extent of changes, I think another review cycle is needed to check the details of the paper again."""
682,"""microGAN: Promoting Variety through Microbatch Discrimination""","['adversarial training', 'gans']","""We propose to tackle the mode collapse problem in generative adversarial networks (GANs) by using multiple discriminators and assigning a different portion of each minibatch, called microbatch, to each discriminator. We gradually change each discriminator's task from distinguishing between real and fake samples to discriminating samples coming from inside or outside its assigned microbatch by using a diversity parameter pseudo-formula . The generator is then forced to promote variety in each minibatch to make the microbatch discrimination harder to achieve by each discriminator. Thus, all models in our framework benefit from having variety in the generated set to reduce their respective losses. We show evidence that our solution promotes sample diversity since early training stages on multiple datasets.""","""The paper proposes an approach to remedying mode collapse problem in GANs. This approach relies on using multiple discriminators and assigning a different portion of each minibatch to each discriminator. + preventing mode collapse in GAN training is an important problem - the exact motivation for the proposed techniques is not fully fleshed out - the evaluation and baselines used are lacking"""
683,"""Measuring Compositionality in Representation Learning""","['compositionality', 'representation learning', 'evaluation']","""Many machine learning algorithms represent input data with vector embeddings or discrete codes. When inputs exhibit compositional structure (e.g. objects built from parts or procedures from subroutines), it is natural to ask whether this compositional structure is reflected in the the inputs learned representations. While the assessment of compositionality in languages has received significant attention in linguistics and adjacent fields, the machine learning literature lacks general-purpose tools for producing graded measurements of compositional structure in more general (e.g. vector-valued) representation spaces. We describe a procedure for evaluating compositionality by measuring how well the true representation-producing model can be approximated by a model that explicitly composes a collection of inferred representational primitives. We use the procedure to provide formal and empirical characterizations of compositional structure in a variety of settings, exploring the relationship between compositionality and learning dynamics, human judgments, representational similarity, and generalization.""","""This paper presents a method for measuring the degree to which some representation for a composed object effectively represents the pieces from which it is composed. All three authors found this to be an important topic for study, and found the paper to be a limited but original and important step toward studying this topic. However, two reviewers expressed serious concerns about clarity, and were not fully satisfied with the revisions made so far. I'm recommending acceptance, but I ask the authors to further revise the paper (especially the introduction) to make sure it includes a blunt and straightforward presentation of the problem under study and the way TRE addresses it. I'm also somewhat concerned at R2's mention of a potential confound in one experiment. The paper has been updated with what appears to be a fix, though, and R2 has not yet responded, so I'm presuming that this issue has been resolved. I also ask the authors to release code shortly upon de-anonymization, as promised."""
684,"""AIM: Adversarial Inference by Matching Priors and Conditionals""","['Generative adversarial network', 'inference', 'generative model']","""Effective inference for a generative adversarial model remains an important and challenging problem. We propose a novel approach, Adversarial Inference by Matching priors and conditionals (AIM), which explicitly matches prior and conditional distributions in both data and code spaces, and puts a direct constraint on the dependency structure of the generative model. We derive an equivalent form of the prior and conditional matching objective that can be optimized efficiently without any parametric assumption on the data. We validate the effectiveness of AIM on the MNIST, CIFAR-10, and CelebA datasets by conducting quantitative and qualitative evaluations. Results demonstrate that AIM significantly improves both reconstruction and generation as compared to other adversarial inference models.""","""The paper proposes a method that aims to combine the strenghts of VAEs and GANs. The paper establishes an interesting bridge between GANs and VAEs. The experimental results are encouraging, even though only relatively small datasets were used. It is encouraging that the method results in better reconstructions then ALI, a related method. Some reviewers think that the paper contains limited novelty compared to the wealth of recent work on this topic (e.g. ALI/BiGAN). The paper's contribution is seen as incremental; e.g. the training is very similar to InfoGAN. Also, the claims of better sample quality over ALI seem insufficiently supported by the data."""
685,"""ADAPTIVE NETWORK SPARSIFICATION VIA DEPENDENT VARIATIONAL BETA-BERNOULLI DROPOUT""","['Bayesian deep learning', 'network pruning']","""While variational dropout approaches have been shown to be effective for network sparsification, they are still suboptimal in the sense that they set the dropout rate for each neuron without consideration of the input data. With such input-independent dropout, each neuron is evolved to be generic across inputs, which makes it difficult to sparsify networks without accuracy loss. To overcome this limitation, we propose adaptive variational dropout whose probabilities are drawn from sparsity-inducing beta-Bernoulli prior. It allows each neuron to be evolved either to be generic or specific for certain inputs, or dropped altogether. Such input-adaptive sparsity-inducing dropout allows the resulting network to tolerate larger degree of sparsity without losing its expressive power by removing redundancies among features. We validate our dependent variational beta-Bernoulli dropout on multiple public datasets, on which it obtains significantly more compact networks than baseline methods, with consistent accuracy improvements over the base networks.""","""The paper proposes Variational Beta-Bernoulli Dropout,, a Bayesian method for sparsifying neural networks. The method adopts a spike-and-slab pior over parameter of the network. The paper proposes Beta hyperpriors over the network, motivated by the Indian Buffet Process, and propose a method for input-conditional priors. The paper is well-written and the material is communicated clearly. The topic is also of interest to the community and might have important implications down the road. The authors, however, failed to convince the reviewers that the paper is ready for publication at ICLR. The proposed method is very similar to earlier work. The reviewers think that the paper is not ready for publication."""
686,"""Cautious Deep Learning""","['Deep Learning', 'Classification', 'Prediction', 'Cautious Methods']","""Most classifiers operate by selecting the maximum of an estimate of the conditional distribution pseudo-formula where pseudo-formula stands for the features of the instance to be classified and pseudo-formula denotes its label. This often results in a hubristic bias: overconfidence in the assignment of a definite label. Usually, the observations are concentrated on a small volume but the classifier provides definite predictions for the entire space. We propose constructing conformal prediction sets which contain a set of labels rather than a single label. These conformal prediction sets contain the true label with probability pseudo-formula . Our construction is based on pseudo-formula rather than pseudo-formula which results in a classifier that is very cautious: it outputs the null set --- meaning ``I don't know'' --- when the object does not resemble the training examples. An important property of our approach is that classes can be added or removed without having to retrain the classifier. We demonstrate the performance on the ImageNet ILSVRC dataset and the CelebA and IMDB-Wiki facial datasets using high dimensional features obtained from state of the art convolutional neural networks. ""","""The paper presents a conformal prediction approach to supervised classification, with the goal of reducing the overconfidence of standard soft-max learning techniques. The proposal is based on previously published methods, which are extended for use with deep learning predictors. Empirical evaluation suggests the proposal results in competitive performance. This work seems to be timely, and the topic is of interest to the community. The reviewers and AC opinions were mixed, with reviewers either being unconvinced about the novelty of the proposed work or expressing issues about the strength of the empirical evidence supporting the claims. Additional experiments would significantly strengthen this submission."""
687,"""ALISTA: Analytic Weights Are As Good As Learned Weights in LISTA""","['sparse recovery', 'neural networks']","""Deep neural networks based on unfolding an iterative algorithm, for example, LISTA (learned iterative shrinkage thresholding algorithm), have been an empirical success for sparse signal recovery. The weights of these neural networks are currently determined by data-driven black-box training. In this work, we propose Analytic LISTA (ALISTA), where the weight matrix in LISTA is computed as the solution to a data-free optimization problem, leaving only the stepsize and threshold parameters to data-driven learning. This signicantly simplies the training. Specically, the data-free optimization problem is based on coherence minimization. We show our ALISTA retains the optimal linear convergence proved in (Chen et al., 2018) and has a performance comparable to LISTA. Furthermore, we extend ALISTA to convolutional linear operators, again determined in a data-free manner. We also propose a feed-forward framework that combines the data-free optimization and ALISTA networks from end to end, one that can be jointly trained to gain robustness to small perturbations in the encoding model.""","""This is a well executed paper that makes clear contributions to the understanding of unrolled iterative optimization and soft thresholding for sparse signal recovery with neural networks."""
688,"""Synthetic Datasets for Neural Program Synthesis""",[],"""The goal of program synthesis is to automatically generate programs in a particular language from corresponding specifications, e.g. input-output behavior. Many current approaches achieve impressive results after training on randomly generated I/O examples in limited domain-specific languages (DSLs), as with string transformations in RobustFill. However, we empirically discover that applying test input generation techniques for languages with control flow and rich input space causes deep networks to generalize poorly to certain data distributions; to correct this, we propose a new methodology for controlling and evaluating the bias of synthetic data distributions over both programs and specifications. We demonstrate, using the Karel DSL and a small Calculator DSL, that training deep networks on these distributions leads to improved cross-distribution generalization performance. ""","""This paper analyzes existing approaches to program induction from I/O pairs, and demonstrates that naively generating I/O pairs results in a non-uniform sampling of salient variables, leading to poor performance. The paper convincingly shows, via strong evaluation, that uniform sampling of these variables can much result in much better models, both for explicit DSL and implicit, neural models. The reviewers feel the observation is an important one, and the paper does a good job providing sufficiently convincing evidence for it. The reviewers and AC note the following potential weaknesses: (1) the paper does not propose a new model, but instead a different data generation strategy, somewhat limiting the novelty, (2) Salient variables that need to be uniformly sampled are still user specified, (3) there were a number of notation and clarity issues that make it difficult to understand the details of the approach, and finally, (4) there are concerns with the use of rejection sampling. The authors provided major revisions that address the clarity issues, including an addition of new proofs, cleaner notation, and removal of unnecessary text. The authors also included additional results, such as KL divergence evaluation to show how uniform the distribution is. The authors also described the need for rejection sampling, especially for Karel dataset, and clarified why the Calculator domain, even though is not ""program synthesis"", still faces similar challenges. The reviewers agreed that not having a new model is not a chief concern, and that using rejection sampling is a reasonable first step, with more efficient techniques left for others for future work. Overall, the reviewers agreed that the paper should be accepted. As reviewer 1 said it best, this paper ""is a timely contribution and I think it is important for future program synthesis papers to take the results and message here to heart""."""
689,"""The Expressive Power of Gated Recurrent Units as a Continuous Dynamical System""","['Gated Recurrent Units', 'Recurrent Neural Network', 'Time Series Predictions', 'interpretable', 'Nonlinear Dynamics', 'Dynamical Systems']","""Gated recurrent units (GRUs) were inspired by the common gated recurrent unit, long short-term memory (LSTM), as a means of capturing temporal structure with less complex memory unit architecture. Despite their incredible success in tasks such as natural and artificial language processing, speech, video, and polyphonic music, very little is understood about the specific dynamic features representable in a GRU network. As a result, it is difficult to know a priori how successful a GRU-RNN will perform on a given data set. In this paper, we develop a new theoretical framework to analyze one and two dimensional GRUs as a continuous dynamical system, and classify the dynamical features obtainable with such system. We found rich repertoire that includes stable limit cycles over time (nonlinear oscillations), multi-stable state transitions with various topologies, and homoclinic orbits. In addition, we show that any finite dimensional GRU cannot precisely replicate the dynamics of a ring attractor, or more generally, any continuous attractor, and is limited to finitely many isolated fixed points in theory. These findings were then experimentally verified in two dimensions by means of time series prediction.""","""The paper analyses GRUs using dynamic systems theory. The paper is well-written and the theory seems to be solid. But there is agreement amongst the reviewers that the application of the method might not scale well beyond rather simple 1- or 2-D GRUs (i.e., with one or two GRUs). This limitation, which is an increasingly serious problem in machine-learning papers, should be solved before the paper should be published. A very recent extension of the simulations to 16 GRUs improves this, but a rigorous analysis of higher-dimensional systems is pending and poses a considerable block for acceptance."""
690,"""L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data""","['Model Interpretation', 'Feature Selection']","""Instancewise feature scoring is a method for model interpretation, which yields, for each test instance, a vector of importance scores associated with features. Methods based on the Shapley score have been proposed as a fair way of computing feature attributions, but incur an exponential complexity in the number of features. This combinatorial explosion arises from the definition of Shapley value and prevents these methods from being scalable to large data sets and complex models. We focus on settings in which the data have a graph structure, and the contribution of features to the target variable is well-approximated by a graph-structured factorization. In such settings, we develop two algorithms with linear complexity for instancewise feature importance scoring on black-box models. We establish the relationship of our methods to the Shapley value and a closely related concept known as the Myerson value from cooperative game theory. We demonstrate on both language and image data that our algorithms compare favorably with other methods using both quantitative metrics and human evaluation.""","""The paper presents two new methods for model-agnostic interpretation of instance-wise feature importance. Pros: Unlike previous approaches based on the Shapley value, which had an exponential complexity in the number of features, the proposed methods have a linear-complexity when the data have a graph structure, which allows approximation based on graph-structured factorization. The proposed methods present solid technical novelty to study the important challenge of instance-wise, model-agnostic, linear-complexity interpretation of features. Cons: All reviewers wanted to see more extensive experimental results. Authors responded with most experiments requested. One issue raised by R3 was the need for comparing the proposed model-agnostic methods to existing model-specific methods. The proposed linear-complexity algorithm relies on the markov assumption, which some reviewers commented to be a potentially invalid assumption to make, but this does not seem to be a deal breaker since it is a relatively common assumption to make when deriving a polynomial-complexity approximation algorithm. Overall, the rebuttal addressed the reviewers' concerns well enough, leading to increased scores. Verdict: Accept. Solid technical novelty with convincing empirical results."""
691,"""Mimicking actions is a good strategy for beginners: Fast Reinforcement Learning with Expert Action Sequences""","['Reinforcement Learning', 'Imitation Learning', 'Atari', 'A3C', 'GA3C']","""Imitation Learning is the task of mimicking the behavior of an expert player in a Reinforcement Learning(RL) Environment to enhance the training of a fresh agent (called novice) beginning from scratch. Most of the Reinforcement Learning environments are stochastic in nature, i.e., the state sequences that an agent may encounter usually follow a Markov Decision Process (MDP). This makes the task of mimicking difficult as it is very unlikely that a new agent may encounter same or similar state sequences as an expert. Prior research in Imitation Learning proposes various ways to learn a mapping between the states encountered and the respective actions taken by the expert while mostly being agnostic to the order in which these were performed. Most of these methods need considerable number of states-action pairs to achieve good results. We propose a simple alternative to Imitation Learning by appending the novices action space with the frequent short action sequences that the expert has taken. This simple modification, surprisingly improves the exploration and significantly outperforms alternative approaches like Dataset Aggregation. We experiment with several popular Atari games and show significant and consistent growth in the score that the new agents achieve using just a few expert action sequences.""","""The paper proposes an interesting idea for more effective imitation learning. The idea is to include short actions sequences as labels (in addition to the basic actions) in imitation learning. Results on a few Atari games demonstrate the potential of this approach. Reviewers generally like the idea, think it is simple, and are encouraged by its empirical support. That said, the work still appears somewhat preliminary in the current stage: (1) some reviewer is still in doubt about the chosen baseline; (2) empirical evidence is all in the similar set of Atari games --- how broadly is this approach applicable?"""
692,"""TarMAC: Targeted Multi-Agent Communication""",[],"""We explore the collaborative multi-agent setting where a team of deep reinforcement learning agents attempt to solve a shared task in partially observable environments. In this scenario, learning an effective communication protocol is key. We propose a communication protocol that allows for targeted communication, where agents learn \emph{what} messages to send and \emph{who} to send them to. Additionally, we introduce a multi-stage communication approach where the agents co-ordinate via several rounds of communication before taking an action in the environment. We evaluate our approach on several cooperative multi-agent tasks, of varying difficulties with varying number of agents, in a variety of environments ranging from 2D grid layouts of shapes and simulated traffic junctions to complex 3D indoor environments. We demonstrate the benefits of targeted as well as multi-stage communication. Moreover, we show that the targeted communication strategies learned by the agents are quite interpretable and intuitive.""","""The reviewers raised a number of concerns including the lack of clarity of various parts of the paper, lack of explanation, incremental novelty, and insufficiently demonstrated significance of the proposed. The authors rebuttal addressed some of the reviewers concerns but not fully. Overall, I believe that the paper presents some interesting extensions for multi-agent communication but in its current form the paper lacks explanations, comparisons and discussions. Hence, I cannot recommend this paper for presentation at ICLR."""
693,"""Learning Gibbs-regularized GANs with variational discriminator reparameterization""","['deep generative models', 'graphical models', 'trajectory forecasting', 'GANs', 'density estimation', 'structured prediction']",""" We propose a novel approach to regularizing generative adversarial networks (GANs) leveraging learned {\em structured Gibbs distributions}. Our method consists of reparameterizing the discriminator to be an explicit function of two densities: the generator PDF pseudo-formula and a structured Gibbs distribution pseudo-formula . Leveraging recent work on invertible pushforward density estimators, this reparameterization is made possible by assuming the generator is invertible, which enables the analytic evaluation of the generator PDF pseudo-formula . We further propose optimizing the Jeffrey divergence, which balances mode coverage with sample quality. The combination of this loss and reparameterization allows us to effectively regularize the generator by imposing structure from domain knowledge on pseudo-formula , as in classical graphical models. Applying our method to a vehicle trajectory forecasting task, we observe that we are able to obtain quantitatively superior mode coverage as well as better-quality samples compared to traditional methods.""","""The paper proposes to define the GAN discriminator as an explicit function of a invertible generator density and a structured Gibbs distribution to tackle the problems of spurious modes and mode collapse. The resulting model is similar to R2P2, i.e. it can be seen as adding an adversarial component to R2P2, and shows competitive (but no better) performance. Reviewers agree, that these limits the novelty of the contribution, and that the paper would be improved by a more extensive empirical evaluation. """
694,"""Learning to Make Analogies by Contrasting Abstract Relational Structure""","['cognitive science', 'analogy', 'psychology', 'cognitive theory', 'cognition', 'abstraction', 'generalization']","""Analogical reasoning has been a principal focus of various waves of AI research. Analogy is particularly challenging for machines because it requires relational structures to be represented such that they can be flexibly applied across diverse domains of experience. Here, we study how analogical reasoning can be induced in neural networks that learn to perceive and reason about raw visual data. We find that the critical factor for inducing such a capacity is not an elaborate architecture, but rather, careful attention to the choice of data and the manner in which it is presented to the model. The most robust capacity for analogical reasoning is induced when networks learn analogies by contrasting abstract relational structures in their input domains, a training method that uses only the input data to force models to learn about important abstract features. Using this technique we demonstrate capacities for complex, visual and symbolic analogy making and generalisation in even the simplest neural network architectures.""",""" pros: - The paper is well-written and includes a lot of interesting connections to cog sci (though see specific clarity concerns) - The tasks considered (visual and symbolic) provide a nice opportunity to study analogy making in different settings. cons: - There was some concerns about baselines and novelty that I think the authors have largely addressed in revision This is an intriguing paper and an exciting direction and I think it merits acceptance."""
695,"""VECTORIZATION METHODS IN RECOMMENDER SYSTEM""",[],"""The most used recommendation method is collaborative filtering, and the key part of collaborative filtering is to compute the similarity. The similarity based on co-occurrence of similar event is easy to implement and can be applied to almost all the situation. So when the word2vec model reach the state-of-art at a lower computation cost in NLP. An correspond model in recommender system item2vec is proposed and reach state-of-art in recommender system. It is easy to see that the position of user and item is interchangeable when their count size gap is not too much, we proposed a user2vec model and show its performance. The similarity based on co-occurrence information suffers from cold start, we proposed a content based similarity model based on doc2vec which is another technology in NLP.""","""The reviewers are unanonymous in their assessment that the paper is not ICLR quality in its current form."""
696,"""Optimization on Multiple Manifolds""","['Optimization', 'Multiple constraints', 'Manifold']","""Optimization on manifold has been widely used in machine learning, to handle optimization problems with constraint. Most previous works focus on the case with a single manifold. However, in practice it is quite common that the optimization problem involves more than one constraints, (each constraint corresponding to one manifold). It is not clear in general how to optimize on multiple manifolds effectively and provably especially when the intersection of multiple manifolds is not a manifold or cannot be easily calculated. We propose a unified algorithm framework to handle the optimization on multiple manifolds. Specifically, we integrate information from multiple manifolds and move along an ensemble direction by viewing the information from each manifold as a drift and adding them together. We prove the convergence properties of the proposed algorithms. We also apply the algorithms into training neural network with batch normalization layers and achieve preferable empirical results.""","""The paper describes a constrained optimization strategy for optimizing on an intersection of two manifolds. Unfortunately, the paper suffers from generally weak presentation quality, with the technical exposition seriously criticized by two out of the three reviewers. (The single positive review is too short and devoid of content to be taken seriously. Even there, concerns are expressed.) This paper requires substantial improvement before it could be considered for publication."""
697,"""Quasi-hyperbolic momentum and Adam for deep learning""","['sgd', 'momentum', 'nesterov', 'adam', 'qhm', 'qhadam', 'optimization']","""Momentum-based acceleration of stochastic gradient descent (SGD) is widely used in deep learning. We propose the quasi-hyperbolic momentum algorithm (QHM) as an extremely simple alteration of momentum SGD, averaging a plain SGD step with a momentum step. We describe numerous connections to and identities with other algorithms, and we characterize the set of two-state optimization algorithms that QHM can recover. Finally, we propose a QH variant of Adam called QHAdam, and we empirically demonstrate that our algorithms lead to significantly improved training in a variety of settings, including a new state-of-the-art result on WMT16 EN-DE. We hope that these empirical results, combined with the conceptual and practical simplicity of QHM and QHAdam, will spur interest from both practitioners and researchers. Code is immediately available.""","""This paper presents quasi-hyperbolic momentum, a generalization of Nesterov Accelerated Gradient. The method can be seen as adding an additional hyperparameter to NAG corresponding to the weighting of the direct gradient term in the update. The contribution is pretty simple, but the paper has good discussion of the relationships with other momentum methods, careful theoretical analysis, and fairly strong experimental results. All the reviewers believe this is a strong paper and should be accepted, and I concur. """
698,"""Learning Corresponded Rationales for Text Matching""","['interpretability', 'rationalization', 'text matching', 'dependent selection']","""The ability to predict matches between two sources of text has a number of applications including natural language inference (NLI) and question answering (QA). While flexible neural models have become effective tools in solving these tasks, they are rarely transparent in terms of the mechanism that mediates the prediction. In this paper, we propose a self-explaining architecture where the model is forced to highlight, in a dependent manner, how spans of one side of the input match corresponding segments of the other side in order to arrive at the overall decision. The text spans are regularized to be coherent and concise, and their correspondence is captured explicitly. The text spans -- rationales -- are learned entirely as latent mechanisms, guided only by the distal supervision from the end-to-end task. We evaluate our model on both NLI and QA using three publicly available datasets. Experimental results demonstrate quantitatively and qualitatively that our method delivers interpretable justification of the prediction without sacrificing state-of-the-art performance. Our code and data split will be publicly available. ""","""This paper attempts at modeling text matching and also generating rationales. The motivation of the paper is good. However there is some shortcomings of the paper, e.g. there is very little comparison with prior work, no human evaluation at scale and also it seems that several prior models that use attention mechanism would generate similar rationales. No characterization of the last aspect has been made here. Hence, addressing these issues could make the paper better for future venues. There is relative consensus between the reviewers that the paper could improve if the reviewers' concerns are addressed when it is submitted to future venues."""
699,"""Lyapunov-based Safe Policy Optimization""","['Reinforcement Learning', 'Safe Learning', 'Lyapunov Functions', 'Constrained Markov Decision Problems']","""In many reinforcement learning applications, it is crucial that the agent interacts with the environment only through safe policies, i.e.,~policies that do not take the agent to certain undesirable situations. These problems are often formulated as a constrained Markov decision process (CMDP) in which the agent's goal is to optimize its main objective while not violating a number of safety constraints. In this paper, we propose safe policy optimization algorithms that are based on the Lyapunov approach to CMDPs, an approach that has well-established theoretical guarantees in control engineering. We first show how to generate a set of state-dependent Lyapunov constraints from the original CMDP safety constraints. We then propose safe policy gradient algorithms that train a neural network policy using DDPG or PPO, while guaranteeing near-constraint satisfaction at every policy update by projecting either the policy parameter or the action onto the set of feasible solutions induced by the linearized Lyapunov constraints. Unlike the existing (safe) constrained PG algorithms, ours are more data efficient as they are able to utilize both on-policy and off-policy data. Furthermore, the action-projection version of our algorithms often leads to less conservative policy updates and allows for natural integration into an end-to-end PG training pipeline. We evaluate our algorithms and compare them with CPO and the Lagrangian method on several high-dimensional continuous state and action simulated robot locomotion tasks, in which the agent must satisfy certain safety constraints while minimizing its expected cumulative cost. ""","""This is an interesting direction but multiple reviewers had concerns about the amount of novelty in the current work, and given the strong pool of other papers, this didn't quite reach the threshold. """
700,"""Adversarial Information Factorization""","['disentangled representations', 'factored representations', 'generative adversarial networks', 'variational auto encoders', 'generative models']","""We propose a novel generative model architecture designed to learn representations for images that factor out a single attribute from the rest of the representation. A single object may have many attributes which when altered do not change the identity of the object itself. Consider the human face; the identity of a particular person is independent of whether or not they happen to be wearing glasses. The attribute of wearing glasses can be changed without changing the identity of the person. However, the ability to manipulate and alter image attributes without altering the object identity is not a trivial task. Here, we are interested in learning a representation of the image that separates the identity of an object (such as a human face) from an attribute (such as 'wearing glasses'). We demonstrate the success of our factorization approach by using the learned representation to synthesize the same face with and without a chosen attribute. We refer to this specific synthesis process as image attribute manipulation. We further demonstrate that our model achieves competitive scores, with state of the art, on a facial attribute classification task.""","""The paper proposes a supervised adversarial method for disentangling the latent space of a VAE into two groups: latents z which are independent of the given attribute y, and \hat{y} which contains information about y. Since the encoder also predicts \hat{y} it can be used for classification and the paper shows competitive results on this task, apart from the attribute manipulation task. Reviewers had raised points about model complexity and connections to prior works which the authors have addressed and the paper is on the borderline based on the scores. Though none of the reviewers explicitly pointed out the similarity of the paper with Fader networks (Lample et al., 2017), the adversarial setup for getting attribute invariant 'z' is exactly same as in Fader networks, as also pointed out in an anonymous comment. The only difference is that encoder in the current paper also predicts the attribute itself (\hat{y}), which is not the case in Fader n/w, and hence the encoder can be used as a classifier as well (authors have also mentioned and discussed this difference in their response). However, the core idea of the paper as outlined in the title of the paper, ie, using adversarial loss for information factorization, is very similar to this earlier work, which diminishes the originality of the work. With the borderline review scores, the paper can go in either of the half-spaces (accept/reject) but I am hesitant to recommend an ""accept"" due to limited originality of the approach. However, if there is space in the program, the paper can be accepted. """
701,"""Massively Parallel Hyperparameter Tuning""","['hyperparameter optimization', 'automl']","""Modern learning models are characterized by large hyperparameter spaces. In order to adequately explore these large spaces, we must evaluate a large number of configurations, typically orders of magnitude more configurations than available parallel workers. Given the growing costs of model training, we would ideally like to perform this search in roughly the same wall-clock time needed to train a single model. In this work, we tackle this challenge by introducing ASHA, a simple and robust hyperparameter tuning algorithm with solid theoretical underpinnings that exploits parallelism and aggressive early-stopping. Our extensive empirical results show that ASHA outperforms state-of-the-art hyperparameter tuning methods; scales linearly with the number of workers in distributed settings; converges to a high quality configuration in half the time taken by Vizier, Google's internal hyperparameter tuning service) in an experiment with 500 workers; and beats the published result for a near state-of-the-art LSTM architecture in under pseudo-formula the time to train a single model.""","""The paper proposes and evaluates an asynchronous hyperparameter optimization algorithm. Strengths: The experiments are certainly thorough, and I appreciated the discussion of the algorithm and side-experiments demonstrating its operation in different settings. Overall the paper is pretty clear. It's a good thing when a proposed method is a simple variant of an existing method. Weaknesses: The first page could have been half the length, and it's not clear why we should care about the stated goal of this work. Isn't the real goal just to get good test performance in a small amount of time? The title is also a bit obnoxious and land-grabby - it could have been used for almost any of the comparison methods. The proposed method is a minor change to SHA. The proposed change is kind of obvious, and the resulting method does have a number of hyper-hyperparameters. Consensus: Ultimately I agree with the reviewers that is just below the bar of acceptance. This does seem like a valid contribution to the hyperparameter-tuning literature, but more of an engineering contribution than a research contribution. It's also getting a little bit away from the subject of machine learning, and might be more appropriate for say, SysML."""
702,"""Rating Continuous Actions in Spatial Multi-Agent Problems""",[],"""We study credit assignment problems in spatial multi-agent environments where agents pursue a joint objective. On the example of soccer, we rate the movements of individual players with respect to their potential for staging a successful attack. We propose a purely data-driven approach to simultaneously learn a model of agent movements as well as their ratings via an agent-centric deep reinforcement learning framework. Our model allows for efficient learning and sampling of ratings in the continuous action space. We empirically observe on historic soccer data that the model accurately rates agent movements w.r.t. their relative contribution to the collective goal.""","""The reviewers raised a number of major concerns including the incremental novelty of the proposed (if any), a poor readability of the presented materials, and, most importantly, insufficient and unconvincing experimental evaluation presented. The authors did not provide any rebuttal. Hence, I cannot suggest this paper for presentation at ICLR."""
703,"""DISTRIBUTIONAL CONCAVITY REGULARIZATION FOR GANS""","['Generative Adversarial Networks', 'regularization', 'optimal transport', 'functional gradient', 'convex analysis']","""We propose Distributional Concavity (DC) regularization for Generative Adversarial Networks (GANs), a functional gradient-based method that promotes the entropy of the generator distribution and works against mode collapse. Our DC regularization is an easy-to-implement method that can be used in combination with the current state of the art methods like Spectral Normalization and Wasserstein GAN with gradient penalty to further improve the performance. We will not only show that our DC regularization can achieve highly competitive results on ILSVRC2012 and CIFAR datasets in terms of Inception score and Fr\'echet inception distance, but also provide a mathematical guarantee that our method can always increase the entropy of the generator distribution. We will also show an intimate theoretical connection between our method and the theory of optimal transport.""","""This paper proposes distributional concavity regularization for GANs which encourages producing generator distributions with higher entropy. The reviewers found the contribution interesting for the ICLR community. R3 initially found the paper lacked clarity, but the authors took the feedback in consideration and made significant improvements in their revision. The reviewers all agreed that the updated paper should be accepted."""
704,"""Evolutionary-Neural Hybrid Agents for Architecture Search""","['Evolutionary', 'Architecture Search', 'NAS']","""Neural Architecture Search has recently shown potential to automate the design of Neural Networks. The use of Neural Network agents trained with Reinforcement Learning can offer the possibility to learn complex patterns, as well as the ability to explore a vast and compositional search space. On the other hand, evolutionary algorithms offer the greediness and sample efficiency needed for such an application, as each sample requires a considerable amount of resources. We propose a class of Evolutionary-Neural hybrid agents (Evo-NAS), that retain the best qualities of the two approaches. We show that the Evo-NAS agent can outperform both Neural and Evolutionary agents, both on a synthetic task, and on architecture search for a suite of text classification datasets.""","""Reviewers are in a consensus and recommended to reject after engaging with the authors. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit. """
705,"""Pushing the bounds of dropout""",[],"""We show that dropout training is best understood as performing MAP estimation concurrently for a family of conditional models whose objectives are themselves lower bounded by the original dropout objective. This discovery allows us to pick any model from this family after training, which leads to a substantial improvement on regularisation-heavy language modelling. The family includes models that compute a power mean over the sampled dropout masks, and their less stochastic subvariants with tighter and higher lower bounds than the fully stochastic dropout objective. We argue that since the deterministic subvariant's bound is equal to its objective, and the highest amongst these models, the predominant view of it as a good approximation to MC averaging is misleading. Rather, deterministic dropout is the best available approximation to the true objective.""","""The paper tried to introduce a new interpretation of dropout and come with improved algorithms. However, the reviewers were not convinced that the presented arguments were correct/novel, and they found the paper difficult to follow. The authors are encouraged to carefully revise their paper to address these concerns. """
706,"""Learning to Understand Goal Specifications by Modelling Reward""","['instruction following', 'reward modelling', 'language understanding']","""Recent work has shown that deep reinforcement-learning agents can learn to follow language-like instructions from infrequent environment rewards. However, this places on environment designers the onus of designing language-conditional reward functions which may not be easily or tractably implemented as the complexity of the environment and the language scales. To overcome this limitation, we present a framework within which instruction-conditional RL agents are trained using rewards obtained not from the environment, but from reward models which are jointly trained from expert examples. As reward models improve, they learn to accurately reward agents for completing tasks for environment configurations---and for instructions---not present amongst the expert data. This framework effectively separates the representation of what instructions require from how they can be executed. In a simple grid world, it enables an agent to learn a range of commands requiring interaction with blocks and understanding of spatial relations and underspecified abstract arrangements. We further show the method allows our agent to adapt to changes in the environment without requiring new expert examples.""","""Pros: - The paper is well-written and clear and presented with helpful illustrations and videos. - The training methodology seems sound (multiple random seeds etc.) - The results are encouraging. Cons: - There was some concern generally about how this work is positioned relative to related work and the completeness of the related work. However, the authors have made this clearer in their rebuttal. There was a considerable amount of discussion between the authors and all reviewers to pin down some unclear aspects of the paper. I believe in the end there was good convergence and I thank both the authors and reviewers for their persistence and dilligence in working through this. The final paper is much better I think and I recommend acceptance."""
707,"""Preconditioner on Matrix Lie Group for SGD""","['preconditioner', 'stochastic gradient descent', 'Newton method', 'Fisher information', 'natural gradient', 'Lie group']","""We study two types of preconditioners and preconditioned stochastic gradient descent (SGD) methods in a unified framework. We call the first one the Newton type due to its close relationship to the Newton method, and the second one the Fisher type as its preconditioner is closely related to the inverse of Fisher information matrix. Both preconditioners can be derived from one framework, and efficiently estimated on any matrix Lie groups designated by the user using natural or relative gradient descent minimizing certain preconditioner estimation criteria. Many existing preconditioners and methods, e.g., RMSProp, Adam, KFAC, equilibrated SGD, batch normalization, etc., are special cases of or closely related to either the Newton type or the Fisher type ones. Experimental results on relatively large scale machine learning problems are reported for performance study.""","""The method presented here adapts an SGD preconditioner by minimizing particular cost functions which are minimized by the inverse Hessian or inverse Fisher matrix. These cost functions are minimized using natural (or relative) gradient on the Lie group, as previously introduced by Amari. This can be extended to learn a Kronecker-factored preconditioner similar to K-FAC, except that the preconditioner is constrained to be upper triangular, which allows the relative gradient to be computed using backsubstitution rather than inversion. Experiments show modest speedups compared to SGD on ImageNet and language modeling. There's a wide divergence in reviewer scores. We can disregard the extremely short review by R2. R1 and R3 each did very careful reviews (R3 even tried out the algorithm), but gave scores of 5 and 8. They agree on most of the particulars, but just emphasized different factors. Because of this, I took a careful look, and indeed I think the paper has significant strengths and weaknesses. The main strength is the novelty of the approach. Combining relative gradient with upper triangular preconditioners is clever, and allows for a K-FAC-like algorithm which avoids matrix inversion. I haven't seen anything similar, and this method seems potentially useful. R3 reports that (s)he tried out the algorithm and found it to work well. Contrary to R1, I think the paper does use Lie groups in a meaningful way. Unfortunately, the writing is below the standards of an ICLR paper. The title is misleading, since the method isn't learning a preconditioner ""on"" the Lie group. The abstract and introduction don't give a clear idea of what the paper is about. While some motivation for the algorithms is given, it's expressed very tersely, and in a way that will only make sense to someone who knows the mathematical toolbox well enough to appreciate why the algorithm makes sense. As the reviewers point out, important details (such as hyperparameter tuning schemes) are left out of the experiments section. The experiments are also somewhat problematic, as pointed out by R1. The paper compares only to SGD and Adam, even though many other second-order optimizers have been proposed (and often with code available). It's unclear how well the baselines were tuned, and at the end of the day, the performance gain is rather limited. The experiments measure only iterations, not wall clock time. On the plus side, the experiments include ImageNet, which is ambitious by the standards of an algorithmic paper, and as mentioned above, R3 got good results from the method. On the whole, I would favor acceptance because of the novelty and potential usefulness of the approach. This would be a pretty solid submission of the writing were improved. (While the authors feel constrained by the 8 page limit, I'd recommend going beyond this for clarity.) However, I emphasize that it is very important to clean up the writing. """
708,"""Learning Latent Semantic Representation from Pre-defined Generative Model""","['Latent space', 'Generative adversarial network', 'variational autoencoder', 'conditioned generation']","""Learning representations of data is an important issue in machine learning. Though GAN has led to significant improvements in the data representations, it still has several problems such as unstable training, hidden manifold of data, and huge computational overhead. GAN tends to produce the data simply without any information about the manifold of the data, which hinders from controlling desired features to generate. Moreover, most of GANs have a large size of manifold, resulting in poor scalability. In this paper, we propose a novel GAN to control the latent semantic representation, called LSC-GAN, which allows us to produce desired data to generate and learns a representation of the data efficiently. Unlike the conventional GAN models with hidden distribution of latent space, we define the distributions explicitly in advance that are trained to generate the data based on the corresponding features by inputting the latent variables that follow the distribution. As the larger scale of latent space caused by deploying various distributions in one latent space makes training unstable while maintaining the dimension of latent space, we need to separate the process of defining the distributions explicitly and operation of generation. We prove that a VAE is proper for the former and modify a loss function of VAE to map the data into the pre-defined latent space so as to locate the reconstructed data as close to the input data according to its characteristics. Moreover, we add the KL divergence to the loss function of LSC-GAN to include this process. The decoder of VAE, which generates the data with the corresponding features from the pre-defined latent space, is used as the generator of the LSC-GAN. Several experiments on the CelebA dataset are conducted to verify the usefulness of the proposed method to generate desired data stably and efficiently, achieving a high compression ratio that can hold about 24 pixels of information in each dimension of latent space. Besides, our model learns the reverse of features such as not laughing (rather frowning) only with data of ordinary and smiling facial expression.""","""Reviewers have expressed concerns about clarity/writing of the paper and technical novelty, which the authors haven't responded to. The paper is not suitable for publication at ICLR."""
709,"""Subgradient Descent Learns Orthogonal Dictionaries""","['Dictionary learning', 'Sparse coding', 'Non-convex optimization', 'Theory']","""This paper concerns dictionary learning, i.e., sparse coding, a fundamental representation learning problem. We show that a subgradient descent algorithm, with random initialization, can recover orthogonal dictionaries on a natural nonsmooth, nonconvex L1 minimization formulation of the problem, under mild statistical assumption on the data. This is in contrast to previous provable methods that require either expensive computation or delicate initialization schemes. Our analysis develops several tools for characterizing landscapes of nonsmooth functions, which might be of independent interest for provable training of deep networks with nonsmooth activations (e.g., ReLU), among other applications. Preliminary synthetic and real experiments corroborate our analysis and show that our algorithm works well empirically in recovering orthogonal dictionaries.""","""This paper studies non smooth and non convex optimization and provides a global analysis for orthogonal dictionary learning. The referees indicate that the analysis is highly nontrivial compared with existing work. The experiments fall a bit short and the relation to the loss landscape of neural networks could be described more clearly. The reviewers pointed out that the experiments section was too short. The revision included a few more experiments. The paper has a theoretical focus, and scores high ratings there. The confidence levels of the reviewers is relatively moderate, with only one confident reviewer. However, all five reviewers regard this paper positively, in particular the confident reviewer. """
710,"""Learning to Schedule Communication in Multi-agent Reinforcement Learning""","['Multi agent reinforcement learning', 'deep reinforcement learning', 'Communication']","""Many real-world reinforcement learning tasks require multiple agents to make sequential decisions under the agents interaction, where well-coordinated actions among the agents are crucial to achieve the target goal better at these tasks. One way to accelerate the coordination effect is to enable multiple agents to communicate with each other in a distributed manner and behave as a group. In this paper, we study a practical scenario when (i) the communication bandwidth is limited and (ii) the agents share the communication medium so that only a restricted number of agents are able to simultaneously use the medium, as in the state-of-the-art wireless networking standards. This calls for a certain form of communication scheduling. In that regard, we propose a multi-agent deep reinforcement learning framework, called SchedNet, in which agents learn how to schedule themselves, how to encode the messages, and how to select actions based on received messages. SchedNet is capable of deciding which agents should be entitled to broadcasting their (encoded) messages, by learning the importance of each agents partially observed information. We evaluate SchedNet against multiple baselines under two different applications, namely, cooperative communication and navigation, and predator-prey. Our experiments show a non-negligible performance gap between SchedNet and other mechanisms such as the ones without communication and with vanilla scheduling methods, e.g., round robin, ranging from 32% to 43%.""","""The authors present a learnt scheduling mechanism for managing communications in bandwidth-constrained, contentious multi-agent RL domains. This is well-positioned in the rapidly advancing field of MARL and the contribution of the paper is both novel, interesting, and effective. The agents learn how to schedule themselves, how to encode messages, and how to select actions. The approach is evaluated against several other methods and achieves a good performance increase. The reviewers had concerns regarding the difficulty of evaluating the overall performance and also about how it would fare in more real-world scenarios, but all agree that this paper should be accepted."""
711,"""On the Relationship between Neural Machine Translation and Word Alignment""","['Neural Machine Translation', 'Word Alignment', 'Neural Network', 'Pointwise Mutual Information']","""Prior researches suggest that attentional neural machine translation (NMT) is able to capture word alignment by attention, however, to our surprise, it almost fails for NMT models with multiple attentional layers except for those with a single layer. This paper introduce two methods to induce word alignment from general neural machine translation models. Experiments verify that both methods obtain much better word alignment than the method by attention. Furthermore, based on one of the proposed method, we design a criterion to divide target words into two categories (i.e. those mostly contributed from source ""CFS"" words and the other words mostly contributed from target ""CFT"" words), and analyze word alignment under these two categories in depth. We find that although NMT models are difficult to capture word alignment for CFT words but these words do not sacrifice translation quality significantly, which provides an explanation why NMT is more successful for translation yet worse for word alignment compared to statistical machine translation. We further demonstrate that word alignment errors for CFS words are responsible for translation errors in some extent by measuring the correlation between word alignment and translation for several NMT systems.""","""This paper examines the relationship between attention and alignment in NMT. The reviewers all agreed that this is a valuable topic that is worth thinking about. However, there were concerns both about the clarity of the paper and the framing with respect to previous work. First, it was hard for some reviewers to understand exactly what the paper was trying to do due to issues of the paper structure, etc. Second, there are a number of previous works that also examine similar concepts, and the description of how the proposed method differs seemed lacking. Due to these issues, I cannot recommend it for acceptance in its current form."""
712,"""Log Hyperbolic Cosine Loss Improves Variational Auto-Encoder""","['Unsupervised Generative Model', 'VAE', 'log hyperbolic cosine loss']","""In Variational Auto-Encoder (VAE), the default choice of reconstruction loss function between the decoded sample and the input is the squared pseudo-formula . We propose to replace it with the log hyperbolic cosine (log-cosh) loss, which behaves as pseudo-formula at small values and as pseudo-formula at large values, and differentiable everywhere. Compared with pseudo-formula , the log-cosh loss improves the reconstruction without damaging the latent space optimization, thus automatically keeping a balance between the reconstruction and the generation. Extensive experiments on MNIST and CelebA datasets show that the log-cosh reconstruction loss significantly improves the performance of VAE and its variants in output quality, measured by sharpness and FID score. In addition, the gradient of the log-cosh is a simple tanh function, which makes the implementation of gradient descent as simple as adding one sentence in coding. ""","""The reviewers agree that the paper is well-written, and they all seem to like the general idea. One of the earlier criticisms was that you did not compare against other robust loss functions, but you have partially rectified that by comparing to L1 in the appendix. As per the request of reviewer 2 I would also compare to the Huber loss. One remaining concern is the lack of theoretical justification, which could help address the comment of reviewer 3 regarding blurry images from location uncertainty. The other concern is that you should compare your method using FID scores from a standard implementation so that your numbers are comparable to other papers. Some of the reviewers were impressed, but confused by your relatively low scores."""
713,"""Trace-back along capsules and its application on semantic segmentation ""","['capsule', 'capsule network', 'semantic segmentation', 'FCN']","""In this paper, we propose a capsule-based neural network model to solve the semantic segmentation problem. By taking advantage of the extractable part-whole dependencies available in capsule layers, we derive the probabilities of the class labels for individual capsules through a recursive, layer-by-layer procedure. We model this procedure as a traceback pipeline and take it as a central piece to build an end-to-end segmentation network. Under the proposed framework, image-level class labels and object boundaries are jointly sought in an explicit manner, which poses a significant advantage over the state-of-the-art fully convolutional network (FCN) solutions. Experiments conducted on modified MNIST and neuroimages demonstrate that our model considerably enhance the segmentation performance compared to the leading FCN variant. ""","""This paper proposes a method for tracing activations in a capsule-based network in order to obtain semantic segmentation from classification predictions. Reviewers 1 and 2 rate the paper as marginally above threshold, while Reviewer 3 rates it as marginally below. Reviewer 3 particularly points to experimental validation as a major weakness, stating: ""not sure if the method will generalize well beyond MNIST"", ""Im concerned that the results are not transferable to other datasets and that the method shines promising just because of the simple datasets only."" The AC shares these concerns and does not believe the current experimental validation is sufficient. MNIST is a toy dataset, and may have been appropriate for introducing capsules as a new concept, but it is simply not difficult enough to serve as a quantitative benchmark to distinguish capsule performance from U-Net. U-Net and Tr-CapsNet appear to have similar performance on both MNIST and the hippocampus dataset; the relatively small advantage to Tr-CapsNet is not convincing. Furthermore, as Reviewer 1 suggests, it would seem appropriate to include experimental comparison to other capsule-based segmentation approaches (e.g. LaLonde and Bagci, Capsules for Object Segmentation, 2018). This related work is mentioned, but not used as an experimental baseline. """
714,"""Learning Actionable Representations with Goal Conditioned Policies""","['Representation Learning', 'Reinforcement Learning']","""Representation learning is a central challenge across a range of machine learning areas. In reinforcement learning, effective and functional representations have the potential to tremendously accelerate learning progress and solve more challenging problems. Most prior work on representation learning has focused on generative approaches, learning representations that capture all the underlying factors of variation in the observation space in a more disentangled or well-ordered manner. In this paper, we instead aim to learn functionally salient representations: representations that are not necessarily complete in terms of capturing all factors of variation in the observation space, but rather aim to capture those factors of variation that are important for decision making -- that are ""actionable"". These representations are aware of the dynamics of the environment, and capture only the elements of the observation that are necessary for decision making rather than all factors of variation, eliminating the need for explicit reconstruction. We show how these learned representations can be useful to improve exploration for sparse reward problems, to enable long horizon hierarchical reinforcement learning, and as a state representation for learning policies for downstream tasks. We evaluate our method on a number of simulated environments, and compare it to prior methods for representation learning, exploration, and hierarchical reinforcement learning.""","""To borrow the succinct summary from R1, ""the paper suggests a method for generating representations that are linked to goals in reinforcement learning. More precisely, it wishes to learn a representation so that two states are similar if the policies leading to them are similar."" The reviewers and AC agree that this is a novel and worthy idea. Concerns about the paper are primarily about the following. (i) the method already requires good solutions as input, i.e., in the form of goal-conditioned policies, (GCPs) and the paper claims that these are easy to learn in any case. As R3 notes, this then begs the question as to why the actionable representations are needed. (ii) reviewers had questions regarding the evaluations, i.e., fairness of baselines, additional comparisons, and additional detail. After much discussion, there is now a fair degree of consensus. While R1 (the low score) still has a remaining issue with evaluation, particularly hyperparameter evaluation, they are also ok with acceptance. The AC is of the opinion that hyperparameter tuning is of course an important issue, but does not see it as the key issue for this particular paper. The AC is of the opinion that the key issue is issue (i), raised by R3. In the discussion, the authors reconcile the inherent contradiction in (i) based on the need of additional downstream tasks that can then benefit from the actionable representation, and as demonstrated in a number of the evaluation examples (at least in the revised version). The AC believes in this logic, but believes that this should be stated more clearly in the final paper. And it should be explained the extent to which training for auxiliary tasks implicitly solve this problem in any case. The AC also suggests nominating R3 for a best-reviewer award."""
715,"""Semi-supervised Learning with Multi-Domain Sentiment Word Embeddings""",[],"""Word embeddings are known to boost performance of many NLP tasks such as text classification, meanwhile they can be enhanced by labels at the document level to capture nuanced meaning such as sentiment and topic. Can one combine these two research directions to benefit from both? In this paper, we propose to jointly train a text classifier with a label-enhanced and domain-aware word embedding model, using an unlabeled corpus and only a few labeled data from non-target domains. The embeddings are trained on the unlabed corpus and enhanced by pseudo labels coming from the classifier, and at the same time are used by the classifier as input and training signals. We formalize this symbiotic cycle in a variational Bayes framework, and show that our method improves both the embeddings and the text classifier, outperforming state-of-the-art domain adaptation and semi-supervised learning techniques. We conduct detailed ablative tests to reveal gains from important components of our approach. The source code and experiment data will be publicly released.""","""Pros: - The paper is well written Cons: - Not very novel - Evaluation only on sentiment classification, whereas approaches applicable in broader context exists - There are question re baselines (R3) Neither reviewer was particularly enthusiastic about the paper, I believe, mostly because of the limited score and novelty. """
716,"""Reliable Uncertainty Estimates in Deep Neural Networks using Noise Contrastive Priors""","['uncertainty estimates', 'out of distribution', 'bayesian neural network', 'neural network priors', 'regression', 'active learning']","""Obtaining reliable uncertainty estimates of neural network predictions is a long standing challenge. Bayesian neural networks have been proposed as a solution, but it remains open how to specify their prior. In particular, the common practice of a standard normal prior in weight space imposes only weak regularities, causing the function posterior to possibly generalize in unforeseen ways on inputs outside of the training distribution. We propose noise contrastive priors (NCPs) to obtain reliable uncertainty estimates. The key idea is to train the model to output high uncertainty for data points outside of the training distribution. NCPs do so using an input prior, which adds noise to the inputs of the current mini batch, and an output prior, which is a wide distribution given these inputs. NCPs are compatible with any model that can output uncertainty estimates, are easy to scale, and yield reliable uncertainty estimates throughout training. Empirically, we show that NCPs prevent overfitting outside of the training distribution and result in uncertainty estimates that are useful for active learning. We demonstrate the scalability of our method on the flight delays data set, where we significantly improve upon previously published results.""","""The paper studies the problem of uncertainty estimation of neural networks and proposes to use Bayesian approach with noice contrastive prior. The reviewers and AC note the potential weaknesses of experimental results: (1) lack of sufficient datasets with moderate-to-high dimensional inputs, (2) arguable choices of hyperparameters and (3) lack of direct evaluations, e.g., measuring network calibration is better than active learning. The paper is well written and potentially interesting. However, AC decided that the paper might not be ready to publish in the current form due to the weakness."""
717,"""Hybrid Policies Using Inverse Rewards for Reinforcement Learning""","['Reinforcement Learning', 'Rewards']","""This paper puts forward a broad-spectrum improvement for reinforcement learning algorithms, which combines the policies using original rewards and inverse (negative) rewards. The policies using inverse rewards are competitive with the original policies, and help the original policies correct their mis-actions. We have proved the convergence of the inverse policies. The experiments for some games in OpenAI gym show that the hybrid polices based on deep Q-learning, double Q-learning, and on-policy actor-critic obtain the rewards up to 63.8%, 97.8%, and 54.7% more than the original algorithms. The improved polices are more stable than the original policies as well.""","""Pros: - an original idea: learn an additional inverse policy (that minimizes reward) to help find actions that should be avoided. Cons: - not clearly presented - conclusions are not not validated - empirical evidence is weak - no rebuttal The three reviewers reached consensus that the paper should be rejected in its current form, but make numerous suggestions for improving it for a future submission. """
718,"""The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervision""","['Neuro-Symbolic Representations', 'Concept Learning', 'Visual Reasoning']","""We propose the Neuro-Symbolic Concept Learner (NS-CL), a model that learns visual concepts, words, and semantic parsing of sentences without explicit supervision on any of them; instead, our model learns by simply looking at images and reading paired questions and answers. Our model builds an object-based scene representation and translates sentences into executable, symbolic programs. To bridge the learning of two modules, we use a neuro-symbolic reasoning module that executes these programs on the latent scene representation. Analogical to human concept learning, the perception module learns visual concepts based on the language description of the object being referred to. Meanwhile, the learned visual concepts facilitate learning new words and parsing new sentences. We use curriculum learning to guide the searching over the large compositional space of images and language. Extensive experiments demonstrate the accuracy and efficiency of our model on learning visual concepts, word representations, and semantic parsing of sentences. Further, our method allows easy generalization to new object attributes, compositions, language concepts, scenes and questions, and even new program domains. It also empowers applications including visual question answering and bidirectional image-text retrieval.""","""Strong paper in an interesting new direction. More work should be done in this area."""
719,"""Controlling Over-generalization and its Effect on Adversarial Examples Detection and Generation""","['Convolutional Neural Networks', 'Adversarial Instances', 'Out-distribution Samples', 'Rejection Option', 'Over-generalization']","""Convolutional Neural Networks (CNNs) significantly improve the state-of-the-art for many applications, especially in computer vision. However, CNNs still suffer from a tendency to confidently classify out-distribution samples from unknown classes into pre-defined known classes. Further, they are also vulnerable to adversarial examples. We are relating these two issues through the tendency of CNNs to over-generalize for areas of the input space not covered well by the training set. We show that a CNN augmented with an extra output class can act as a simple yet effective end-to-end model for controlling over-generalization. As an appropriate training set for the extra class, we introduce two resources that are computationally efficient to obtain: a representative natural out-distribution set and interpolated in-distribution samples. To help select a representative natural out-distribution set among available ones, we propose a simple measurement to assess an out-distribution set's fitness. We also demonstrate that training such an augmented CNN with representative out-distribution natural datasets and some interpolated samples allows it to better handle a wide range of unseen out-distribution samples and black-box adversarial examples without training it on any adversaries. Finally, we show that generation of white-box adversarial attacks using our proposed augmented CNN can become harder, as the attack algorithms have to get around the rejection regions when generating actual adversaries.""","""The reviewers agree the paper is not ready for publication at ICLR."""
720,"""Normalization Gradients are Least-squares Residuals""","['Deep Learning', 'Normalization', 'Least squares', 'Gradient regression']","""Batch Normalization (BN) and its variants have seen widespread adoption in the deep learning community because they improve the training of deep neural networks. Discussions of why this normalization works so well remain unsettled. We make explicit the relationship between ordinary least squares and partial derivatives computed when back-propagating through BN. We recast the back-propagation of BN as a least squares fit, which zero-centers and decorrelates partial derivatives from normalized activations. This view, which we term {\em gradient-least-squares}, is an extensible and arithmetically accurate description of BN. To further explore this perspective, we motivate, interpret, and evaluate two adjustments to BN.""","""This paper interprets batch norm in terms of normalizing the backpropagated gradients. All of the reviewers believe this interpretation is novel and potentially interesting, but that the paper doesn't make the case that this helps explain batch norm, or provide useful insights into how to improve it. The authors have responded to the original set of reviews by toning down some of the claims in the original paper, but haven't addressed the reviewers' more substantive concerns. There may potentially be interesting ideas here, but I don't think it's ready for publication at ICLR. """
721,"""Deep, Skinny Neural Networks are not Universal Approximators""","['neural network', 'universality', 'expressability']","""In order to choose a neural network architecture that will be effective for a particular modeling problem, one must understand the limitations imposed by each of the potential options. These limitations are typically described in terms of information theoretic bounds, or by comparing the relative complexity needed to approximate example functions between different architectures. In this paper, we examine the topological constraints that the architecture of a neural network imposes on the level sets of all the functions that it is able to approximate. This approach is novel for both the nature of the limitations and the fact that they are independent of network depth for a broad family of activation functions.""","""The paper shows limitations on the types of functions that can be represented by deep skinny networks for certain classes of activation functions, independently of the number of layers. With many other works discussing capabilities but not limitations, the paper contributes to a relatively underexplored topic. The settings capture a large family of activation functions, but exclude others, such as polynomial activations, for which the considered type of obstructions would not apply. Also a concern is raised about it not being clear how this theoretical result can shed insight on the empirical study of neural networks. The authors have responded to some of the comments of the reviewers, but not to all comments, in particular comments of reviewer 1, who's positive review is conditional on the authors addressing some points. The reviewers are all confident and are moderately positive, positive, or very positive about this paper. """
722,"""Metric-Optimized Example Weights""",[],"""Real-world machine learning applications often have complex test metrics, and may have training and test data that follow different distributions. We propose addressing these issues by using a weighted loss function with a standard convex loss, but with weights on the training examples that are learned to optimize the test metric of interest on the validation set. These metric-optimized example weights can be learned for any test metric, including black box losses and customized metrics for specific applications. We illustrate the performance of our proposal with public benchmark datasets and real-world applications with domain shift and custom loss functions that balance multiple objectives, impose fairness policies, and are non-convex and non-decomposable.""","""While there was some support for the ideas presented, the majority of reviewers did not think this paper was ready for publication at ICLR. In particular the experiments need more work, including the protocol for validation, and attention to overfitting."""
723,"""What a difference a pixel makes: An empirical examination of features used by CNNs for categorisation""","['deep learning', 'shape bias', 'vision', 'feature selection']","""Convolutional neural networks (CNNs) were inspired by human vision and, in some settings, achieve a performance comparable to human object recognition. This has lead to the speculation that both systems use similar mechanisms to perform recognition. In this study, we conducted a series of simulations that indicate that there is a fundamental difference between human vision and CNNs: while object recognition in humans relies on analysing shape, CNNs do not have such a shape-bias. We teased apart the type of features selected by the model by modifying the CIFAR-10 dataset so that, in addition to containing objects with shape, the images concurrently contained non-shape features, such as a noise-like mask. When trained on these modified set of images, the model did not show any bias towards selecting shapes as features. Instead it relied on whichever feature allowed it to perform the best prediction -- even when this feature was a noise-like mask or a single predictive pixel amongst 50176 pixels. We also found that regularisation methods, such as batch normalisation or Dropout, did not change this behaviour and neither did past or concurrent experience with images from other datasets.""","""This paper claims to demonstrate that CNNs, unlike human vision, do not have a bias towards reliance on shape for object recognition. Both AnonReviewer1 and AnonReviewer2 point to fundamental flaws in the paper's argument, which the rebuttal fails to resolve. (AnonReviewer1's criticisms are unfortunately conflated with AnonReviewer1's reluctance to view neuroscience or biological vision as an appropriate topic for ICLR; nonetheless AnonReviewer1's technical criticism stands). These observations are: AnonReviewer2: ""Authors have carefully designed a set of experiments which shows CNNs will [overfit] to non-shape features that they added to training images. However, this outcome is not surprising."" AnonReviewer1: ""The experiments don't seem to effectively demonstrate the main claim of the paper that categorization CNNs do not have inductive shape bias"" ""The best way to demonstrate this would have been to subject a trained image-categorization CNN to test data with object shapes in a way that the appearance information couldnt be used to predict the object label. The paper doesnt do this. None of the experiments logically imply that with an unaltered training regime, a trained network would not be predictive of the category label if shapes corresponding to that category are presented."" The AC agrees with both of these observations. CNN behavior is partially a product of the training regime. To examine the scientific question of whether CNNs have similar biases as human vision, the training regimes should be similar. Conversely, if human vision evolved in an environment in which shortcut recognition cues were available via indicator pixels, perhaps it would not have a shape bias. This paper appears fundamentally flawed in its approach. The results are not informative about differences between human vision and CNNs, nor are they surprising to machine learning practitioners."""
724,"""SHAMANN: Shared Memory Augmented Neural Networks""","['memory networks', 'deep learning', 'medical image segmentation']","""Current state-of-the-art methods for semantic segmentation use deep neural networks to learn the segmentation mask from the input image signal as an image-to-image mapping. While these methods effectively exploit global image context, the learning and computational complexities are high. We propose shared memory augmented neural network actors as a dynamically scalable alternative. Based on a decomposition of the image into a sequence of local patches, we train such actors to sequentially segment each patch. To further increase the robustness and better capture shape priors, an external memory module is shared between different actors, providing an implicit mechanism for image information exchange. Finally, the patch-wise predictions are aggregated to a complete segmentation mask. We demonstrate the benefits of the new paradigm on a challenging lung segmentation problem based on chest X-Ray images, as well as on two synthetic tasks based on the MNIST dataset. On the X-Ray data, our method achieves state-of-the-art accuracy with a significantly reduced model size compared to reference methods. In addition, we reduce the number of failure cases by at least half.""","""The paper addresses the problem semantic segmentation using a sequential patch-based model. I agree with the reviewers that the contributions of the paper are not enough for a machine learning venue: (1) there has been prior work on using sequence models for segmentation and (2) the complexity of the proposed approach is not fully justified. The authors did not submit a rebuttal. I encourage the authors to take the feedback into account and improve the paper."""
725,"""NLProlog: Reasoning with Weak Unification for Natural Language Question Answering""","['symbolic reasoning', 'neural networks', 'natural language processing', 'question answering', 'sentence embeddings', 'evolution strategies']","""Symbolic logic allows practitioners to build systems that perform rule-based reasoning which is interpretable and which can easily be augmented with prior knowledge. However, such systems are traditionally difficult to apply to problems involving natural language due to the large linguistic variability of language. Currently, most work in natural language processing focuses on neural networks which learn distributed representations of words and their composition, thereby performing well in the presence of large linguistic variability. We propose to reap the benefits of both approaches by applying a combination of neural networks and logic programming to natural language question answering. We propose to employ an external, non-differentiable Prolog prover which utilizes a similarity function over pretrained sentence encoders. We fine-tune these representations via Evolution Strategies with the goal of multi-hop reasoning on natural language. This allows us to create a system that can apply rule-based reasoning to natural language and induce domain-specific natural language rules from training data. We evaluate the proposed system on two different question answering tasks, showing that it complements two very strong baselines BIDAF (Seo et al., 2016a) and FASTQA (Weissenborn et al.,2017) and outperforms both when used in an ensemble.""","""This paper combines Prolog-like reasoning with distributional semantics, applied to natural language question answering. Given the importance of combining neural and symbolic techniques, this paper provides an important contribution. Further, the proposed method complements standard QA models as it can be easily combined with them. The reviewers and AC note the following potential weaknesses: (1) The evaluation consisted primarily on small subsets of existing benchmarks, (2) the reviewers were concerned that the handcrafted rules were introducing domain information into the model, and (3) were unconvinced that the benefits of the proposed approach were actually complementary to existing neural models. The authors addressed a number of these concerns in the response and their revision. They discussed how OpenIE affects the performance, and other questions the reviewers had. Further, they clarified that the rule templates are really high-level/generic and not ""prior knowledge"" as the reviewers had initially assumed. The revision also provided more error analysis, and heavily edited the paper for clarity. Although these changes increased the reviewer scores, a critical concern still remains: the evaluation is not performed on the complete question-answering benchmark, but on small subsets of the data, and the benefits are not significant. This makes the evaluation quite weak, and the authors are encouraged to identify appropriate evaluation benchmarks. There is disagreement in the reviewer scores; even though all of them identified the weak evaluation as a concern, some are more forgiving than others, partly due to the other improvements made to the paper. The AC, however, agrees with reviewer 2 that the empirical results need to be sound for this paper to have an impact, and thus is recommending a rejection. Please note that paper was incredibly close to an acceptance, but identifying appropriate benchmarks will make the paper much stronger."""
726,"""Differentiable Expected BLEU for Text Generation""","['text generation', 'BLEU', 'differentiable', 'gradient descent', 'maximum likelihood learning', 'policy gradient', 'machine translation']","""Neural text generation models such as recurrent networks are typically trained by maximizing data log-likelihood based on cross entropy. Such training objective shows a discrepancy from test criteria like the BLEU metric. Recent work optimizes expected BLEU under the model distribution using policy gradient, while such algorithm can suffer from high variance and become impractical. In this paper, we propose a new Differentiable Expected BLEU (DEBLEU) objective that permits direct optimization of neural generation models with gradient descent. We leverage the decomposability and sparsity of BLEU, and reformulate it with moderate approximations, making the evaluation of the objective and its gradient efficient, comparable to common cross-entropy loss. We further devise a simple training procedure with ground-truth masking and annealing for stable optimization. Experiments on neural machine translation and image captioning show our method significantly improves over both cross-entropy and policy gradient training.""","""The paper presents a differentiable approximation of BLEU score, which can be directly optimized using SGD. The reviewers raised concerns about (1) direct evaluation of the quality of the approximation and (2) the significance of the experimental results. There is also a concern (3) regarding the significance of BLEU score in the first place, and whether BLEU is the right metric that one needs to directly optimize. The authors did not provide a response, and based on the concerns above (especially 1-2) I believe that the paper does not pass the bar for acceptance at ICLR."""
727,"""Guiding Policies with Language via Meta-Learning""","['meta-learning', 'language grounding', 'interactive']","""Behavioral skills or policies for autonomous agents are conventionally learned from reward functions, via reinforcement learning, or from demonstrations, via imitation learning. However, both modes of task specification have their disadvantages: reward functions require manual engineering, while demonstrations require a human expert to be able to actually perform the task in order to generate the demonstration. Instruction following from natural language instructions provides an appealing alternative: in the same way that we can specify goals to other humans simply by speaking or writing, we would like to be able to specify tasks for our machines. However, a single instruction may be insufficient to fully communicate our intent or, even if it is, may be insufficient for an autonomous agent to actually understand how to perform the desired task. In this work, we propose an interactive formulation of the task specification problem, where iterative language corrections are provided to an autonomous agent, guiding it in acquiring the desired skill. Our proposed language-guided policy learning algorithm can integrate an instruction and a sequence of corrections to acquire new skills very quickly. In our experiments, we show that this method can enable a policy to follow instructions and corrections for simulated navigation and manipulation tasks, substantially outperforming direct, non-interactive instruction following.""","""The paper proposes a meta-learning approach to ""language guided policy learning"" where instructions are provided in the form of natural language instructions, rather than in the form of a reward function or through demonstration. A particularly interesting novel feature of the proposed approach is that it can seamlessly incorporate natural language corrections after an initial attempt to solve the task, opening up the direction towards natural instructions through interactive dialogue. The method is empirically shown to be able to learn to navigate environments and manipulate objects more sample efficiently (on test tasks) than approaches without instructions. The reviewers noted several potential weaknesses: while the problem setting was considered interesting, the empirical validation was seen to be limited. Reviewers noted that only one (simple) domain was studied, and it was unclear if results would hold up in more complex domains. They also note lack of comparison to baselines based on prior work (e.g., pre-training). The authors provided very detailed replies to the reviewer comments, and added very substantial new experiments, including an entire new domain and newly implemented baselines. Reviewers indicated that they are satisfied with the revisions. The AC reviewed the reviewer suggestions and revisions and notes that the additional experiments significantly improve the contribution of the paper. The resulting consensus is that the paper should be accepted. The AC would like to note that several figures are very small and unreadable when the paper is printed, e.g., figure 7, and suggests that the authors increase figure size (and font size within figures) to ensure legibility."""
728,"""Towards More Theoretically-Grounded Particle Optimization Sampling for Deep Learning""",[],"""Many deep-learning based methods such as Bayesian deep learning (DL) and deep reinforcement learning (RL) have heavily relied on the ability of a model being able to efficiently explore via Bayesian sampling. Particle-optimization sampling (POS) is a recently developed technique to generate high-quality samples from a target distribution by iteratively updating a set of interactive particles, with a representative algorithm the Stein variational gradient descent (SVGD). Though obtaining significant empirical success, the {\em non-asymptotic} convergence behavior of SVGD remains unknown. In this paper, we generalize POS to a stochasticity setting by injecting random noise in particle updates, called stochastic particle-optimization sampling (SPOS). Notably, for the first time, we develop {\em non-asymptotic convergence theory} for the SPOS framework, characterizing convergence of a sample approximation w.r.t.\! the number of particles and iterations under both convex- and noncovex-energy-function settings. Interestingly, we provide theoretical understanding of a pitfall of SVGD that can be avoided in the proposed SPOS framework, {\it i.e.}, particles tend to collapse to a local mode in SVGD under some particular conditions. Our theory is based on the analysis of nonlinear stochastic differential equations, which serves as an extension and a complementary development to the asymptotic convergence theory for SVGD such as (Liu, 2017). With such theoretical guarantees, SPOS can be safely and effectively applied on both Bayesian DL and deep RL tasks. Extensive results demonstrate the effectiveness of our proposed framework.""","""This paper proposes a combination of SVGD and SLGD and analyzes its non-asymptotic properties based on gradient flow. This is an interesting direction to explore. Unfortunately, two major concerns have been raised regarding this paper: 1) the reviewers identified multiple technical flaws. Authors provided rebuttal and addressed some of the problems. But the reviewers think it requires significantly more improvement and clarification to fully address the issues. 2) the motivation of the combination of SVGD and SLGD, despite of being very interesting, is not very clearly motivated; by combining SVGD and SLGD, one get convergence rate for free from the SLGD part, but not much insight is shed on the SVGD part (meaning if the contribution of SLGD is zero, then the bound because vacuum). This could be misleading given that one of the claimed contribution is non-asymptotic theory of ''SVGD-style algorithms"" (rather than SLGD style..). We encourage the authors to addresses the technical questions and clarify the contribution and motivation of the paper in revision for future submissions. """
729,"""Inter-BMV: Interpolation with Block Motion Vectors for Fast Semantic Segmentation on Video""","['semantic segmentation', 'video', 'efficient inference', 'video segmentation', 'video compression']","""Models optimized for accuracy on single images are often prohibitively slow to run on each frame in a video, especially on challenging dense prediction tasks, such as semantic segmentation. Recent work exploits the use of optical flow to warp image features forward from select keyframes, as a means to conserve computation on video. This approach, however, achieves only limited speedup, even when optimized, due to the accuracy degradation introduced by repeated forward warping, and the inference cost of optical flow estimation. To address these problems, we propose a new scheme that propagates features using the block motion vectors (BMV) present in compressed video (e.g. H.264 codecs), instead of optical flow, and bi-directionally warps and fuses features from enclosing keyframes to capture scene context on each video frame. Our technique, interpolation-BMV, enables us to accurately estimate the features of intermediate frames, while keeping inference costs low. We evaluate our system on the CamVid and Cityscapes datasets, comparing to both a strong single-frame baseline and related work. We find that we are able to substantially accelerate segmentation on video, achieving near real-time frame rates (20+ frames per second) on large images (e.g. 960 x 720 pixels), while maintaining competitive accuracy. This represents an improvement of almost 6x over the single-frame baseline and 2.5x over the fastest prior work.""","""Strengths: Paper uses an efficient inference procedure cutting inference time on intermediate frames by 53%, & yields better accuracy and IOU compared to the one recent closely related work. The ablation study seems sufficient and well-designed. The paper presents two feature propagation strategies and three feature fusion methods. The experiments compare these different settings, and show that interpolation-BMV is indeed a better feature propagation. Weaknesses: Reviewers believed the work to be of limited novelty. The algorithm is close to the optical-flow based models Shelhamer et al. (2016) and Zhu et al. (2017). Reviewer asserts that the main difference is that the optical-flow is replaced with BMV, which is a byproduct of modern cameras. R3 felt that there was Insufficient experimental comparison with other baselines and that technical details were not clear enough. Contention: Authors assert that Shelhamer et al. (2016) does not use optical flow, and instead simply copies features from frame to frame (and schedules this copying). Zhu et al. (2017) then proposes an improvement to this scheme, forward feature warping with optical flow. In general, both these techniques fail to achieve speedups beyond small multiples of the baseline (< 3x), without impacting accuracy. Consensus: It was disappointing that some of the reviewers did not engage after the author review (perhaps initial impressions were just too low). However, after the author rebuttal R1 did respond and held to the position that the work should not be accepted, justified by the assertion that other modern architectures that are lighter weight and are able to produce fast predictions. """
730,"""Adaptivity of deep ReLU network for learning in Besov and mixed smooth Besov spaces: optimal rate and curse of dimensionality""","['deep learning theory', 'approximation analysis', 'generalization error analysis', 'Besov space', 'minimax optimality']","""Deep learning has shown high performances in various types of tasks from visual recognition to natural language processing, which indicates superior flexibility and adaptivity of deep learning. To understand this phenomenon theoretically, we develop a new approximation and estimation error analysis of deep learning with the ReLU activation for functions in a Besov space and its variant with mixed smoothness. The Besov space is a considerably general function space including the Holder space and Sobolev space, and especially can capture spatial inhomogeneity of smoothness. Through the analysis in the Besov space, it is shown that deep learning can achieve the minimax optimal rate and outperform any non-adaptive (linear) estimator such as kernel ridge regression, which shows that deep learning has higher adaptivity to the spatial inhomogeneity of the target function than other estimators such as linear ones. In addition to this, it is shown that deep learning can avoid the curse of dimensionality if the target function is in a mixed smooth Besov space. We also show that the dependency of the convergence rate on the dimensionality is tight due to its minimax optimality. These results support high adaptivity of deep learning and its superior ability as a feature extractor. ""","""The paper extends the results in Yarotsky (2017) from Sobolev spaces to Besov spaces, stating that once the target function lies in certain Besov spaces, there exists some deep neural networks with ReLU activation that approximate the target in the minimax optimal rates. Such adaptive networks can be found by empirical risk minimization, which however is not yet known to be found by SGDs etc. This gap is the key weakness of applying approximation theory to the study of constructive deep neural networks of certain approximation spaces, which lacks algorithmic guarantees. The gap is hoped to be filled in future studies. Despite the incompleteness of approximation theory, this paper is still a good solid work. Based on fact that the majority of reviewers suggest accept (6,8,6), with some concerns on the clarity, the paper is proposed as probable accept. """
731,"""Universal Successor Features for Transfer Reinforcement Learning""","['Reinforcement Learning', 'Successor Features', 'Successor Representations', 'Transfer Learning', 'Representation Learning']","""Transfer in Reinforcement Learning (RL) refers to the idea of applying knowledge gained from previous tasks to solving related tasks. Learning a universal value function (Schaul et al., 2015), which generalizes over goals and states, has previously been shown to be useful for transfer. However, successor features are believed to be more suitable than values for transfer (Dayan, 1993; Barreto et al.,2017), even though they cannot directly generalize to new goals. In this paper, we propose (1) Universal Successor Features (USFs) to capture the underlying dynamics of the environment while allowing generalization to unseen goals and (2) a flexible end-to-end model of USFs that can be trained by interacting with the environment. We show that learning USFs is compatible with any RL algorithm that learns state values using a temporal difference method. Our experiments in a simple gridworld and with two MuJoCo environments show that USFs can greatly accelerate training when learning multiple tasks and can effectively transfer knowledge to new tasks.""","""In considering the reviews and the author response, I would summarize the evaluation of the paper as following: The main idea in the paper -- to combine goal-conditioning with successor features -- is an interesting direction for research, but is somewhat incremental in light of the prior work in the area. Most of the reviewers generally agreed on this point. While a relatively incremental technical contribution could still result in a successful paper with a thorough empirical analysis and compelling results, the evaluation in the paper is unfortunately not very extensive: the provided tasks are very simple, and the difference from prior methods is not very large. All of the tasks are equivalent to either grid worlds or reaching, which are very simple. Without a deeper technical contribution or a more extensive empirical evaluation, I do not think the paper is ready for publication in ICLR."""
732,"""Towards GAN Benchmarks Which Require Generalization""","['evaluation', 'generative adversarial networks', 'adversarial divergences']","""For many evaluation metrics commonly used as benchmarks for unconditional image generation, trivially memorizing the training set attains a better score than models which are considered state-of-the-art; we consider this problematic. We clarify a necessary condition for an evaluation metric not to behave this way: estimating the function must require a large sample from the model. In search of such a metric, we turn to neural network divergences (NNDs), which are defined in terms of a neural network trained to distinguish between distributions. The resulting benchmarks cannot be ``won'' by training set memorization, while still being perceptually correlated and computable only from samples. We survey past work on using NNDs for evaluation, implement an example black-box metric based on these ideas, and validate experimentally that it can measure a notion of generalization. ""","""The paper argues for a GAN evaluation metric that needs sufficiently large number of generated samples to evaluate. Authors propose a metric based on existing set of divergences computed with neural net representations. R2 and R3 appreciate the motivation behind the proposed method and the discussion in the paper to that end. The proposed NND based metric has some limitations as pointed out by R2/R3 and also acknowledged by the authors -- being biased towards GANs learned with the same NND metric; challenge in choosing the capacity of the metric neural network; being computationally expensive, etc. However, these points are discussed well in the paper, and R2 and R3 are in favor of accepting the paper (with R3 bumping their score up after the author response). R1's main concern is the lack of rigorous theoretical analysis of the proposed metric, which the AC agrees with, but is willing to overlook, given that it is nontrivial and most existing evaluation metrics in the literature also lack this. Overall, this is a borderline paper but falling on the accept side according to the AC. """
733,"""Backprop with Approximate Activations for Memory-efficient Network Training""","['Back-propagation', 'Memory Efficient Training', 'Approximate Gradients', 'Deep Learning']","""With innovations in architecture design, deeper and wider neural network models deliver improved performance on a diverse variety of tasks. But the increased memory footprint of these models presents a challenge during training, when all intermediate layer activations need to be stored for back-propagation. Limited GPU memory forces practitioners to make sub-optimal choices: either train inefficiently with smaller batches of examples; or limit the architecture to have lower depth and width, and fewer layers at higher spatial resolutions. This work introduces an approximation strategy that significantly reduces a network's memory footprint during training, but has negligible effect on training performance and computational expense. During the forward pass, we replace activations with lower-precision approximations immediately after they have been used by subsequent layers, thus freeing up memory. The approximate activations are then used during the backward pass. This approach limits the accumulation of errors across the forward and backward pass---because the forward computation across the network still happens at full precision, and the approximation has a limited effect when computing gradients to a layer's input. Experiments, on CIFAR and ImageNet, show that using our approach with 8- and even 4-bit fixed-point approximations of 32-bit floating-point activations has only a minor effect on training and validation performance, while affording significant savings in memory usage.""","""This work proposes to reduce memory use in network training by quantizing the activations during backprop. It shows that this leads to only small drops in accuracy for resnets on CIFAR-10 and Imagenet for factors up to 8. The reviewers raised concerns about comparison to other approaches such as checkpointing, and questioned the technical novelty of the approach. The authors were able to properly address the concerns around comparisons, but the issue around novelty remained. This could be compensated by strengthening the experimental results and leveraging the memory saving for instance to train larger networks. Resubmission is encouraged."""
734,"""CEM-RL: Combining evolutionary and gradient-based methods for policy search""","['evolution strategy', 'deep reinforcement learning']","""Deep neuroevolution and deep reinforcement learning (deep RL) algorithms are two popular approaches to policy search. The former is widely applicable and rather stable, but suffers from low sample efficiency. By contrast, the latter is more sample efficient, but the most sample efficient variants are also rather unstable and highly sensitive to hyper-parameter setting. So far, these families of methods have mostly been compared as competing tools. However, an emerging approach consists in combining them so as to get the best of both worlds. Two previously existing combinations use either an ad hoc evolutionary algorithm or a goal exploration process together with the Deep Deterministic Policy Gradient (DDPG) algorithm, a sample efficient off-policy deep RL algorithm. In this paper, we propose a different combination scheme using the simple cross-entropy method (CEM) and Twin Delayed Deep Deterministic policy gradient (TD3), another off-policy deep RL algorithm which improves over DDPG. We evaluate the resulting method, CEM-RL, on a set of benchmarks classically used in deep RL. We show that CEM-RL benefits from several advantages over its competitors and offers a satisfactory trade-off between performance and sample efficiency.""","""This paper combines two different types of existing optimization methods, CEM/CMA-ES and DDPG/TD3, for policy optimization. The approach resembles ERL but demonstrates good better performance on a variety of continuous control benchmarks. Although I feel the novelty of the paper is limited, the provided promising results may justify the acceptance of the paper."""
735,"""Unification of Recurrent Neural Network Architectures and Quantum Inspired Stable Design ""","['theory and analysis of RNNs architectures', 'reversibe evolution', 'stability of deep neural network', 'learning representations of outputs or states', 'quantum inspired embedding']","""Various architectural advancements in the design of recurrent neural networks~(RNN) have been focusing on improving the empirical stability and representability by sacrificing the complexity of the architecture. However, more remains to be done to fully understand the fundamental trade-off between these conflicting requirements. Towards answering this question, we forsake the purely bottom-up approach of data-driven machine learning to understand, instead, the physical origin and dynamical properties of existing RNN architectures. This facilitates designing new RNNs with smaller complexity overhead and provable stability guarantee. First, we define a family of deep recurrent neural networks, pseudo-formula - pseudo-formula -ORNN, according to the order of nonlinearity pseudo-formula and the range of temporal memory scale pseudo-formula in their underlying dynamics embodied in the form of discretized ordinary differential equations. We show that most of the existing proposals of RNN architectures belong to different orders of pseudo-formula - pseudo-formula -ORNNs. We then propose a new RNN ansatz, namely the Quantum-inspired Universal computing Neural Network~(QUNN), to leverage the reversibility, stability, and universality of quantum computation for stable and universal RNN. QUNN provides a complexity reduction in the number of training parameters from being polynomial in both data and correlation time to only linear in correlation time. Compared to Long-Short-Term Memory (LSTM), QUNN of the same number of hidden layers facilitates higher nonlinearity and longer memory span with provable stability. Our work opens new directions in designing minimal RNNs based on additional knowledge about the dynamical nature of both the data and different training architectures.""","""although the way in which the authors characterize existing rnn variants and how they derive a new type of rnn are interesting, the submission lacks justification (either empirical or theoretical) that supports whether and how the proposed rnn's behave in a ""learning"" setting different from the existing rnn variants."""
736,"""The Variational Deficiency Bottleneck""","['Variational Information Bottleneck', 'Blackwell Sufficiency', 'Le Cam Deficiency', 'Information Channel']","""We introduce a bottleneck method for learning data representations based on channel deficiency, rather than the more traditional information sufficiency. A variational upper bound allows us to implement this method efficiently. The bound itself is bounded above by the variational information bottleneck objective, and the two methods coincide in the regime of single-shot Monte Carlo approximations. The notion of deficiency provides a principled way of approximating complicated channels by relatively simpler ones. The deficiency of one channel w.r.t. another has an operational interpretation in terms of the optimal risk gap of decision problems, capturing classification as a special case. Unsupervised generalizations are possible, such as the deficiency autoencoder, which can also be formulated in a variational form. Experiments demonstrate that the deficiency bottleneck can provide advantages in terms of minimal sufficiency as measured by information bottleneck curves, while retaining a good test performance in classification and reconstruction tasks. ""","""Strengths: The paper presents an alternative regularized training objective for supervised learning that has a reasonable theoretical justification. It also has a simple computational formula. Weaknesses: The experiments are minimal proofs of concept on MNIST and fashion MNIST, and the authors didn't find an example where this formulation makes a large difference. The resulting formula is very close to existing methods. Finally the paper is a bit dense and the intuitions we should gain from this theory aren't made clear. Points of contention: One reviewer pointed out the close connection of the new objective to IWAE, and the authors added a discussion of the relation and showed that they're not mathematically equivalent. However, as far as I can tell they're almost identical in purpose: As k -> \infty in IWAE, the encoder ceases to matter. And as M -> \infty in VDB, we take the max over all encoders. Could the method proposed in this paper lead to an alternative to IWAE in the VAE setting? Consensus: Consensus wasn't reached, but the ""7"" reviewer did not appear to have put much though into their review."""
737,"""Deep Lagrangian Networks: Using Physics as Model Prior for Deep Learning""","['Deep Model Learning', 'Robot Control']","""Deep learning has achieved astonishing results on many tasks with large amounts of data and generalization within the proximity of training data. For many important real-world applications, these requirements are unfeasible and additional prior knowledge on the task domain is required to overcome the resulting problems. In particular, learning physics models for model-based control requires robust extrapolation from fewer samples often collected online in real-time and model errors may lead to drastic damages of the system. Directly incorporating physical insight has enabled us to obtain a novel deep model learning approach that extrapolates well while requiring fewer samples. As a first example, we propose Deep Lagrangian Networks (DeLaN) as a deep network structure upon which Lagrangian Mechanics have been imposed. DeLaN can learn the equations of motion of a mechanical system (i.e., system dynamics) with a deep network efficiently while ensuring physical plausibility. The resulting DeLaN network performs very well at robot tracking control. The proposed method did not only outperform previous model learning approaches at learning speed but exhibits substantially improved and more robust extrapolation to novel trajectories and learns online in real-time.""","""The paper looks at a novel form of physics-constrained system identification for a multi-link robot, although it could also be applied more generally. The contributions is in many simple; this is seen in a good light (R1, R3) or more modestly (R2). R3 notes surprise that this hasn't been done before. Results are demonstrated on a simualted 2-dof robot and real Barrett WAM arm, better than a pure neural network modeling approach, PID control, or an analytic model. Some aspects of the writing needed to be addressed, i.e., PDE vs ODE notations. The point of biggest concern is related to positioning the work relative to other system-identification literature, where there has been an abundance of work in the robotics and control literature. There is no final consensus on this point for R3; R3 did not receive the email notification of the author's detailed reply, and notes that the author has clarified some respects, but still has concerns, and did not have time to further provide feedback on short notice. In balance, the AC believes that this kind of constrained learning of models is underexplored, and notes that the reviewers (who have considerable shared expertise in robotics-related work) believe that this is a step in the right direction and that it is surprising this type of approach has not been investigated yet. The authors have further reconciled their work with earlier sys-ID work, and can further describe how their work is situated with respect to prior art in sys-ID (as they do in their discussion comments). The AC recommends that: (a) the abstract explicitly mention ""system identification"" as a relevant context for the work in this paper, given that the ML audience should be (or can be) made aware of this terminology; and (b) push more of the math related to the development of the necessary derivatives to an appendix, given that the particular use of the derivations seems to be more in support of obtaining the performance necessary for online use, rather than something that cannot be accomplished with autodiff. """
738,"""Characterizing Malicious Edges targeting on Graph Neural Networks""",[],"""Deep neural networks on graph structured data have shown increasing success in various applications. However, due to recent studies about vulnerabilities of machine learning models, researchers are encouraged to explore the robustness of graph neural networks (GNNs). So far there are two work targeting to attack GNNs by adding/deleting edges to fool graph based classification tasks. Such attacks are challenging to be detected since the manipulation is very subtle compared with traditional graph attacks. In this paper we propose the first detection mechanism against these two proposed attacks. Given a perturbed graph, we propose a novel graph generation method together with link prediction as preprocessing to detect potential malicious edges. We also propose novel features which can be leveraged to perform outlier detection when the number of added malicious edges are large. Different detection components are proposed and tested, and we also evaluate the performance of final detection pipeline. Extensive experiments are conducted to show that the proposed detection mechanism can achieve AUC above 90% against the two attack strategies on both Cora and Citeseer datasets. We also provide in-depth analysis of different attack strategies and corresponding suitable detection methods. Our results shed light on several principles for detecting different types of attacks.""","""All reviewers recommended rejecting this submission so I will as well. However, I do not believe it is fundamentally misguided or anything of that nature. Unfortunately, reviewers did not participate as much in discussions with the authors as I believe they should. However, this paper concerns a relatively niche problem of modest interest to the ICLR community. I believe a stronger version of this work would be a more application-focused paper that delved into practical details about a specific case study where this work provides a clear benefit."""
739,"""Learning Implicitly Recurrent CNNs Through Parameter Sharing""","['deep learning', 'architecture search', 'computer vision']","""We introduce a parameter sharing scheme, in which different layers of a convolutional neural network (CNN) are defined by a learned linear combination of parameter tensors from a global bank of templates. Restricting the number of templates yields a flexible hybridization of traditional CNNs and recurrent networks. Compared to traditional CNNs, we demonstrate substantial parameter savings on standard image classification tasks, while maintaining accuracy. Our simple parameter sharing scheme, though defined via soft weights, in practice often yields trained networks with near strict recurrent structure; with negligible side effects, they convert into networks with actual loops. Training these networks thus implicitly involves discovery of suitable recurrent architectures. Though considering only the aspect of recurrent links, our trained networks achieve accuracy competitive with those built using state-of-the-art neural architecture search (NAS) procedures. Our hybridization of recurrent and convolutional networks may also represent a beneficial architectural bias. Specifically, on synthetic tasks which are algorithmic in nature, our hybrid networks both train faster and extrapolate better to test examples outside the span of the training set.""","""This paper proposed an interesting approach to weight sharing among CNN layers via shared weight templates to save parameters. It's well written with convincing results. Reviewers have a consensus on accept."""
740,"""BNN+: Improved Binary Network Training""","['Binary Network', 'Binary Training', 'Model Compression', 'Quantization']","""Deep neural networks (DNN) are widely used in many applications. However, their deployment on edge devices has been difficult because they are resource hungry. Binary neural networks (BNN) help to alleviate the prohibitive resource requirements of DNN, where both activations and weights are limited to 1-bit. We propose an improved binary training method (BNN+), by introducing a regularization function that encourages training weights around binary values. In addition to this, to enhance model performance we add trainable scaling factors to our regularization functions. Furthermore, we use an improved approximation of the derivative of the sign activation function in the backward computation. These additions are based on linear operations that are easily implementable into the binary training framework. We show experimental results on CIFAR-10 obtaining an accuracy of 86.5%, on AlexNet and 91.3% with VGG network. On ImageNet, our method also outperforms the traditional BNN method and XNOR-net, using AlexNet by a margin of 4% and 2% top-1 accuracy respectively.""","""The paper makes two fairly incremental contributions regarding training binarized neural networks: (1) the swish-based STE, and (2) a regularization that pushes weights to take on values in {-1, +1}. Reviewer1 and reviewer2 both pointed out concerns about the incremental contribution, the thoroughness of the evaluation, the poor clarity and consistency of the writing. Reviewer3 was muted during the discussion. Given the valid concerns from reviewer1/2, this paper is recommended for rejection. """
741,"""Hierarchical interpretations for neural network predictions""","['interpretability', 'natural language processing', 'computer vision']","""Deep neural networks (DNNs) have achieved impressive predictive performance due to their ability to learn complex, non-linear relationships between variables. However, the inability to effectively visualize these relationships has led to DNNs being characterized as black boxes and consequently limited their applications. To ameliorate this problem, we introduce the use of hierarchical interpretations to explain DNN predictions through our proposed method: agglomerative contextual decomposition (ACD). Given a prediction from a trained DNN, ACD produces a hierarchical clustering of the input features, along with the contribution of each cluster to the final prediction. This hierarchy is optimized to identify clusters of features that the DNN learned are predictive. We introduce ACD using examples from Stanford Sentiment Treebank and ImageNet, in order to diagnose incorrect predictions, identify dataset bias, and extract polarizing phrases of varying lengths. Through human experiments, we demonstrate that ACD enables users both to identify the more accurate of two DNNs and to better trust a DNN's outputs. We also find that ACD's hierarchy is largely robust to adversarial perturbations, implying that it captures fundamental aspects of the input and ignores spurious noise.""","""The paper receives a unanimous accept over reviewers, though some concerns on novelty exist. So it is suggested to be a probable accept. """
742,"""Towards Robust, Locally Linear Deep Networks""","['robust derivatives', 'transparency', 'interpretability']","""Deep networks realize complex mappings that are often understood by their locally linear behavior at or around points of interest. For example, we use the derivative of the mapping with respect to its inputs for sensitivity analysis, or to explain (obtain coordinate relevance for) a prediction. One key challenge is that such derivatives are themselves inherently unstable. In this paper, we propose a new learning problem to encourage deep networks to have stable derivatives over larger regions. While the problem is challenging in general, we focus on networks with piecewise linear activation functions. Our algorithm consists of an inference step that identifies a region around a point where linear approximation is provably stable, and an optimization step to expand such regions. We propose a novel relaxation to scale the algorithm to realistic models. We illustrate our method with residual and recurrent networks on image and sequence datasets.""","""The paper aims to encourage deep networks to have stable derivatives over larger regions under networks with piecewise linear activation functions. All reviewers and AC note the significance of the paper. AC also thinks this is also a very timely work and potentially of broader interest of ICLR audience."""
743,"""An adaptive homeostatic algorithm for the unsupervised learning of visual features""","['Sparse Coding', 'Unsupervised Learning', 'Natural Scene Statistics', 'Biologically Plausible Deep Networks', 'Visual Perception', 'Computer Vision']","""The formation of structure in the brain, that is, of the connections between cells within neural populations, is by large an unsupervised learning process: the emergence of this architecture is mostly self-organized. In the primary visual cortex of mammals, for example, one may observe during development the formation of cells selective to localized, oriented features. This leads to the development of a rough representation of contours of the retinal image in area V1. We modeled these mechanisms using sparse Hebbian learning algorithms. These algorithms alternate a coding step to encode the information with a learning step to find the proper encoder. A major difficulty faced by these algorithms is to deduce a good representation while knowing immature encoders, and to learn good encoders with a non-optimal representation. To address this problem, we propose to introduce a new regulation process between learning and coding, called homeostasis. Our homeostasis is compatible with a neuro-mimetic architecture and allows for the fast emergence of localized filters sensitive to orientation. The key to this algorithm lies in a simple adaptation mechanism based on non-linear functions that reconciles the antagonistic processes that occur at the coding and learning time scales. We tested this unsupervised algorithm with this homeostasis rule for a range of existing unsupervised learning algorithms coupled with different neural coding algorithms. In addition, we propose a simplification of this optimal homeostasis rule by implementing a simple heuristic on the probability of activation of neurons. Compared to the optimal homeostasis rule, we show that this heuristic allows to implement a more rapid unsupervised learning algorithm while keeping a large part of its effectiveness. These results demonstrate the potential application of such a strategy in machine learning and we illustrate this with one result in a convolutional neural network.""","""This paper shows how to obtain more homogeneous activation of atoms in a dictionary. As reviewers point out, the paper is well written and indeed shows that the propose scheme results in a more uniform activation. However, the value of this contribution rests on making a case that uniformity is indeed a desirable outcome per se. As two reviewers explain, this crucial point is left unaddressed, which makes the paper too weak for ICLR."""
744,"""Improved Language Modeling by Decoding the Past""","['language modeling', 'regularization', 'LSTM']","""Highly regularized LSTMs achieve impressive results on several benchmark datasets in language modeling. We propose a new regularization method based on decoding the last token in the context using the predicted distribution of the next token. This biases the model towards retaining more contextual information, in turn improving its ability to predict the next token. With negligible overhead in the number of parameters and training time, our Past Decode Regularization (PDR) method achieves a word level perplexity of 55.6 on the Penn Treebank and 63.5 on the WikiText-2 datasets using a single softmax. We also show gains by using PDR in combination with a mixture-of-softmaxes, achieving a word level perplexity of 53.8 and 60.5 on these datasets. In addition, our method achieves 1.169 bits-per-character on the Penn Treebank Character dataset for character level language modeling. These results constitute a new state-of-the-art in their respective settings.""","""The paper proposes an additional module to train language models, adding a new loss that tries to predict the previous token given the next one, thus enforcing the model to remember the past. Two out of 3 reviewers recommend to accept the paper; the third one said it was misleading to claim SOTA since authors didn't try the mixture-of-softmax model that is actually currently SOTA. The authors acknowledged and modified the paper accordingly, and added a few more experiments. The reviewer still thinks the improvements are not important enough to claim significant novelty. Overall, I think the idea is simple and adds some structure to language modeling, but I also concur with the reviewer about limited improvements, which makes it a borderline paper. When calibrating with other area chairs, I decided to recommend to reject the paper."""
745,"""A fully automated periodicity detection in time series""","['Time series', 'feature engineering', 'period detection', 'machine learning']","""This paper presents a method to autonomously find periodicities in a signal. It is based on the same idea of using Fourier Transform and autocorrelation function presented in Vlachos et al. 2005. While showing interesting results this method does not perform well on noisy signals or signals with multiple periodicities. Thus, our method adds several new extra steps (hints clustering, filtering and detrending) to fix these issues. Experimental results show that the proposed method outperforms the state of the art algorithms. ""","""This paper presents an heuristic method to detect periodicity in a time-series such that it can handle noise and multiple periods. All reviewers agreed that this paper falls off the scope of ICLR since it does not discuss any learning-related question. Moreover, the authors did not provide any response nor updated manuscript addressing the reviewers remarks. The AC thus recommends rejection. """
746,"""Switching Linear Dynamics for Variational Bayes Filtering""","['sequence model', 'switching linear dynamical systems', 'variational bayes', 'filter', 'variational inference', 'stochastic recurrent neural network']","""System identification of complex and nonlinear systems is a central problem for model predictive control and model-based reinforcement learning. Despite their complexity, such systems can often be approximated well by a set of linear dynamical systems if broken into appropriate subsequences. This mechanism not only helps us find good approximations of dynamics, but also gives us deeper insight into the underlying system. Leveraging Bayesian inference and Variational Autoencoders, we show how to learn a richer and more meaningful state space, e.g. encoding joint constraints and collisions with walls in a maze, from partial and high-dimensional observations. This representation translates into a gain of accuracy of the learned dynamics which we showcase on various simulated tasks.""","""The overall view of the reviewers is that the paper is not quite good enough as it stands. The reviewers also appreciates the contributions so taking the comments into account and resubmit elsewhere is encouraged. """
747,"""Investigating CNNs' Learning Representation under label noise""","['learning with noisy labels', 'deep learning', 'convolutional neural networks']","""Deep convolutional neural networks (CNNs) are known to be robust against label noise on extensive datasets. However, at the same time, CNNs are capable of memorizing all labels even if they are random, which means they can memorize corrupted labels. Are CNNs robust or fragile to label noise? Much of researches focusing on such memorization uses class-independent label noise to simulate label corruption, but this setting is simple and unrealistic. In this paper, we investigate the behavior of CNNs under class-dependently simulated label noise, which is generated based on the conceptual distance between classes of a large dataset (i.e., ImageNet-1k). Contrary to previous knowledge, we reveal CNNs are more robust to such class-dependent label noise than class-independent label noise. We also demonstrate the networks under class-dependent noise situations learn similar representation to the no noise situation, compared to class-independent noise situations.""","""The paper analyzes the performance of CNN models when data is mislabelled in different manners. The reviewers and AC note the critical limitation of novelty of this paper to meet the high standard of ICLR. AC thinks the proposed method has potential and is interesting, but decided that the authors need more works to publish."""
748,"""Sample Efficient Deep Neuroevolution in Low Dimensional Latent Space""","['Neuroevolution', 'Reinforcement Learning']","""Current deep neuroevolution models are usually trained in a large parameter search space for complex learning tasks, e.g. playing video games, which needs billions of samples and thousands of search steps to obtain significant performance. This raises a question of whether we can make use of sequential data generated during evolution, encode input samples, and evolve in low dimensional parameter space with latent state input in a fast and efficient manner. Here we give an affirmative answer: we train a VAE to encode input samples, then an RNN to model environment dynamics and handle temporal information, and last evolve our low dimensional policy network in latent space. We demonstrate that this approach is surprisingly efficient: our experiments on Atari games show that within 10M frames and 30 evolution steps of training, our algorithm could achieve competitive result compared with ES, A3C, and DQN which need billions of frames.""","""Pros: - compelling idea to use VAEs to reduce the dimensionality of the space in which to run evolution - non-trivial benchmark results - clearly written, solid background Cons: - moderate novelty (as compared to [1]) - performance results are sup-par - no rebuttal, despite constructive and detailed review comments (and an explicit willingness to raise scores by multiple points!) The reviewers agree that the paper should be rejected in its current form, but would plausibly have been willing to reassess their scores for a major revision -- which did not materialize."""
749,"""Learning Representations of Categorical Feature Combinations via Self-Attention""","['Learning Representations', 'Feature Combinations', 'Self-Attention']","""Self-attention has been widely used to model the sequential data and achieved remarkable results in many applications. Although it can be used to model dependencies without regard to positions of sequences, self-attention is seldom applied to non-sequential data. In this work, we propose to learn representations of multi-field categorical data in prediction tasks via self-attention mechanism, where features are orderless but have intrinsic relations over different fields. In most current DNN based models, feature embeddings are simply concatenated for further processing by networks. Instead, by applying self-attention to transform the embeddings, we are able to relate features in different fields and automatically learn representations of their combinations, which are known as the factors of many prevailing linear models. To further improve the effect of feature combination mining, we modify the original self-attention structure by restricting the similarity weight to have at most k non-zero values, which additionally regularizes the model. We experimentally evaluate the effectiveness of our self-attention model on non-sequential data. Across two click through rate prediction benchmark datasets, i.e., Cretio and Avazu, our model with top-k restricted self-attention achieves the state-of-the-art performance. Compared with the vanilla MLP, the gain by adding self-attention is significantly larger than that by modifying the network structures, which most current works focus on.""","""All reviewers agree in their assessment that this paper is not ready for acceptance into ICLR and the authors did not respond during the rebuttal phase."""
750,"""Prob2Vec: Mathematical Semantic Embedding for Problem Retrieval in Adaptive Tutoring""","['personalized learning', 'e-learning', 'text embedding', 'Skip-gram', 'imbalanced data set', 'data level classification methods']","""We propose a new application of embedding techniques to problem retrieval in adaptive tutoring. The objective is to retrieve problems similar in mathematical concepts. There are two challenges: First, like sentences, problems helpful to tutoring are never exactly the same in terms of the underlying concepts. Instead, good problems mix concepts in innovative ways, while still displaying continuity in their relationships. Second, it is difficult for humans to determine a similarity score consistent across a large enough training set. We propose a hierarchical problem embedding algorithm, called Prob2Vec, that consists of an abstraction and an embedding step. Prob2Vec achieves 96.88\% accuracy on a problem similarity test, in contrast to 75\% from directly applying state-of-the-art sentence embedding methods. It is surprising that Prob2Vec is able to distinguish very fine-grained differences among problems, an ability humans need time and effort to acquire. In addition, the sub-problem of concept labeling with imbalanced training data set is interesting in its own right. It is a multi-label problem suffering from dimensionality explosion, which we propose ways to ameliorate. We propose the novel negative pre-training algorithm that dramatically reduces false negative and positive ratios for classification, using an imbalanced training data set.""","""I tend to agree with reviewers. This is a bit more of an applied type of work and does not lead to new insights in learning representations. Lack of technical novelty Dataset too small"""
751,"""Local Stability and Performance of Simple Gradient Penalty pseudo-formula -Wasserstein GAN""","['WGAN', 'gradient penalty', 'stability', 'measure valued differentiation']","""Wasserstein GAN(WGAN) is a model that minimizes the Wasserstein distance between a data distribution and sample distribution. Recent studies have proposed stabilizing the training process for the WGAN and implementing the Lipschitz constraint. In this study, we prove the local stability of optimizing the simple gradient penalty pseudo-formula -WGAN(SGP pseudo-formula -WGAN) under suitable assumptions regarding the equilibrium and penalty measure pseudo-formula . The measure valued differentiation concept is employed to deal with the derivative of the penalty terms, which is helpful for handling abstract singular measures with lower dimensional support. Based on this analysis, we claim that penalizing the data manifold or sample manifold is the key to regularizing the original WGAN with a gradient penalty. Experimental results obtained with unintuitive penalty measures that satisfy our assumptions are also provided to support our theoretical results.""","""All three reviewers expressed concerns about the assumptions made for the local stability analysis. The AC thus recommends ""revise and resubmit""."""
752,"""Three continual learning scenarios and a case for generative replay""","['continual learning', 'generative models', 'replay', 'distillation', 'variational autoencoder']","""Standard artificial neural networks suffer from the well-known issue of catastrophic forgetting, making continual or lifelong learning problematic. Recently, numerous methods have been proposed for continual learning, but due to differences in evaluation protocols it is difficult to directly compare their performance. To enable more meaningful comparisons, we identified three distinct continual learning scenarios based on whether task identity is known and, if it is not, whether it needs to be inferred. Performing the split and permuted MNIST task protocols according to each of these scenarios, we found that regularization-based approaches (e.g., elastic weight consolidation) failed when task identity needed to be inferred. In contrast, generative replay combined with distillation (i.e., using class probabilities as soft targets) achieved superior performance in all three scenarios. In addition, we reduced the computational cost of generative replay by integrating the generative model into the main model.""","""The authors have proposed 3 continual learning variants which are all based on MNIST and which vary in terms of whether task ids are given and what the classification task is, and they have proposed a method which incorporates a symmetric VAE for generative replay with a class discriminator. The proposed method does work well on the continual learning scenarios and the incorporation of the generative model with the classifier is more efficient than keeping them separate. The discussion of the different CL scenarios and of related work is nice to read. However, the authors imply that these scenarios cover the space of important CL variants, yet they do not consider many other settings, such as when tasks continually change rather than having sharp boundaries. The authors have also only focused on the catastrophic forgetting aspect of continual learning, without considering scenarios where, e.g., strong forward transfer (or backwards transfer) is very important. Regarding the proposed architecture that combines a VAE with a softmax classifier for efficiency, the reviewers all felt that this was not novel enough to recommend publication."""
753,"""Neural Persistence: A Complexity Measure for Deep Neural Networks Using Algebraic Topology""","['Algebraic topology', 'persistent homology', 'network complexity', 'neural network']","""While many approaches to make neural networks more fathomable have been proposed, they are restricted to interrogating the network with input data. Measures for characterizing and monitoring structural properties, however, have not been developed. In this work, we propose neural persistence, a complexity measure for neural network architectures based on topological data analysis on weighted stratified graphs. To demonstrate the usefulness of our approach, we show that neural persistence reflects best practices developed in the deep learning community such as dropout and batch normalization. Moreover, we derive a neural persistence-based stopping criterion that shortens the training process while achieving comparable accuracies as early stopping based on validation loss.""","""The paper presents a topological complexity measure of neural networks based on persistence 0-homology of the weights in each layer. Some lower and upper bounds of the p-norm persistence diagram are derived that leads to normalized persistence metric. The main discovery of such a topological complexity measure is that it leads to a stability-based early stopping criterion without a statistical cross-validation, as well as distinct characterizations on random initialization, batch normalization and drop out. Experiments are conducted with simple networks and MNIST, Fashion-MNIST, CIFAR10, IMDB datasets. The main concerns from the reviewers are that experimental studies are still preliminary and the understanding on the observed interesting phenomenon is premature. The authors make comprehensive responses to the raised questions with new experiments and some reviewers raise the rating. The reviewers all agree that the paper presents a novel study on neural network from an algebraic topology perspective with interesting results that has not been seen before. The paper is thus suggested to be borderline lean accept. """
754,"""Generative predecessor models for sample-efficient imitation learning""","['Imitation Learning', 'Generative Models', 'Deep Learning']","""We propose Generative Predecessor Models for Imitation Learning (GPRIL), a novel imitation learning algorithm that matches the state-action distribution to the distribution observed in expert demonstrations, using generative models to reason probabilistically about alternative histories of demonstrated states. We show that this approach allows an agent to learn robust policies using only a small number of expert demonstrations and self-supervised interactions with the environment. We derive this approach from first principles and compare it empirically to a state-of-the-art imitation learning method, showing that it outperforms or matches its performance on two simulated robot manipulation tasks and demonstrate significantly higher sample efficiency by applying the algorithm on a real robot.""","""This paper proposes to estimate the predecessor state dynamics for more sample-efficient imitation learning. While backward models have been used in the past in reinforcement learning, the application to imitation learning has not been previously studied. The paper is well-written and the results are good, showing clear improvements over GAIL in the presented experiments. The primary weakness of the paper is the lack of comparisons to the baselines suggested by reviewer 1 (a jumpy forward model and a single step predecessor model) to fully evaluate the contribution, and to SAIL and AIRL. Despite these weaknesses, the paper slightly exceeds the bar for acceptance at ICLR. The authors are strongly encouraged to include these comparisons in the final version."""
755,"""Representation Degeneration Problem in Training Natural Language Generation Models""","['Natural Language Processing', 'Representation Learning']","""We study an interesting problem in training neural network-based models for natural language generation tasks, which we call the \emph{representation degeneration problem}. We observe that when training a model for natural language generation tasks through likelihood maximization with the weight tying trick, especially with big training datasets, most of the learnt word embeddings tend to degenerate and be distributed into a narrow cone, which largely limits the representation power of word embeddings. We analyze the conditions and causes of this problem and propose a novel regularization method to address it. Experiments on language modeling and machine translation show that our method can largely mitigate the representation degeneration problem and achieve better performance than baseline algorithms.""","""although i (ac) believe the contribution is fairly limited (e.g., (1) only looking at the word embedding which goes through many nonlinear layers, in which case it's not even clear whether how word vectors are distributed matters much, (2) only considering the case of tied embeddings, which is not necessarily the most common setting, ...), all the reviewers found the execution of the submission (motivation, analysis and experimentation) to be done well, and i'll go with the reviewers' opinion."""
756,"""From Nodes to Networks: Evolving Recurrent Neural Networks""","['Recurrent neural networks', 'evolutionary algorithms', 'genetic programming']","""Gated recurrent networks such as those composed of Long Short-Term Memory (LSTM) nodes have recently been used to improve state of the art in many sequential processing tasks such as speech recognition and machine translation. However, the basic structure of the LSTM node is essentially the same as when it was first conceived 25 years ago. Recently, evolutionary and reinforcement learning mechanisms have been employed to create new variations of this structure. This paper proposes a new method, evolution of a tree-based encoding of the gated memory nodes, and shows that it makes it possible to explore new variations more effectively than other methods. The method discovers nodes with multiple recurrent paths and multiple memory cells, which lead to significant improvement in the standard language modeling benchmark task. Remarkably, this node did not perform well in another task, music modeling, but it was possible to evolve a different node that did, demonstrating that the approach discovers customized structure for each task. The paper also shows how the search process can be speeded up by training an LSTM network to estimate performance of candidate structures, and by encouraging exploration of novel solutions. Thus, evolutionary design of complex neural network structures promises to improve performance of deep learning architectures beyond human ability to do so.""","""In this work, the authors explore using genetic programming to search over network architectures. The reviewers noted that the proposed approach is simple and fast. However, the reviewers expressed concerns about the experimental validation (e.g., experiments were conducted on small tasks; issues with comparisons (cf. feedback from Reviewer2)), and the fact that the method were not compared against various baseline methods related to architecture search. """
757,"""Adversarial Attacks on Node Embeddings""","['node embeddings', 'adversarial attacks']","""The goal of network representation learning is to learn low-dimensional node embeddings that capture the graph structure and are useful for solving downstream tasks. However, despite the proliferation of such methods there is currently no study of their robustness to adversarial attacks. We provide the first adversarial vulnerability analysis on the widely used family of methods based on random walks. We derive efficient adversarial perturbations that poison the network structure and have a negative effect on both the quality of the embeddings and the downstream tasks. We further show that our attacks are transferable since they generalize to many models, and are successful even when the attacker is restricted.""","""The paper provides a novel analysis of the robustness to adversarial attacks in network representation learning. It appears to be a useful contribution for important class of models; however, the detailed reviews (1 and 2) raise some concerns that may require a bit of further work (though partially addressed in revised version)."""
758,"""DEFactor: Differentiable Edge Factorization-based Probabilistic Graph Generation""","['molecular graphs', 'conditional autoencoder', 'graph autoencoder']","""Generating novel molecules with optimal properties is a crucial step in many industries such as drug discovery. Recently, deep generative models have shown a promising way of performing de-novo molecular design. Although graph generative models are currently available they either have a graph size dependency in their number of parameters, limiting their use to only very small graphs or are formulated as a sequence of discrete actions needed to construct a graph, making the output graph non-differentiable w.r.t the model parameters, therefore preventing them to be used in scenarios such as conditional graph generation. In this work we propose a model for conditional graph generation that is computationally efficient and enables direct optimisation of the graph. We demonstrate favourable performance of our model on prototype-based molecular graph conditional generation tasks.""","""Since the reviewers unanimously recommended rejecting this paper, I am also recommending against publication. The paper considers an interesting problem and expresses some interesting modeling ideas. However, I concur with the reviewers that a more extensive and convincing set of experiments would be important to add. Especially important would be more experiments with simple extensions of previous approaches and much simpler models designed to solve one of the tasks directly, even if it is in an ad hoc way. If we assume that we only care about results, we should first make sure these particular benchmarks are difficult (this should not be too hard to establish more convincingly if it is true) and that obvious things to try do not work well."""
759,"""Causal importance of orientation selectivity for generalization in image recognition""","['deep learning', 'generalization', 'selectivity', 'neuroscience']","""Although both our brain and deep neural networks (DNNs) can perform high-level sensory-perception tasks such as image or speech recognition, the inner mechanism of these hierarchical information-processing systems is poorly understood in both neuroscience and machine learning. Recently, Morcos et al. (2018) examined the effect of class-selective units in DNNs, i.e., units with high-level selectivity, on network generalization, concluding that hidden units that are selectively activated by specific input patterns may harm the network's performance. In this study, we revisit their hypothesis, considering units with selectivity for lower-level features, and argue that selective units are not always harmful to the network performance. Specifically, by using DNNs trained for image classification (7-layer CNNs and VGG16 trained on CIFAR-10 and ImageNet, respectively), we analyzed the orientation selectivity of individual units. Orientation selectivity is a low-level selectivity widely studied in visual neuroscience, in which, when images of bars with several orientations are presented to the eye, many neurons in the visual cortex respond selectively to a specific orientation. We found that orientation-selective units exist in both lower and higher layers of these DNNs, as in our brain. In particular, units in the lower layers become more orientation-selective as the generalization performance improves during the course of training of the DNNs. Consistently, networks that generalize better are more orientation-selective in the lower layers. We finally reveal that ablating these selective units in the lower layers substantially degrades the generalization performance, at least by disrupting the shift-invariance of the higher layers. These results suggest to the machine-learning community that, contrary to the triviality of units with high-level selectivity, lower-layer units with selectivity for low-level features can be indispensable for generalization, and for neuroscientists, orientation selectivity can play a causally important role in object recognition.""","""The authors conduct experiments to study orientation selectivity in neural networks. The reviewers generally agreed that the paper was clearly written and easy to follow. Further, the experimental analysis demonstrates that contrary to what was claimed in some previous work, the learned orientation selectivity can be useful for generalization. However, the reviewers also raised a number of concerns: 1) that the conclusions are drawn on the basis of a couple of neural network architectures; the authors attempted to add results using a Resnet50 model, but this analysis was ultimately removed when the authors discovered a bug; 2) in the context of the contributions in neuroscience it was not clear that the limited results on the two artificial networks are sufficient to help draw such conclusions, and that 3) since the network is trained to recognize objects, it would seem natural that the model would learn neurons that are sensitive to orientation and that it is not clear how the authors observations might lead to better trained models. While the reviewers were not completely unanimous in their scores, the AC agrees with a majority of the reviewers that the work while interesting could be strengthened by additional experiments on other architectures. """
760,"""Deep Neuroevolution: Genetic Algorithms are a Competitive Alternative for Training Deep Neural Networks for Reinforcement Learning""","['Neuroevolution', 'Reinforcement Learning']","""Deep artificial neural networks (DNNs) are typically trained via gradient-based learning algorithms, namely backpropagation. Evolution strategies (ES) can rival backprop-based algorithms such as Q-learning and policy gradients on challenging deep reinforcement learning (RL) problems. However, ES can be considered a gradient-based algorithm because it performs stochastic gradient descent via an operation similar to a finite-difference approximation of the gradient. That raises the question of whether non-gradient-based evolutionary algorithms can work at DNN scales. Here we demonstrate they can: we evolve the weights of a DNN with a simple, gradient-free, population-based genetic algorithm (GA) and it performs well on hard deep RL problems, including Atari and humanoid locomotion. The Deep GA successfully evolves networks with over four million free parameters, the largest neural networks ever evolved with a traditional evolutionary algorithm. These results (1) expand our sense of the scale at which GAs can operate, (2) suggest intriguingly that in some cases following the gradient is not the best choice for optimizing performance, and (3) make immediately available the multitude of neuroevolution techniques that improve performance. We demonstrate the latter by showing that combining DNNs with novelty search, which encourages exploration on tasks with deceptive or sparse reward functions, can solve a high-dimensional problem on which reward-maximizing algorithms (e.g.\ DQN, A3C, ES, and the GA) fail. Additionally, the Deep GA is faster than ES, A3C, and DQN (it can train Atari in {\raise.17ex\hbox{ pseudo-formula }}4 hours on one workstation or {\raise.17ex\hbox{ pseudo-formula }}1 hour distributed on 720 cores), and enables a state-of-the-art, up to 10,000-fold compact encoding technique. ""","""This paper presents an empirical study of the applicability of genetic algorithms to deep RL problems. Major concerns of the paper include: 1. paper organization, especially the presentation of the results, is hard to follow; 2. the results are not strong enough to support that claims made in this paper, as GAs are currently not strong enough when compared to the SOTA RL algorithms; 3. Not quite clear why or when GAs are better than RL or ES; Lack of insights. Overall, this paper cannot be accepted yet. """
761,"""Sentence Encoding with Tree-Constrained Relation Networks""","['sentence encoder', 'relation networks', 'tree', 'machine translation']","""The meaning of a sentence is a function of the relations that hold between its words. We instantiate this relational view of semantics in a series of neural models based on variants of relation networks (RNs) which represent a set of objects (for us, words forming a sentence) in terms of representations of pairs of objects. We propose two extensions to the basic RN model for natural language. First, building on the intuition that not all word pairs are equally informative about the meaning of a sentence, we use constraints based on both supervised and unsupervised dependency syntax to control which relations influence the representation. Second, since higher-order relations are poorly captured by a sum of pairwise relations, we use a recurrent extension of RNs to propagate information so as to form representations of higher order relations. Experiments on sentence classification, sentence pair classification, and machine translation reveal that, while basic RNs are only modestly effective for sentence representation, recurrent RNs with latent syntax are a reliably powerful representational device.""","""This paper presents two extensions of Relation Networks (RNs) to represent a sentence as a set of relations between words: (1) dependency-based constraints to control the influence of different relations within a sentence and (2) recurrent extension of RNs to propagate information through the tree structure of relations. Pros: The notion of relation networks for sentence representation is potentially interesting. Cons: The significance of the proposed methods compared to existing variants of TreeRNNs is not clear (R1). R1 requested empirical comparisons against TreeRNNs (since the proposed methods are also of tree shape), but the authors argued back that such experiments are necessary beyond BiLSTM baselines. Verdict: Reject. The proposed methods build on relatively incremental ideas and the empirical results are rather inconclusive."""
762,"""On the Computational Inefficiency of Large Batch Sizes for Stochastic Gradient Descent""","['Deep learning', 'large batch training', 'scaling rules', 'stochastic gradient descent']","""Increasing the mini-batch size for stochastic gradient descent offers significant opportunities to reduce wall-clock training time, but there are a variety of theoretical and systems challenges that impede the widespread success of this technique (Daset al., 2016; Keskar et al., 2016). We investigate these issues, with an emphasis on time to convergence and total computational cost, through an extensive empirical analysis of network training across several architectures and problem domains, including image classification, image segmentation, and language modeling. Although it is common practice to increase the batch size in order to fully exploit available computational resources, we find a substantially more nuanced picture. Our main finding is that across a wide range of network architectures and problem domains, increasing the batch size beyond a certain point yields no decrease in wall-clock time to convergence for either train or test loss. This batch size is usually substantially below the capacity of current systems. We show that popular training strategies for large batch size optimization begin to fail before we can populate all available compute resources, and we show that the point at which these methods break down depends more on attributes like model architecture and data complexity than it does directly on the size of the dataset.""","""The paper presents an interesting empirical analysis showing that increasing the batch size beyond a certain point yields no decrease in time to convergence. This is an interesting finding, since it indicates that parallelisation approaches might have their limits. On the other hand, the study does not allow the practitioners to tune their hyperparamters since the optimal batch size is dependent on the model architecture and the dataset. Furthermore, as also pointed out in an anonymous comment, the batch size is VERY large compared to the size of the benchmark sets. Therefore, it would be nice to see if the observation carries over to large-scale data sets, where the number of samples in the mini-batch is still small compared to the total number of samples. """
763,"""Pixel Chem: A Representation for Predicting Material Properties with Neural Network""","['material property prediction', 'neural network', 'material structure representation', 'chemistry']","""In this work we developed a new representation of the chemical information for the machine learning models, with benefits from both the real space (R-space) and energy space (K-space). Different from the previous symmetric matrix presentations, the charge transfer channel based on Paulings electronegativity is derived from the dependence on real space distance and orbitals for the hetero atomic structures. This representation can work for the bulk materials as well as the low dimensional nano materials, and can map the R-space and K-space into the pixel space (P-space) by training and testing 130k structures. P-space can well reproduce the R-space quantities within error 0.53. This new asymmetric matrix representation double the information storage than the previous symmetric representations.This work provides a new dimension for the computational chemistry towards the machine learning architecture. ""","""I would like to highlight to the PCs that reviewers highlighted clear evidence of plagiarism from prior work, which I was able to easily verify (a full paragraph of text was copied, word-for-word, from a paper describing one of the baselines the current work compares against). Further, all reviewers unanimously agreed that the paper was poorly written, and contains no useful advances for the ICLR audience. I recommend a rejection, and further, examination by the PCs of the conduct of the authors."""
764,"""BlackMarks: Black-box Multi-bit Watermarking for Deep Neural Networks""","['Digital Watermarking', 'IP Protection', 'Deep Neural Networks']","""Deep Neural Networks (DNNs) are increasingly deployed in cloud servers and autonomous agents due to their superior performance. The deployed DNN is either leveraged in a white-box setting (model internals are publicly known) or a black-box setting (only model outputs are known) depending on the application. A practical concern in the rush to adopt DNNs is protecting the models against Intellectual Property (IP) infringement. We propose BlackMarks, the first end-to-end multi-bit watermarking framework that is applicable in the black-box scenario. BlackMarks takes the pre-trained unmarked model and the owners binary signature as inputs. The output is the corresponding marked model with specific keys that can be later used to trigger the embedded watermark. To do so, BlackMarks first designs a model-dependent encoding scheme that maps all possible classes in the task to bit 0 and bit 1. Given the owners watermark signature (a binary string), a set of key image and label pairs is designed using targeted adversarial attacks. The watermark (WM) is then encoded in the distribution of output activations of the DNN by fine-tuning the model with a WM-specific regularized loss. To extract the WM, BlackMarks queries the model with the WM key images and decodes the owners signature from the corresponding predictions using the designed encoding scheme. We perform a comprehensive evaluation of BlackMarks performance on MNIST, CIFAR-10, ImageNet datasets and corroborate its effectiveness and robustness. BlackMarks preserves the functionality of the original DNN and incurs negligible WM embedding overhead as low as 2.054%.""","""The reviews agree the paper is not ready for publication at ICLR. """
765,"""Differentiable Learning-to-Normalize via Switchable Normalization""","['normalization', 'deep learning', 'CNN', 'computer vision']","""We address a learning-to-normalize problem by proposing Switchable Normalization (SN), which learns to select different normalizers for different normalization layers of a deep neural network. SN employs three distinct scopes to compute statistics (means and variances) including a channel, a layer, and a minibatch. SN switches between them by learning their importance weights in an end-to-end manner. It has several good properties. First, it adapts to various network architectures and tasks (see Fig.1). Second, it is robust to a wide range of batch sizes, maintaining high performance even when small minibatch is presented (e.g. 2 images/GPU). Third, SN does not have sensitive hyper-parameter, unlike group normalization that searches the number of groups as a hyper-parameter. Without bells and whistles, SN outperforms its counterparts on various challenging benchmarks, such as ImageNet, COCO, CityScapes, ADE20K, and Kinetics. Analyses of SN are also presented. We hope SN will help ease the usage and understand the normalization techniques in deep learning. The code of SN will be released.""","""This paper proposes Switchable Normalization (SN) that leans how to combine three existing normalization techniques for improved performance. There is a general consensus that that the paper has good quality and clarity, is well motivated, is sufficiently novel, makes clear contributions for training deep neural networks, and provides convincing experimental results to show the advantages of the proposed SN."""
766,"""Improving On-policy Learning with Statistical Reward Accumulation""",[],"""Deep reinforcement learning has obtained significant breakthroughs in recent years. Most methods in deep-RL achieve good results via the maximization of the reward signal provided by the environment, typically in the form of discounted cumulative returns. Such reward signals represent the immediate feedback of a particular action performed by an agent. However, tasks with sparse reward signals are still challenging to on-policy methods. In this paper, we introduce an effective characterization of past reward statistics (which can be seen as long-term feedback signals) to supplement this immediate reward feedback. In particular, value functions are learned with multi-critics supervision, enabling complex value functions to be more easily approximated in on-policy learning, even when the reward signals are sparse. We also introduce a novel exploration mechanism called ``hot-wiring'' that can give a boost to seemingly trapped agents. We demonstrate the effectiveness of our advantage actor multi-critic (A2MC) method across the discrete domains in Atari games as well as continuous domains in the MuJoCo environments. A video demo is provided at pseudo-url and source codes will be made available upon paper acceptance.""","""The paper proposes an interesting idea for efficient exploration of on-policy learning in sparse reward RL problems. The empirical results are promising, which is the main strength of the paper. On the other hand, reviewers generally feel that the proposed algorithm is rather ad hoc, sometimes with not-so-transparent algorithmic choices. As a result, it is really unclear whether the idea works only on the test problems, or applies to a broader set of problems. The author responses and new results are helpful and appreciated by all reviewers, but do not change the reviewers' concerns."""
767,"""Adaptive Mixture of Low-Rank Factorizations for Compact Neural Modeling""","['Low-Rank Factorization', 'Compact Neural Nets', 'Efficient Modeling', 'Mixture models']","""Modern deep neural networks have a large amount of weights, which make them difficult to deploy on computation constrained devices such as mobile phones. One common approach to reduce the model size and computational cost is to use low-rank factorization to approximate a weight matrix. However, performing standard low-rank factorization with a small rank can hurt the model expressiveness and significantly decrease the performance. In this work, we propose to use a mixture of multiple low-rank factorizations to model a large weight matrix, and the mixture coefficients are computed dynamically depending on its input. We demonstrate the effectiveness of the proposed approach on both language modeling and image classification tasks. Experiments show that our method not only improves the computation efficiency but also maintains (sometimes outperforms) its accuracy compared with the full-rank counterparts.""","""The paper is clearly written and well motivated, but there are remaining concerns on contributions and comparisons. The paper received mixed initial reviews. After extensive discussions, while the authors successfully clarified several important issues (such as computation efficiency w.r.t splitting) pointed out by Reviewer 4 (an expert in the field), they were not able to convince him/her about the significance of the proposed network compression method. Reviewer 4 has the following remaining concerns: 1) This is a typical paper showing only FLOPs reduction but with an intent of real-time acceleration. However, wall-clock speedup is different from FLOPs reduction. It may not be beneficial to change the current computing flow optimized in modern software/hardware. This is one of major reasons why the reported wall-clock time even slows down. The problem may be alleviated with optimization efforts on software or hardware, then it is unclear how good/worse will it be when compared with fine-grain pruning solutions (Han et al. 2015b, Han et al. 2016 & Han et al. 2017), which achieved a higher FLOP reduction and a great wall-clock speedup with hardware optimized (using ASIC and FPGA); 2) If it is OK to target on FLOPs reduction (without comparison with fine-grain pruning solutions), 2.1) In LSTM experiments, the major producer of FLOPs -- the output layer, is excluded and this exclusion was hidden in the first version. Although the author(s) claimed that an output layer could be compressed, it is not shown in the paper. Compressing output layer will reduce model capacity, making other layers more difficult being compressed. 2.2) In CNN experiments, the improvements of CIFAR-10 is within a random range and not statistically significant. In table 2, ""Regular low-rank MobileNet"" improves the original MobileNet, showing that the original MobileNet (an arXiv paper) is not well designed. ""Adaptive Low-rank MobileNet"" improves accuracy upon ""Regular low-rank MobileNet"", but using 0.3M more parameters. The trade-off is unclear. In addition to these remaining concerns of Reviewer 4, the AC feels that the paper essentially modifies the original network structure in a very specific way: adding a particular nonlinear layer between two adjacent layers. Thus it seems a little bit unfair to mainly use low-rank factorization (which can be considered as a compression technique that barely changes the network architecture) for comparison. Adding comparisons with fine-grain pruning solutions (Han et al. 2015b, Han et al. 2016 & Han et al. 2017) and a large number of more recent related references inspired by the low-rank baseline (M. Jaderberg et al 2014) , as listed by Reviewer 4, will make the proposed method much more convincing. """
768,"""Learning to Progressively Plan""",[],"""For problem solving, making reactive decisions based on problem description is fast but inaccurate, while search-based planning using heuristics gives better solutions but could be exponentially slow. In this paper, we propose a new approach that improves an existing solution by iteratively picking and rewriting its local components until convergence. The rewriting policy employs a neural network trained with reinforcement learning. We evaluate our approach in two domains: job scheduling and expression simplification. Compared to common effective heuristics, baseline deep models and search algorithms, our approach efficiently gives solutions with higher quality.""","""This paper provides a new approach for progressive planning on discrete state and action spaces. The authors use LSTM architectures to iteratively select and improve local segments of an existing plan. They formulate the rewriting task as a reinforcement learning problem where the action space is the application of a set of possible rewriting rules. These models are then evaluated on a simulated job scheduling dataset and Halide expression simplification. This is an interesting paper dealing with an important problem. The proposed solution based on combining several existing pieces is novel. On the negative side, the reviewers thought the writing could be improved, and the main ideas are not explained clearly. Furthermore, the experimental evaluation is weak."""
769,"""Unsupervised Meta-Learning for Reinforcement Learning""","['Meta-Learning', 'Reinforcement Learning', 'Exploration', 'Unsupervised']","""Meta-learning is a powerful tool that learns how to quickly adapt a model to new tasks. In the context of reinforcement learning, meta-learning algorithms can acquire reinforcement learning procedures to solve new problems more efficiently by meta-learning prior tasks. The performance of meta-learning algorithms critically depends on the tasks available for meta-training: in the same way that supervised learning algorithms generalize best to test points drawn from the same distribution as the training points, meta-learning methods generalize best to tasks from the same distribution as the meta-training tasks. In effect, meta-reinforcement learning offloads the design burden from algorithm design to task design. If we can automate the process of task design as well, we can devise a meta-learning algorithm that is truly automated. In this work, we take a step in this direction, proposing a family of unsupervised meta-learning algorithms for reinforcement learning. We describe a general recipe for unsupervised meta-reinforcement learning, and describe an effective instantiation of this approach based on a recently proposed unsupervised exploration technique and model-agnostic meta-learning. We also discuss practical and conceptual considerations for developing unsupervised meta-learning methods. Our experimental results demonstrate that unsupervised meta-reinforcement learning effectively acquires accelerated reinforcement learning procedures without the need for manual task design, significantly exceeds the performance of learning from scratch, and even matches performance of meta-learning methods that use hand-specified task distributions.""","""This paper introduces unsupervised meta-learning algorithms for RL. Major concerns of the paper include: 1. Lack of clarity. The presentation of the method can be improved. 2. The motivation and justification of applying unsupervised meta-learning needs to be strengthened. More discussions and better motivating examples may be useful. 3. Experimental details are not sufficient and comparisons may not be sufficient to support the aim. Overall, this paper cannot be accepted yet. """
770,"""A Closer Look at Few-shot Classification""","['few shot classification', 'meta-learning']","""Few-shot classication aims to learn a classier to recognize unseen classes during training with limited labeled examples. While signicant progress has been made, the growing complexity of network designs, meta-learning algorithms, and differences in implementation details make a fair comparison difcult. In this paper, we present 1) a consistent comparative analysis of several representative few-shot classication algorithms, with results showing that deeper backbones signicantly reduce the gap across methods including the baseline, 2) a slightly modied baseline method that surprisingly achieves competitive performance when compared with the state-of-the-art on both the mini-ImageNet and the CUB datasets, and 3) a new experimental setting for evaluating the cross-domain generalization ability for few-shot classication algorithms. Our results reveal that reducing intra-class variation is an important factor when the feature backbone is shallow, but not as critical when using deeper backbones. In a realistic, cross-domain evaluation setting, we show that a baseline method with a standard ne-tuning practice compares favorably against other state-of-the-art few-shot learning algorithms.""","""This paper provides a number of interesting experiments for few-shot learning using the CUB and miniImagenet datasets. One of the especially intriguing experiments is the analysis of backbone depth in the architecture, as it relates to few-shot performance. The strong performance of the baseline and baseline++ are quite surprising. Overall the reviewers agree that this paper raises a number of questions about current few-shot learning approaches, especially how they relate to architecture and dataset characteristics. A few minor comments: - In table 1, matching nets are mistakenly attributed to Ravi and Larochelle. Should be Vinyals et al. - The notation for cosine similarity in section 3.2 is odd. It looks like youre computing some cosine function of two vectors which doesnt make sense. Please clarify this. - There are a few results that were promised after the revision deadline, please be sure to include these in the final draft. """
771,"""Learning to Learn without Forgetting by Maximizing Transfer and Minimizing Interference""",[],"""Lack of performance when it comes to continual learning over non-stationary distributions of data remains a major challenge in scaling neural network learning to more human realistic settings. In this work we propose a new conceptualization of the continual learning problem in terms of a temporally symmetric trade-off between transfer and interference that can be optimized by enforcing gradient alignment across examples. We then propose a new algorithm, Meta-Experience Replay (MER), that directly exploits this view by combining experience replay with optimization based meta-learning. This method learns parameters that make interference based on future gradients less likely and transfer based on future gradients more likely. We conduct experiments across continual lifelong supervised learning benchmarks and non-stationary reinforcement learning environments demonstrating that our approach consistently outperforms recently proposed baselines for continual learning. Our experiments show that the gap between the performance of MER and baseline algorithms grows both as the environment gets more non-stationary and as the fraction of the total experiences stored gets smaller. ""","""Pros: - novel method for continual learning - clear, well written - good results - no need for identified tasks - detailed rebuttal, new results in revision Cons: - experiments could be on more realistic/challenging domains The reviewers agree that the paper should be accepted."""
772,"""Provable Guarantees on Learning Hierarchical Generative Models with Deep CNNs""","['deep learning', 'theory']","""Learning deep networks is computationally hard in the general case. To show any positive theoretical results, one must make assumptions on the data distribution. Current theoretical works often make assumptions that are very far from describing real data, like sampling from Gaussian distribution or linear separability of the data. We describe an algorithm that learns convolutional neural network, assuming the data is sampled from a deep generative model that generates images level by level, where lower resolution images correspond to latent semantic classes. We analyze the convergence rate of our algorithm assuming the data is indeed generated according to this model (as well as additional assumptions). While we do not pretend to claim that the assumptions are realistic for natural images, we do believe that they capture some true properties of real data. Furthermore, we show that on CIFAR-10, the algorithm we analyze achieves results in the same ballpark with vanilla convolutional neural networks that are trained with SGD.""","""This manuscript proposes a generative model for images, then proposes a training procedure for fitting a convolutional neural network based on this model. One novelty if this result is that the generative procedure seems to be more complex than generative assumptions required for previous work. It is clear that the problem addressed -- training methods that may improve on SGD, with convergence guarantees -- is of significant interest to the community. The reviewers and AC note several issue (i) the initial version of the manuscript includes several assumptions that are not clearly stated. This seems to have been fixed in the updated manuscript (ii) reviewers suspect that the accumulation of stated assumptions may result in an easily separable generative model -- limiting the generality of the results (iii) experiemental results are underwhelming, and only comparable to much older published results."""
773,"""Discriminator Rejection Sampling""","['GANs', 'rejection sampling']","""We propose a rejection sampling scheme using the discriminator of a GAN to approximately correct errors in the GAN generator distribution. We show that under quite strict assumptions, this will allow us to recover the data distribution exactly. We then examine where those strict assumptions break down and design a practical algorithmcalled Discriminator Rejection Sampling (DRS)that can be used on real data-sets. Finally, we demonstrate the efficacy of DRS on a mixture of Gaussians and on the state of the art SAGAN model. On ImageNet, we train an improved baseline that increases the best published Inception Score from 52.52 to 62.36 and reduces the Frechet Inception Distance from 18.65 to 14.79. We then use DRS to further improve on this baseline, improving the Inception Score to 76.08 and the FID to 13.75.""","""The paper proposes a discriminator dependent rejection sampling scheme for improving the quality of samples from a trained GAN. The paper is clearly written, presents an interesting idea and the authors extended and improved the experimental analyses as suggested by the reviewers."""
774,"""Beyond Games: Bringing Exploration to Robots in Real-world""","['Exploration', 'curiosity', 'manipulation']","""Exploration has been a long standing problem in both model-based and model-free learning methods for sensorimotor control. While there has been major advances over the years, most of these successes have been demonstrated in either video games or simulation environments. This is primarily because the rewards (even the intrinsic ones) are non-differentiable since they are function of the environment (which is a black-box). In this paper, we focus on the policy optimization aspect of the intrinsic reward function. Specifically, by using a local approximation, we formulate intrinsic reward as a differentiable function so as to perform policy optimization using likelihood maximization -- much like supervised learning instead of reinforcement learning. This leads to a significantly sample efficient exploration policy. Our experiments clearly show that our approach outperforms both on-policy and off-policy optimization approaches like REINFORCE and DQN respectively. But most importantly, we are able to implement an exploration policy on a robot which learns to interact with objects completely from scratch just using data collected via the differentiable exploration module. See project videos at pseudo-url""","""The authors propose implementing intrinsic motivation as a differentiable supervised loss coming from the error of a forward model, rather than the black box style of curiosity reward. The motivation is that this approach will lead to more sample efficient exploration for real robots. The use of a differentiable loss for policy optimization is interesting and has some novelty. However, the reviewers were unanimous in their criticism of the paper for poor baselines, unclear experiments and results, and unsupported claims. Even after substantial revisions to the paper, the AC and reviewers were unconvinced of the basic claims of the paper."""
775,"""Contextual Recurrent Convolutional Model for Robust Visual Learning""","['contextual modulation', 'recurrent convolutional network', 'robust visual learning']","""Feedforward convolutional neural network has achieved a great success in many computer vision tasks. While it validly imitates the hierarchical structure of biological visual system, it still lacks one essential architectural feature: contextual recurrent connections with feedback, which widely exists in biological visual system. In this work, we designed a Contextual Recurrent Convolutional Network with this feature embedded in a standard CNN structure. We found that such feedback connections could enable lower layers to ``rethink"" about their representations given the top-down contextual information. We carefully studied the components of this network, and showed its robustness and superiority over feedforward baselines in such tasks as noise image classification, partially occluded object recognition and fine-grained image classification. We believed this work could be an important step to help bridge the gap between computer vision models and real biological visual system.""","""This paper explores the addition of feedback connections to popular CNN architectures. All three reviewers suggest rejecting the paper, pointing to limited novelty with respect to other recent publications, and unconvincing experiments. The AC agrees with the reviewers. """
776,"""Infinitely Deep Infinite-Width Networks""","['Infinite-width networks', 'initialisation', 'kernel methods', 'reproducing kernel Hilbert spaces', 'Gaussian processes']","""Infinite-width neural networks have been extensively used to study the theoretical properties underlying the extraordinary empirical success of standard, finite-width neural networks. Nevertheless, until now, infinite-width networks have been limited to at most two hidden layers. To address this shortcoming, we study the initialisation requirements of these networks and show that the main challenge for constructing them is defining the appropriate sampling distributions for the weights. Based on these observations, we propose a principled approach to weight initialisation that correctly accounts for the functional nature of the hidden layer activations and facilitates the construction of arbitrarily many infinite-width layers, thus enabling the construction of arbitrarily deep infinite-width networks. The main idea of our approach is to iteratively reparametrise the hidden-layer activations into appropriately defined reproducing kernel Hilbert spaces and use the canonical way of constructing probability distributions over these spaces for specifying the required weight distributions in a principled way. Furthermore, we examine the practical implications of this construction for standard, finite-width networks. In particular, we derive a novel weight initialisation scheme for standard, finite-width networks that takes into account the structure of the data and information about the task at hand. We demonstrate the effectiveness of this weight initialisation approach on the MNIST, CIFAR-10 and Year Prediction MSD datasets.""","""The paper studies how to construct infinitely deep infinite-width networks from a theoretical point of view, and uses the results of its theoretical analysis to design a weight initialization scheme for finite-width networks. While the idea is interesting and the paper may contain novel theoretical contributions, the experimental results are weak, as pointed out by all three reviewers from several different perspectives. In particular, it seems that the presented theoretical analysis is useful mainly for weight initialization and hence has limited potential impacts. In addition, the authors have responded to neither the AC's question, nor a detailed anonymous comment that challenges the value of Proposition 1 given the previous work by Aronszajn."""
777,"""Multiple Encoder-Decoders Net for Lane Detection""",[],"""For semantic image segmentation and lane detection, nets with a single spatial pyramid structure or encoder-decoder structure are usually exploited. Convolutional neural networks (CNNs) show great results on both high-level and low-level features representations, however, the capability has not been fully embodied for lane detection task. In especial, it's still a challenge for model-based lane detection to combine the multi-scale context with a pixel-level accuracy because of the weak visual appearance and strong prior information. In this paper, we we propose an novel network for lane detection, the three main contributions are as follows. First, we employ multiple encoder-decoders module in end-to-end ways and show the promising results for lane detection. Second, we analysis different configurations of multiple encoder-decoders nets. Third, we make our attempts to rethink the evaluation methods of lane detection for the limitation of the popular methods based on IoU.""","""As the reviewers point out, the paper is below the acceptance standard of ICLR due to low novelty, unclear presentation, and lack of experimental comparison against the state-of-the-art baselines."""
778,"""NSGA-Net: A Multi-Objective Genetic Algorithm for Neural Architecture Search""","['neural architecture search', 'evolutionary algorithms']","""This paper introduces NSGA-Net, an evolutionary approach for neural architecture search (NAS). NSGA-Net is designed with three goals in mind: (1) a NAS procedure for multiple, possibly conflicting, objectives, (2) efficient exploration and exploitation of the space of potential neural network architectures, and (3) output of a diverse set of network architectures spanning a trade-off frontier of the objectives in a single run. NSGA-Net is a population-based search algorithm that explores a space of potential neural network architectures in three steps, namely, a population initialization step that is based on prior-knowledge from hand-crafted architectures, an exploration step comprising crossover and mutation of architectures and finally an exploitation step that applies the entire history of evaluated neural architectures in the form of a Bayesian Network prior. Experimental results suggest that combining the objectives of minimizing both an error metric and computational complexity, as measured by FLOPS, allows NSGA-Net to find competitive neural architectures near the Pareto front of both objectives on two different tasks, object classification and object alignment. NSGA-Net obtains networks that achieve 3.72% (at 4.5 million FLOP) error on CIFAR-10 classification and 8.64% (at 26.6 million FLOP) error on the CMU-Car alignment task.""","""Pros: - an explicitly multi-objective approach to neural architecture search - multiple datasets - ablation experiments Cons: - lack of baselines like hyperparameter search - ill-justified increase in capacity after search - ineffective use of the multiple objectives in assessment - not clearly beating random search baseline The reviewers adjusted their scores upward after the rebuttal, but serious concerns remain, and the consensus is still to (borderline) reject the paper."""
779,"""Learning and Planning with a Semantic Model""","['deep reinforcement learning', 'generalization', 'semantic structure', 'model-based']","""Building deep reinforcement learning agents that can generalize and adapt to unseen environments remains a fundamental challenge for AI. This paper describes progresses on this challenge in the context of man-made environments, which are visually diverse but contain intrinsic semantic regularities. We propose a hybrid model-based and model-free approach, LEArning and Planning with Semantics (LEAPS), consisting of a multi-target sub-policy that acts on visual inputs, and a Bayesian model over semantic structures. When placed in an unseen environment, the agent plans with the semantic model to make high-level decisions, proposes the next sub-target for the sub-policy to execute, and updates the semantic model based on new observations. We perform experiments in visual navigation tasks using House3D, a 3D environment that contains diverse human-designed indoor scenes with real-world objects. LEAPS outperforms strong baselines that do not explicitly plan using the semantic content.""","""The paper presents LEAPS, a hybrid model-based and model-free algorithm that uses a Bayesian approach to reason/plan over semantic features, while low level behavior is learned in a model-free manner. The approach is designed for human-made environments with semantic similarity, such as indoor navigation, and is empirically validated in a virtual indoor navigation task, House3D. Reviewers and AC note the interesting approach to this challenging problem. The presented approach can provide an elegant way to incorporate domain knowledge into RL approaches. The reviewers and AC note several potential weaknesses. The reviewers are concerned about the very low success rate, and critiqued the use of success rate as a key metric itself, given that random search with a sufficiently high cut-off could solve the task. The authors added additional results in a metric that incorporates path length, and provided clarifying details. However, key concerns remained given the low success rates. The AC notes that e.g., results in the top and middle row of figure 4 show very similar results for LEAPS and the reported baselines. Further, ""figure 5"" shows no confidence / error bars, and it is not possible to assess whether any differences are statistically significant. Overall, the questions of whether something substantial has been learned, should be addressed with a detailed error analysis of the proposed approach and the baselines, to provide insight into whether and how the approaches solve the task. At the moment, the paper presents a potentially valuable approach, but does not provide convincing evidence and conceptual insights into the approach's effectiveness."""
780,"""Unsupervised Image to Sequence Translation with Canvas-Drawer Networks""","['image', 'translation', 'unsupervised', 'model-based']","""Encoding images as a series of high-level constructs, such as brush strokes or discrete shapes, can often be key to both human and machine understanding. In many cases, however, data is only available in pixel form. We present a method for generating images directly in a high-level domain (e.g. brush strokes), without the need for real pairwise data. Specifically, we train a canvas network to imitate the mapping of high-level constructs to pixels, followed by a high-level drawing network which is optimized through this mapping towards solving a desired image recreation or translation task. We successfully discover sequential vector representations of symbols, large sketches, and 3D objects, utilizing only pixel data. We display applications of our method in image segmentation, and present several ablation studies comparing various configurations.""","""This paper was reviewed by three experts. After the author response, R2 and R3 recommend rejecting this paper citing concerns of experimental evaluation and poor quality of the manuscript. All three reviewers continue to have questions for the authors, which the authors have not responded to. The AC finds no basis for accepting this paper in this state. """
781,"""CHEMICAL NAMES STANDARDIZATION USING NEURAL SEQUENCE TO SEQUENCE MODEL""","['Chemical Names Standardization', 'Byte Pair Encoding', 'Sequence to Sequence Model']","""Chemical information extraction is to convert chemical knowledge in text into true chemical database, which is a text processing task heavily relying on chemical compound name identification and standardization. Once a systematic name for a chemical compound is given, it will naturally and much simply convert the name into the eventually required molecular formula. However, for many chemical substances, they have been shown in many other names besides their systematic names which poses a great challenge for this task. In this paper, we propose a framework to do the auto standardization from the non-systematic names to the corresponding systematic names by using the spelling error correction, byte pair encoding tokenization and neural sequence to sequence model. Our framework is trained end to end and is fully data-driven. Our standardization accuracy on the test dataset achieves 54.04% which has a great improvement compared to previous state-of-the-art result.""","""The area chair agrees with reviewer 1 and 2 that this paper does not have sufficient machine learning novelty for ICLR. This is competent work and the problem is interesting, but ICLR is not the right venue since the main contributions are on defining the task. All the models that are then applied are standard."""
782,"""SIMILE: Introducing Sequential Information towards More Effective Imitation Learning""","['Reinforcement Learning', 'Imitation Learning', 'Sequential Information']","""Reinforcement learning (RL) is a metaheuristic aiming at teaching an agent to interact with an environment and maximizing the reward in a complex task. RL algorithms often encounter the difficulty in defining a reward function in a sparse solution space. Imitation learning (IL) deals with this issue by providing a few expert demonstrations, and then either mimicking the expert's behavior (behavioral cloning, BC) or recovering the reward function by assuming the optimality of the expert (inverse reinforcement learning, IRL). Conventional IL approaches formulate the agent policy by mapping one single state to a distribution over actions, which did not consider sequential information. This strategy can be less accurate especially in IL, a weakly supervised learning environment, especially when the number of expert demonstrations is limited. This paper presents an effective approach named Sequential IMItation LEarning (SIMILE). The core idea is to introduce sequential information, so that an agent can refer to both the current state and past state-action pairs to make a decision. We formulate our approach into a recurrent model, and instantiate it using LSTM so as to fuse both long-term and short-term information. SIMILE is a generalized IL framework which is easily applied to BL and IRL, two major types of IL algorithms. Experiments are performed on several robot controlling tasks in OpenAI Gym. SIMILE not only achieves performance gain over the baseline approaches, but also enjoys the benefit of faster convergence and better stability of testing performance. These advantages verify a higher learning efficiency of SIMILE, and implies its potential applications in real-world scenarios, i.e., when the agent-environment interaction is more difficult and/or expensive.""","""This paper explores the use of sequential information to improve imitation learning, essentially using recurrent networks (LSTM) instead of a simple NN in several existing imitation learning models (BC, GAIL, etc.). On the positive side, the empirical results are good, showing improvement in terms of attained rewards, convergence speed and stability. There are however some significant issues with the way the way the approach is motivated and positioned with respect to existsing work. In particular, the issue described in the paper is due to the fact they consider POMDPs (not MDPs): this should have been more clearly explained. There are also issues with the Related Work section. For these reasons, the paper is not quite ready for publication. """
783,"""Scalable Unbalanced Optimal Transport using Generative Adversarial Networks""","['unbalanced optimal transport', 'generative adversarial networks', 'population modeling']","""Generative adversarial networks (GANs) are an expressive class of neural generative models with tremendous success in modeling high-dimensional continuous measures. In this paper, we present a scalable method for unbalanced optimal transport (OT) based on the generative-adversarial framework. We formulate unbalanced OT as a problem of simultaneously learning a transport map and a scaling factor that push a source measure to a target measure in a cost-optimal manner. We provide theoretical justification for this formulation, showing that it is closely related to an existing static formulation by Liero et al. (2018). We then propose an algorithm for solving this problem based on stochastic alternating gradient updates, similar in practice to GANs, and perform numerical experiments demonstrating how this methodology can be applied to population modeling.""","""After revision, all reviewers agree that this paper makes an interesting contribution to ICLR by proposing a new methodology for unbalanced optimal transport using GANs and should be accepted."""
784,"""Pixel Redrawn For A Robust Adversarial Defense""","['adversarial machine learning', 'deep learning', 'adversarial example']","""Recently, an adversarial example becomes a serious problem to be aware of because it can fool trained neural networks easily. To prevent the issue, many researchers have proposed several defense techniques such as adversarial training, input transformation, stochastic activation pruning, etc. In this paper, we propose a novel defense technique, Pixel Redrawn (PR) method, which redraws every pixel of training images to convert them into distorted images. The motivation for our PR method is from the observation that the adversarial attacks have redrawn some pixels of the original image with the known parameters of the trained neural network. Mimicking these attacks, our PR method redraws the image without any knowledge of the trained neural network. This method can be similar to the adversarial training method but our PR method can be used to prevent future attacks. Experimental results on several benchmark datasets indicate our PR method not only relieves the over-fitting issue when we train neural networks with a large number of epochs, but it also boosts the robustness of the neural network.""","""Based on the majority of reviewers with reject (ratings: 4,6,3), the current version of paper is proposed as reject. """
785,"""NICE: noise injection and clamping estimation for neural network quantization""","['Efficient inference', 'Hardware-efficient model architectures', 'Quantization']","""Convolutional Neural Networks (CNN) are very popular in many fields including computer vision, speech recognition, natural language processing, to name a few. Though deep learning leads to groundbreaking performance in these domains, the networks used are very demanding computationally and are far from real-time even on a GPU, which is not power efficient and therefore does not suit low power systems such as mobile devices. To overcome this challenge, some solutions have been proposed for quantizing the weights and activations of these networks, which accelerate the runtime significantly. Yet, this acceleration comes at the cost of a larger error. The NICE method proposed in this work trains quantized neural networks by noise injection and a learned clamping, which improve the accuracy. This leads to state-of-the-art results on various regression and classification tasks, e.g., ImageNet classification with architectures such as ResNet-18/34/50 with low as 3-bit weights and 3 -bit activations. We implement the proposed solution on an FPGA to demonstrate its applicability for low power real-time applications.""","""This paper addresses an important problem, quantizing deep neural network models to reduce the cost of implementing them on hardware such as FPGAs without severely affecting task performance. The approach explored in the paper combines three ideas: (1) injecting noise into the network to simulate the effects of quantization noise, (2) a smart initialization of the parameter and activation clamping along with learning of the activation clamping using the straight-through estimator, and (3) a gradual approach to quantization. While the reviewers agreed that the problem is important, they raised concerns about the novelty of the proposed approach and the quality of the experiments. The authors did not respond to the reviewers in the discussion period, and did not revise their submission."""
786,"""Self-Monitoring Navigation Agent via Auxiliary Progress Estimation""","['visual grounding', 'textual grounding', 'instruction-following', 'navigation agent']","""The Vision-and-Language Navigation (VLN) task entails an agent following navigational instruction in photo-realistic unknown environments. This challenging task demands that the agent be aware of which instruction was completed, which instruction is needed next, which way to go, and its navigation progress towards the goal. In this paper, we introduce a self-monitoring agent with two complementary components: (1) visual-textual co-grounding module to locate the instruction completed in the past, the instruction required for the next action, and the next moving direction from surrounding images and (2) progress monitor to ensure the grounded instruction correctly reflects the navigation progress. We test our self-monitoring agent on a standard benchmark and analyze our proposed approach through a series of ablation studies that elucidate the contributions of the primary components. Using our proposed method, we set the new state of the art by a significant margin (8% absolute increase in success rate on the unseen test set). Code is available at pseudo-url.""","""The authors have described a navigation method that uses co-grounding between language and vision as well as an explicit self-assessment of progress. The method is used for room 2 room navigation and is tested in unseen environments. On the positive side, the approach is well-analyzed, with multiple ablations and baseline comparisons. The method is interesting and could be a good starting point for a more ambitious grounded language-vision agent. The approach seems to work well and achieves a high score using the metric of successful goal acquisition. On the negative side, the method relies on beam search, which is certainly unrealistic for real-world navigation, the evaluation metric is very simple and may be misleading, and the architecture is quite complex, may not scale or survive the test of time, and has little relevance for the greater ML community. There was a long discussion between the authors and the reviewers and other members of the public that resolved many of these points, with the authors being extremely responsive in giving additional results and details, and the reviewers' conclusion is that the paper should be accepted. """
787,"""Automatically Composing Representation Transformations as a Means for Generalization""","['compositionality', 'deep learning', 'metareasoning']","""A generally intelligent learner should generalize to more complex tasks than it has previously encountered, but the two common paradigms in machine learning -- either training a separate learner per task or training a single learner for all tasks -- both have difficulty with such generalization because they do not leverage the compositional structure of the task distribution. This paper introduces the compositional problem graph as a broadly applicable formalism to relate tasks of different complexity in terms of problems with shared subproblems. We propose the compositional generalization problem for measuring how readily old knowledge can be reused and hence built upon. As a first step for tackling compositional generalization, we introduce the compositional recursive learner, a domain-general framework for learning algorithmic procedures for composing representation transformations, producing a learner that reasons about what computation to execute by making analogies to previously seen problems. We show on a symbolic and a high-dimensional domain that our compositional approach can generalize to more complex problems than the learner has previously encountered, whereas baselines that are not explicitly compositional do not.""",""" pros: - the paper is well-written and presents a nice framing of the composition problem - good comparison to prior work - very important research direction cons: - from an architectural standpoint the paper is somewhat incremental over Routing Networks [Rosenbaum et al] - as Reviewers 2 and 3 point out, the experiments are a bit weak, relying on heuristics such as a window over 3 symbols in the multi-lingual arithmetic case, and a pre-determined set of operations (scaling, translation, rotation, identity) in the MNIST case. As the authors state, there are three core ideas in this paper (my paraphrase): (1) training on a set of compositional problems (with the right architecture/training procedure) can encourage the model to learn modules which can be composed to solve new problems, enabling better generalization. (2) treating the problem of selecting functions for composition as a sequential decision-making problem in an MDP (3) jointly learning the parameters of the functions and the (meta-level) composition policy. As discussed during the review period, these three ideas are already present in the Routing Networks (RN) architecture of Rosenbaum et al. However CRL offers insights and improvements over RN algorithmically in a several ways: (1) CRL uses a curriculum learning strategy. This seems to be key in achieving good results and makes a lot of sense for naturally compositional problems. (2) The focus in RN was on using the architecture to solve multi-task problems in object recognition. The solutions learned in image domains while ""compositional"" are less clearly interpretable. In this paper (CRL) the focus is more squarely on interpretable compositional tasks like arithmetic and explores extrapolation. (3) The RN architecture does support recursion (and there are some experiments in this mode) but it was not the main focus. In this paper (CRL) recursion is given a clear, prominent role. I appreciate that the authors' engagement in the discussion period. My feeling is that the paper offers nice improvements, a useful framing of the problem, a clear recursive formulation, and a more central focus on naturally compositional problems. I am recommending the paper for acceptance but suggest that the authors remove or revise their contributions (3) and (4) on pg. 2 in light of the discussion on routing nets. Routing Networks, Adaptive Selection of Non-Linear Functions for Multi-task Learning, ICLR 2018"""
788,"""On Accurate Evaluation of GANs for Language Generation""","['GANs', 'Evaluation', 'Generative Models']","""Generative Adversarial Networks (GANs) are a promising approach to language generation. The latest works introducing novel GAN models for language generation use n-gram based metrics for evaluation and only report single scores of the best run. In this paper, we argue that this often misrepresents the true picture and does not tell the full story, as GAN models can be extremely sensitive to the random initialization and small deviations from the best hyperparameter choice. In particular, we demonstrate that the previously used BLEU score is not sensitive to semantic deterioration of generated texts and propose alternative metrics that better capture the quality and diversity of the generated samples. We also conduct a set of experiments comparing a number of GAN models for text with a conventional Language Model (LM) and find that none of the considered models performs convincingly better than the LM.""","""This paper conducts experiments evaluating several different metrics for evaluating GAN-based language generation models. This is a worthy pursuit, and some of the evaluation is interesting. However as noted by Reviewer 2, there are a number of concerns with the execution of the paper: evaluation of metrics with respect to human judgement is insufficient, the diversity of the text samples is not evaluated, and there are clarity issues. I feel that with a major re-write and tighter experiments this paper could potentially become something nice, but in its current form it seems below the ICLR quality threshold. """
789,"""Constraining Action Sequences with Formal Languages for Deep Reinforcement Learning""","['reinforcement learning', 'constraints', 'finite state machines']","""We study the problem of deep reinforcement learning where the agent's action sequences are constrained, e.g., prohibition of dithering or overactuating action sequences that might damage a robot, drone, or other physical device. Our model focuses on constraints that can be described by automata such as DFAs or PDAs. We then propose multiple approaches to augment the state descriptions of the Markov decision process (MDP) with summaries of recent action histories. We empirically evaluate these methods applying DQN to three Atari games, training with reward shaping. We found that our approaches are effective in significantly reducing, and even eliminating, constraint violations while maintaining high reward. We also observed that the total reward achieved by an agent can be highly sensitive to how much the constraints encourage or discourage exploration of potentially effective actions during training, and, in addition to helping ensure safe policies, the use of constraints can enhance exploration during training.""","""The paper studies the problem of reinforcement learning under certain constraints on action sequences. The reviewers raised important concerns regarding (1) the general motivation, (2) the particular formulation of constraints in terms of action sequences and (3) the relevance and significance of experimental results. The authors did not submit a rebuttal. Given the concerns raised by the reviewers, I encourage the authors to improve the paper to possibly resubmit to another venue."""
790,"""Principled Deep Neural Network Training through Linear Programming""","['deep learning theory', 'neural network training', 'empirical risk minimization', 'non-convex optimization', 'treewidth']","""Deep Learning has received significant attention due to its impressive performance in many state-of-the-art learning tasks. Unfortunately, while very powerful, Deep Learning is not well understood theoretically and in particular only recently results for the complexity of training deep neural networks have been obtained. In this work we show that large classes of deep neural networks with various architectures (e.g., DNNs, CNNs, Binary Neural Networks, and ResNets), activation functions (e.g., ReLUs and leaky ReLUs), and loss functions (e.g., Hinge loss, Euclidean loss, etc) can be trained to near optimality with desired target accuracy using linear programming in time that is exponential in the input data and parameter space dimension and polynomial in the size of the data set; improvements of the dependence in the input dimension are known to be unlikely assuming NP$, and improving the dependence on the parameter space dimension remains open. In particular, we obtain polynomial time algorithms for training for a given fixed network architecture. Our work applies more broadly to empirical risk minimization problems which allows us to generalize various previous results and obtain new complexity results for previously unstudied architectures in the proper learning setting.""","""The strength of the paper is that it designs an LP-based algorithm for training neural networks with runtime exponential in the number of parameters and linear in the size of the datasets and the algorithm works for worst-case datasets. As reviewer 2 and reviewer 3 pointed out, the cons include a) it's not clear why the algorithm provides any theoretical insights on how to design in the future polynomial time algorithm --- it seems that the algorithms are inherently exponential time and b) it's not clear whether the algorithm is practically at all relevant. The AC also noted that brute-force search algorithm is also exponential in # parameters and linear in size of datasets, and the authors agreed with it. This leaves the main contribution of the paper be that it works for the worst datasets. However, theoretically speaking, it's not clear this should be counted as a feature for algorithm design because we cannot go beyond the intractability without making assumptions on the data and in the AC's opinion, the big open question is how to make additional assumptions on the data (instead of removing them.) In summary, the drawback b) makes this a purely theoretical paper and the theoretical significance of the paper is unclear due to a). Therefore based on a), the AC decided to recommend reject, although the AC suggested the authors to re-submit to other top theory or ML theory conferences which may better evaluate the theoretical significance of the paper. """
791,"""FlowQA: Grasping Flow in History for Conversational Machine Comprehension""","['Machine Comprehension', 'Conversational Agent', 'Natural Language Processing', 'Deep Learning']","""Conversational machine comprehension requires a deep understanding of the conversation history. To enable traditional, single-turn models to encode the history comprehensively, we introduce Flow, a mechanism that can incorporate intermediate representations generated during the process of answering previous questions, through an alternating parallel processing structure. Compared to shallow approaches that concatenate previous questions/answers as input, Flow integrates the latent semantics of the conversation history more deeply. Our model, FlowQA, shows superior performance on two recently proposed conversational challenges (+7.2% F1 on CoQA and +4.0% on QuAC). The effectiveness of Flow also shows in other tasks. By reducing sequential instruction understanding to conversational machine comprehension, FlowQA outperforms the best models on all three domains in SCONE, with +1.8% to +4.4% improvement in accuracy.""","""Interesting and novel approach of modeling context (mainly external documents with information about the conversation content) for the conversational question answering task, demonstrating significant improvements on the newly released conversational QA datasets. The first version of the paper was weaker on motivation and lacked a clearer presentation of the approach as mentioned by the reviewers, but the paper was updated as explained in the responses to the reviewers. The ablation studies are useful in demonstration of the proposed FLOW approach. A question still remains after the reviews (this was not raised by the reviewers): How does the approach perform in comparison to the state of the art for the single question and answer tasks? If each question was asked in isolation, would it still be the best? """
792,"""Probabilistic Program Induction for Intuitive Physics Game Play""","['intuitive physics', 'probabilistic programming', 'computational cognitive science', 'probabilistic models']","""Recent findings suggest that humans deploy cognitive mechanism of physics simulation engines to simulate the physics of objects. We propose a framework for bots to deploy similar tools for interacting with intuitive physics environments. The framework employs a physics simulation in a probabilistic way to infer about moves performed by an agent in a setting governed by Newtonian laws of motion. However, methods of probabilistic programs can be slow in such setting due to their need to generate many samples. We complement the model with a model-free approach to aid the sampling procedures in becoming more efficient through learning from experience during game playing. We present an approach where a myriad of model-free approaches (a convolutional neural network in our model) and model-based approaches (probabilistic physics simulation) is able to achieve what neither could alone. This way the model outperforms an all model-free or all model-based approach. We discuss a case study showing empirical results of the performance of the model on the game of Flappy Bird. ""","""The paper presents the combination of a model-based (probabilistic program representing the physics) and model-free (CNN trained with DQN) to play Flappy Bird. The approach is interesting, but the paper is hard to follow at times, and the solution seems too specific to the Flappy Bird game. This feels more like a tech report on what was done to get this score on Flappy Bird, than a scientific paper with good comparisons on this environment (in terms of models, algorithms, approaches), and/or other environments to evaluate the method. We encourage the authors to do this additional work."""
793,"""A Self-Supervised Method for Mapping Human Instructions to Robot Policies""",[],"""In this paper, we propose a modular approach which separates the instruction-to-action mapping procedure into two separate stages. The two stages are bridged via an intermediate representation called a goal, which stands for the result after a robot performs a specific task. The first stage maps an input instruction to a goal, while the second stage maps the goal to an appropriate policy selected from a set of robot policies. The policy is selected with an aim to guide the robot to reach the goal as close as possible. We implement the above two stages as a framework consisting of two distinct modules: an instruction-goal mapping module and a goal-policy mapping module. Given a human instruction in the evaluation phase, the instruction-goal mapping module first translates the instruction to a robot-interpretable goal. Once a goal is derived by the instruction-goal mapping module, the goal-policy mapping module then follows up to search through the goal-policy pairs to look for policy to be mapped by the instruction. Our experimental results show that the proposed method is able to learn an effective instruction-to-action mapping procedure in an environment with a given instruction set more efficiently than the baselines. In addition to the impressive data-efficiency, the results also show that our method can be adapted to a new instruction set and a new robot action space much faster than the baselines. The evidence suggests that our modular approach does lead to better adaptability and efficiency. ""","""The paper proposes a novel approach to interfacing robots with humans, or rather vv: by mapping instructions to goals, and goals to robot actions. A possibly nice idea, and possibly good for more efficient learning. But the technical realisation is less strong than the initial idea. The original idea merits a good evaluation, and the authors are strongly encouraged to follow up on this idea and realise it, towards a stronger publication. It be noted that the authors refrained from using the rebuttal phase."""
794,"""Learning Neuron Non-Linearities with Kernel-Based Deep Neural Networks""","['Activation functions', 'Kernel methods', 'Recurrent networks']","""The effectiveness of deep neural architectures has been widely supported in terms of both experimental and foundational principles. There is also clear evidence that the activation function (e.g. the rectier and the LSTM units) plays a crucial role in the complexity of learning. Based on this remark, this paper discusses an optimal selection of the neuron non-linearity in a functional framework that is inspired from classic regularization arguments. A representation theorem is given which indicates that the best activation function is a kernel expansion in the training set, that can be effectively approximated over an opportune set of points modeling 1-D clusters. The idea can be naturally extended to recurrent networks, where the expressiveness of kernel-based activation functions turns out to be a crucial ingredient to capture long-term dependencies. We give experimental evidence of this property by a set of challenging experiments, where we compare the results with neural architectures based on state of the art LSTM cells.""","""This paper proposes to automatically learn the form of the non-linearities of neural networks in deep neural networks, which the reviewers noted to be an interesting albeit significantly studied direction. Overall, this paper falls just below the bar, with no reviewer really willing to champion for acceptance. Reviewer 3 found the paper to be marginally above the acceptance threshold and found the insights provided in the paper (in Section 2) to be a neat and strong contribution. Reviewers 1-2, however, found the paper marginally below the bar and seemed confused by the presentation of the paper. They seemed to believe in the motivation and idea, but they found the paper hard to follow and not particularly clearly written. It would seem that the paper could significantly benefit from careful editing and restructuring to disambiguate contributions from motivation and existing literature. Also, the authors should provide clear justification for their design choices and modeling assumptions. """
795,"""An Empirical Study of Example Forgetting during Deep Neural Network Learning""","['catastrophic forgetting', 'sample weighting', 'deep generalization']","""Inspired by the phenomenon of catastrophic forgetting, we investigate the learning dynamics of neural networks as they train on single classification tasks. Our goal is to understand whether a related phenomenon occurs when data does not undergo a clear distributional shift. We define a ``forgetting event'' to have occurred when an individual training example transitions from being classified correctly to incorrectly over the course of learning. Across several benchmark data sets, we find that: (i) certain examples are forgotten with high frequency, and some not at all; (ii) a data set's (un)forgettable examples generalize across neural architectures; and (iii) based on forgetting dynamics, a significant fraction of examples can be omitted from the training data set while still maintaining state-of-the-art generalization performance.""","""This paper is an analysis of the phenomenon of example forgetting in deep neural net training. The empirical study is the first of its kind and features convincing experiments with architectures that achieve near state-of-the-art results. It shows that a portion of the training set can be seen as support examples. The reviewers noted weaknesses such as in the measurement of the forgetting itself and the training regiment. However, they agreed that their concerns we addressed by the rebuttal. They also noted that the paper is not forthcoming with insights, but found enough value in the systematic empirical study it provides."""
796,"""Consistency-based anomaly detection with adaptive multiple-hypotheses predictions""","['Anomaly detection', 'outlier detection', 'generative models', 'VAE', 'GAN']","""In one-class-learning tasks, only the normal case can be modeled with data, whereas the variation of all possible anomalies is too large to be described sufficiently by samples. Thus, due to the lack of representative data, the wide-spread discriminative approaches cannot cover such learning tasks, and rather generative models, which attempt to learn the input density of the normal cases, are used. However, generative models suffer from a large input dimensionality (as in images) and are typically inefficient learners. We propose to learn the data distribution more efficiently with a multi-hypotheses autoencoder. Moreover, the model is criticized by a discriminator, which prevents artificial data modes not supported by data, and which enforces diversity across hypotheses. This consistency-based anomaly detection (ConAD) framework allows the reliable identification of outof- distribution samples. For anomaly detection on CIFAR-10, it yields up to 3.9% points improvement over previously reported results. On a real anomaly detection task, the approach reduces the error of the baseline models from 6.8% to 1.5%.""","""This paper proposes an anomaly-detection approach by augmenting VAE encoder with a network multiple hypothesis network and then using a discriminator in the decoder to select one of the hypothesis. The idea is interesting although the reviewers found the paper to be poorly written and the approach to be a bit confusing and complicated. Revisions and rebuttal have certainly helped to improve the quality of the work. However, the reviewers believe that the paper require more work before it can be accepted at ICLR. For this reason, I recommend to reject this paper in its current state. """
797,"""Modeling Dynamics of Biological Systems with Deep Generative Neural Networks""","['neural networks', 'markovian dynamics', 'single-cell biology', 'calcium imaging', 'stochastic dynamics', 'generative models']","""Biological data often contains measurements of dynamic entities such as cells or organisms in various states of progression. However, biological systems are notoriously difficult to describe analytically due to their many interacting components, and in many cases, the technical challenge of taking longitudinal measurements. This leads to difficulties in studying the features of the dynamics, for examples the drivers of the transition. To address this problem, we present a deep neural network framework we call Dynamics Modeling Network or DyMoN. DyMoN is a neural network framework trained as a deep generative Markov model whose next state is a probability distribution based on the current state. DyMoN is well-suited to the idiosyncrasies of biological data, including noise, sparsity, and the lack of longitudinal measurements in many types of systems. Thus, DyMoN can be trained using probability distributions derived from the data in any way, such as trajectories derived via dimensionality reduction methods, and does not require longitudinal measurements. We show the advantage of learning deep models over shallow models such as Kalman filters and hidden Markov models that do not learn representations of the data, both in terms of learning embeddings of the data and also in terms training efficiency, accuracy and ability to multitask. We perform three case studies of applying DyMoN to different types of biological systems and extracting features of the dynamics in each case by examining the learned model. ""","""The paper tackles an interesting problem, which is effectively modeling biological time-series data. The advantages of deep neural networks over structured models like HMMs are their ability to learn features from the data, whereas probabilistic graphical models suffer from ""model mismatch"", where the available data must be carefully processed in order to fit the assumptions of the PGM. Any work advancing this topic would be extremely welcome in the world of machine learning in biology. However, the reviewers each raised individual concerns about the paper regarding its clarity and quality, and the authors did not respond. Thus, the reviewers scores remain unchanged, and the rough consensus is a rejection."""
798,"""Predicted Variables in Programming""","['predicted variables', 'machine learning', 'programming', 'computing systems', 'reinforcement learning']","""We present Predicted Variables, an approach to making machine learning (ML) a first class citizen in programming languages. There is a growing divide in approaches to building systems: using human experts (e.g. programming) on the one hand, and using behavior learned from data (e.g. ML) on the other hand. PVars aim to make using ML in programming easier by hybridizing the two. We leverage the existing concept of variables and create a new type, a predicted variable. PVars are akin to native variables with one important distinction: PVars determine their value using ML when evaluated. We describe PVars and their interface, how they can be used in programming, and demonstrate the feasibility of our approach on three algorithmic problems: binary search, QuickSort, and caches. We show experimentally that PVars are able to improve over the commonly used heuristics and lead to a better performance than the original algorithms. As opposed to previous work applying ML to algorithmic problems, PVars have the advantage that they can be used within the existing frameworks and do not require the existing domain knowledge to be replaced. PVars allow for a seamless integration of ML into existing systems and algorithms. Our PVars implementation currently relies on standard Reinforcement Learning (RL) methods. To learn faster, PVars use the heuristic function, which they are replacing, as an initial function. We show that PVars quickly pick up the behavior of the initial function and then improve performance beyond that without ever performing substantially worse -- allowing for a safe deployment in critical applications.""","""The paper proposes a framework at the intersection of programming and machine learning, where some variables in a program are replaced by PVars - variables whose values are learned using machine learning from data. The paper presents an API that is designed to support this scenario, as well as three case studies: binary search, quick sort, and caching - all implemented with PVars. The reviewers and the AC agree that the paper presents and potentially valuable new idea, and shows concrete applications in the presented case studies. They provide example code in the paper, and present a detailed analysis of the obtained results. The reviewers and AC also not several potential weaknesses - the AC will focus on a subset for the present discussion. The paper is unusual in that it presents a programming API rather than e.g., a thorough empirical comparison, a novel approach, or new theoretical insights. Paper at the intersection of systems and machine learning can make valuable contributions to the ICLR community, but need to provide a clear contributions which are supported in the paper by empirical or theoretical results. The research contributions of the present paper are vague, even after the revision phase. The main contribution claimed is the introduction of the API, and that such an API / system is feasible. This is an extremely weak claim. A stronger claim would be if e.g., the present approach would advance the state of the art beyond an existing such framework, e.g., probabilistic programming, either conceptually or empirically. I want to particularly highlight probabilistic programming here, as it is mentioned by the authors - this is a well developed research area, with existing approaches and widely used tools. The authors dismiss this approach in their related work section, saying that probabilistic programming is ""specialized on working with distributions"". Many would see the latter as a benefit, so the authors should clearly motivate how their approach improves over these existing methods, and how it would enable novel uses or otherwise provide benefits. At the current stage, the paper is not ready for publication."""
799,"""Exploiting Cross-Lingual Subword Similarities in Low-Resource Document Classification""","['cross-lingual transfer', 'character-based method', 'low-resource language']","""Text classification must sometimes be applied in situations with no training data in a target language. However, training data may be available in a related language. We introduce a cross-lingual document classification framework CACO between related language pairs. To best use limited training data, our transfer learning scheme exploits cross-lingual subword similarity by jointly training a character-based embedder and a word-based classifier. The embedder derives vector representations for input words from their written forms, and the classifier makes predictions based on the word vectors. We use a joint character representation for both the source language and the target language, which allows the embedder to generalize knowledge about source language words to target language words with similar forms. We propose a multi-task objective that can further improve the model if additional cross-lingual or monolingual resources are available. CACO models trained under low-resource settings rival cross-lingual word embedding models trained under high-resource settings on related language pairs. ""","""The paper's contribution lies in using cross-lingual sharing of subword representations for improving document classification. The paper presents interesting models and results. While the paper is good (two out of three reviewers are happy about it), I do agree with the reviewer who suggests the experimentation with relatively dissimilar languages and showing whether or not the approach works for those cases. I am also not very happy with the author response to the reviewer. Moreover, I think the paper could improve further if the authors presented experiments on more tasks apart from document classification."""
800,"""ROBUST ESTIMATION VIA GENERATIVE ADVERSARIAL NETWORKS""","['robust statistics', 'neural networks', 'minimax rate', 'data depth', 'contamination model', 'Tukey median', 'GAN']","""Robust estimation under Huber's pseudo-formula -contamination model has become an important topic in statistics and theoretical computer science. Rate-optimal procedures such as Tukey's median and other estimators based on statistical depth functions are impractical because of their computational intractability. In this paper, we establish an intriguing connection between f-GANs and various depth functions through the lens of f-Learning. Similar to the derivation of f-GAN, we show that these depth functions that lead to rate-optimal robust estimators can all be viewed as variational lower bounds of the total variation distance in the framework of f-Learning. This connection opens the door of computing robust estimators using tools developed for training GANs. In particular, we show that a JS-GAN that uses a neural network discriminator with at least one hidden layer is able to achieve the minimax rate of robust mean estimation under Huber's pseudo-formula -contamination model. Interestingly, the hidden layers of the neural net structure in the discriminator class are shown to be necessary for robust estimation.""",""" * Strengths This paper presents a very interesting connection between GANs and robust estimation in the presence of corrupted training data. The conceptual ideas are novel and can likely be extended in many further directions. I would not be surprised if this opens up a new line of research. * Weaknesses The paper is poorly written. Due to disagreement among the authors and my interest in the topic, I read the paper in detail myself. I think it would be difficult for a non-expert to understand the key ideas and I strongly encourage the authors to carefully revise the paper to reach a broader audience and highlight the key insights. Additionally, the experiments are only on toy data. * Discussion One of the reviewers was concerned about the lack of efficiency guarantees for the proposed algorithm (indeed, the algorithm requires training GANs which are currently beyond the reach of theory and finicky in practice). That reviewer points to the fact that most papers in the robustness literature are concerned with computational efficiency and is concerned that ignoring this sidesteps one of the key challenges. The reviewer is also concerned about the restriction to parametric or nearly-parametric families (e.g. Gaussians and elliptical distributions). Other reviewers were more positive and did not see these as major issues. * Decision In my opinion, the lack of efficiency guarantees is not a huge issue, as the primary contribution of the paper is pointing out a non-obvious conceptual connection between two literatures. The restriction to parametric families is more concerning, but it seems possible this could be removed with further developments. The main reason for accepting the paper (despite concerns about the writing) is the importance of the conceptual connection. I think this connection is likely to lead to a new line of research and would like to get it out there as soon as possible. * Comments Despite the accept decision, I again urge the authors to improve the quality of exposition to ensure that a large audience can appreciate the ideas."""
801,"""Online Learning for Supervised Dimension Reduction""","['Online Learning', 'Supervised Dimension Reduction', 'Incremental Sliced Inverse Regression', 'Effective Dimension Reduction Space']",""" Online learning has attracted great attention due to the increasing demand for systems that have the ability of learning and evolving. When the data to be processed is also high dimensional and dimension reduction is necessary for visualization or prediction enhancement, online dimension reduction will play an essential role. The purpose of this paper is to propose new online learning approaches for supervised dimension reduction. Our first algorithm is motivated by adapting the sliced inverse regression (SIR), a pioneer and effective algorithm for supervised dimension reduction, and making it implementable in an incremental manner. The new algorithm, called incremental sliced inverse regression (ISIR), is able to update the subspace of significant factors with intrinsic lower dimensionality fast and efficiently when new observations come in. We also refine the algorithm by using an overlapping technique and develop an incremental overlapping sliced inverse regression (IOSIR) algorithm. We verify the effectiveness and efficiency of both algorithms by simulations and real data applications.""","""The paper investigates an incremental form of Sliced Inverse Regression (SIR) for supervised dimensionality reduction. Unfortunately, the experimental evaluation is insufficient as a serious evaluation of the proposed techniques. More importantly, the paper does not appear to contribute a significant advance over the extensive literature on fast generalized eigenvalue decompositions in machine learning. No responses were offered to counter such an opinion."""
802,"""Meta-Learning to Guide Segmentation""","['meta-learning', 'few-shot learning', 'visual segmentation']","""There are myriad kinds of segmentation, and ultimately the `""right"" segmentation of a given scene is in the eye of the annotator. Standard approaches require large amounts of labeled data to learn just one particular kind of segmentation. As a first step towards relieving this annotation burden, we propose the problem of guided segmentation: given varying amounts of pixel-wise labels, segment unannotated pixels by propagating supervision locally (within an image) and non-locally (across images). We propose guided networks, which extract a latent task representation---guidance---from variable amounts and classes (categories, instances, etc.) of pixel supervision and optimize our architecture end-to-end for fast, accurate, and data-efficient segmentation by meta-learning. To span the few-shot and many-shot learning regimes, we examine guidance from as little as one pixel per concept to as much as 1000+ images, and compare to full gradient optimization at both extremes. To explore generalization, we analyze guidance as a bridge between different levels of supervision to segment classes as the union of instances. Our segmentor concentrates different amounts of supervision of different types of classes into an efficient latent representation, non-locally propagates this supervision across images, and can be updated quickly and cumulatively when given more supervision.""","""Paper proposes a meta-learning approach to interactive segmentation. After the author response, R2 and R3 recommend rejecting this paper citing concerns of limited novelty and insufficient experimental evaluation (given the popularity of this topic in computer vision). R1 does not seem be familiar with the extensive literature on interactive segmentation and their positive recommendation has been discounted. The AC finds no basis for accepting this paper. """
803,"""Gradient Acceleration in Activation Functions""","['Gradient Acceleration', 'Saturation Areas', 'Dropout', 'Coadaptation']","""Dropout has been one of standard approaches to train deep neural networks, and it is known to regularize large models to avoid overfitting. The effect of dropout has been explained by avoiding co-adaptation. In this paper, however, we propose a new explanation of why dropout works and propose a new technique to design better activation functions. First, we show that dropout can be explained as an optimization technique to push the input towards the saturation area of nonlinear activation function by accelerating gradient information flowing even in the saturation area in backpropagation. Based on this explanation, we propose a new technique for activation functions, {\em gradient acceleration in activation function (GAAF)}, that accelerates gradients to flow even in the saturation area. Then, input to the activation function can climb onto the saturation area which makes the network more robust because the model converges on a flat region. Experiment results support our explanation of dropout and confirm that the proposed GAAF technique improves performances with expected properties.""","""Reviewers are in a consensus and recommended to reject. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit."""
804,"""Interpreting Layered Neural Networks via Hierarchical Modular Representation""","['interpretabile machine learning', 'neural network', 'hierarchical clustering']","""Interpreting the prediction mechanism of complex models is currently one of the most important tasks in the machine learning field, especially with layered neural networks, which have achieved high predictive performance with various practical data sets. To reveal the global structure of a trained neural network in an interpretable way, a series of clustering methods have been proposed, which decompose the units into clusters according to the similarity of their inference roles. The main problems in these studies were that (1) we have no prior knowledge about the optimal resolution for the decomposition, or the appropriate number of clusters, and (2) there was no method with which to acquire knowledge about whether the outputs of each cluster have a positive or negative correlation with the input and output dimension values. In this paper, to solve these problems, we propose a method for obtaining a hierarchical modular representation of a layered neural network. The application of a hierarchical clustering method to a trained network reveals a tree-structured relationship among hidden layer units, based on their feature vectors defined by their correlation with the input and output dimension values. ""","""All reviewers agree to reject. While there were many positive points to this work, reviewers believed that it was not yet ready for acceptance."""
805,"""Bayesian Prediction of Future Street Scenes using Synthetic Likelihoods""","['bayesian inference', 'segmentation', 'anticipation', 'multi-modality']","""For autonomous agents to successfully operate in the real world, the ability to anticipate future scene states is a key competence. In real-world scenarios, future states become increasingly uncertain and multi-modal, particularly on long time horizons. Dropout based Bayesian inference provides a computationally tractable, theoretically well grounded approach to learn different hypotheses/models to deal with uncertain futures and make predictions that correspond well to observations -- are well calibrated. However, it turns out that such approaches fall short to capture complex real-world scenes, even falling behind in accuracy when compared to the plain deterministic approaches. This is because the used log-likelihood estimate discourages diversity. In this work, we propose a novel Bayesian formulation for anticipating future scene states which leverages synthetic likelihoods that encourage the learning of diverse models to accurately capture the multi-modal nature of future scene states. We show that our approach achieves accurate state-of-the-art predictions and calibrated probabilities through extensive experiments for scene anticipation on Cityscapes dataset. Moreover, we show that our approach generalizes across diverse tasks such as digit generation and precipitation forecasting.""","""This paper proposes a method to encourage diversity of Bayesian dropout method. A discriminator is used to facilitate diversity, which the method deal with multi-modality. Empirical results show good improvement over existing methods. This is a good paper and should be accepted. """
806,"""Label super-resolution networks""","['weakly supervised segmentation', 'land cover mapping', 'medical imaging']","""We present a deep learning-based method for super-resolving coarse (low-resolution) labels assigned to groups of image pixels into pixel-level (high-resolution) labels, given the joint distribution between those low- and high-resolution labels. This method involves a novel loss function that minimizes the distance between a distribution determined by a set of model outputs and the corresponding distribution given by low-resolution labels over the same set of outputs. This setup does not require that the high-resolution classes match the low-resolution classes and can be used in high-resolution semantic segmentation tasks where high-resolution labeled data is not available. Furthermore, our proposed method is able to utilize both data with low-resolution labels and any available high-resolution labels, which we show improves performance compared to a network trained only with the same amount of high-resolution data. We test our proposed algorithm in a challenging land cover mapping task to super-resolve labels at a 30m resolution to a separate set of labels at a 1m resolution. We compare our algorithm with models that are trained on high-resolution data and show that 1) we can achieve similar performance using only low-resolution data; and 2) we can achieve better performance when we incorporate a small amount of high-resolution data in our training. We also test our approach on a medical imaging problem, resolving low-resolution probability maps into high-resolution segmentation of lymphocytes with accuracy equal to that of fully supervised models.""","""This paper formulates a method for training deep networks to produce high-resolution semantic segmentation output using only low-resolution ground-truth labels. Reviewers agree that this is a useful contribution, but with the limitation that joint distribution between low- and high-resolution labels must be known. Experimental results are convincing. The technique introduced by the paper could be applicable to many semantic segmentation problems and is likely to be of general interest. """
807,"""Probabilistic Planning with Sequential Monte Carlo methods""","['control as inference', 'probabilistic planning', 'sequential monte carlo', 'model based reinforcement learning']","""In this work, we propose a novel formulation of planning which views it as a probabilistic inference problem over future optimal trajectories. This enables us to use sampling methods, and thus, tackle planning in continuous domains using a fixed computational budget. We design a new algorithm, Sequential Monte Carlo Planning, by leveraging classical methods in Sequential Monte Carlo and Bayesian smoothing in the context of control as inference. Furthermore, we show that Sequential Monte Carlo Planning can capture multimodal policies and can quickly learn continuous control tasks.""","""This paper presents a new approach for posing control as inference that leverages Sequential Monte Carlo and Bayesian smoothing. There is significant interest from the reviewers into this method, and also an active discussion about this paper, particularly with respect to the optimism bias issue. The paper is borderline and the authors are encouraged to address the desired clarifications and changes from the reviewers. """
808,"""ON THE EFFECTIVENESS OF TASK GRANULARITY FOR TRANSFER LEARNING""","['Transfer Learning', 'Video Understanding', 'Fine-grained Video Classification', 'Video Captioning', 'Common Sense', 'Something-Something Dataset.']","""We describe a DNN for video classification and captioning, trained end-to-end, with shared features, to solve tasks at different levels of granularity, exploring the link between granularity in a source task and the quality of learned features for transfer learning. For solving the new task domain in transfer learning, we freeze the trained encoder and fine-tune an MLP on the target domain. We train on the Something-Something dataset with over 220, 000 videos, and multiple levels of target granularity, including 50 action groups, 174 fine-grained action categories and captions. Classification and captioning with Something-Something are challenging because of the subtle differences between actions, applied to thousands of different object classes, and the diversity of captions penned by crowd actors. Our model performs better than existing classification baselines for SomethingSomething, with impressive fine-grained results. And it yields a strong baseline on the new Something-Something captioning task. Experiments reveal that training with more fine-grained tasks tends to produce better features for transfer learning.""","""This paper presents the empirical relation between the task granularity and transfer learning, when applied between video classification and video captioning. The key take away message is that more fine-grained tasks support better transfer in the case of classification---captioning transfer on 20BN-something-something dataset. Pros: The paper presents a new empirical study on transfer learning between video classification and video captioning performed on the recent 20BN-something-something dataset (220,000 videos concentrating on 174 action categories). The paper presents a lot of experimental results, albeit focused primarily on the 20BN dataset. Cons: The investigation presented by this paper on the effect of the task granularity is rather application-specific and empirical. As a result, it is unclear what generalizable knowledge or insights we gain for a broad range of other applications. The methodology used in the paper is relatively standard and not novel. Also, according to the 20BN-something-something leaderboard (pseudo-url), the performance reported in the paper does not seem competitive compared to current state-of-the-art. There were some clarification questions raised by the reviewers but the authors did not respond. Verdict: Reject. The study presented by the paper is a bit too application-specific with relatively narrow impact for ICLR. Relatively weak novelty and empirical results."""
809,"""End-to-end learning of pharmacological assays from high-resolution microscopy images""","['Convolutional Neural Networks', 'High-resolution images', 'Multiple-Instance Learning', 'Drug Discovery', 'Molecular Biology']","""Predicting the outcome of pharmacological assays based on high-resolution microscopy images of treated cells is a crucial task in drug discovery which tremendously increases discovery rates. However, end-to-end learning on these images with convolutional neural networks (CNNs) has not been ventured for this task because it has been considered infeasible and overly complex. On the largest available public dataset, we compare several state-of-the-art CNNs trained in an end-to-end fashion with models based on a cell-centric approach involving segmentation. We found that CNNs operating on full images containing hundreds of cells perform significantly better at assay prediction than networks operating on a single-cell level. Surprisingly, we could predict 29% of the 209 pharmacological assays at high predictive performance (AUC > 0.9). We compared a novel CNN architecture called GapNet against four competing CNN architectures and found that it performs on par with the best methods and at the same time has the lowest training time. Our results demonstrate that end-to-end learning on high-resolution imaging data is not only possible but even outperforms cell-centric and segmentation-dependent approaches. Hence, the costly cell segmentation and feature extraction steps are not necessary, in fact they even hamper predictive performance. Our work further suggests that many pharmacological assays could be replaced by high-resolution microscopy imaging together with convolutional neural networks.""","""This work studies the performance of several end-to-end CNN architectures for the prediction of biomedical assays in microscopy images. One of the architectures, GAPnet, is a minor modification of existing global average pooling (GAP) networks, involving skip connections and concatenations. The technical novelties are low, as outlined by several reviewers and confirmed by the authors, as most of the value of the work lies in the empirical evaluation of existing methods, or minor variants thereof. Given the low technical novelty and reviewer consensus, recommend reject, however area chair recognizes that the discovered utility may be of value for the biomedical community. Authors are encouraged to use reviewer feedback to improve the work, and submit to a biomedical imaging venue for dissemination to the appropriate communities. """
810,"""Amortized Bayesian Meta-Learning""","['variational inference', 'meta-learning', 'few-shot learning', 'uncertainty quantification']","""Meta-learning, or learning-to-learn, has proven to be a successful strategy in attacking problems in supervised learning and reinforcement learning that involve small amounts of data. State-of-the-art solutions involve learning an initialization and/or learning algorithm using a set of training episodes so that the meta learner can generalize to an evaluation episode quickly. These methods perform well but often lack good quantification of uncertainty, which can be vital to real-world applications when data is lacking. We propose a meta-learning method which efficiently amortizes hierarchical variational inference across tasks, learning a prior distribution over neural network weights so that a few steps of Bayes by Backprop will produce a good task-specific approximate posterior. We show that our method produces good uncertainty estimates on contextual bandit and few-shot learning benchmarks.""","""This paper combines two ideas: MAML, and the hierarchical Bayesian inference approach of Amit and Meir (2018). The idea is fairly straightforward but well-motivated, and it seems to work well in practice. The paper is well-written and includes good discussion of the relevant literature. The experiments show improvements on various tests of Bayesian inference, and include some good analysis beyond simply reporting better numbers. On the whole, the reviewers are fairly positive about the paper. (While the numerical scores are slightly below the cutoff, the reviewers are more positive in the discussion.) The reviewers' main complaint is the lack of comparisons against recently published methods, especially Gordon et al. (2018). The lack of comparison to this paper doesn't strike me as a big problem; the preprint was released only a few months before the deadline, their approach was very different from the proposed one, and the proposed approach has some plausible advantages (simplicity, computational efficiency), so I don't think a direct comparison is required for acceptance. Overall, I recommend acceptance. """
811,"""Gradient descent aligns the layers of deep linear networks""","['implicit regularization', 'alignment of layers', 'deep linear networks', 'gradient descent', 'separable data']","""This paper establishes risk convergence and asymptotic weight matrix alignment --- a form of implicit regularization --- of gradient flow and gradient descent when applied to deep linear networks on linearly separable data. In more detail, for gradient flow applied to strictly decreasing loss functions (with similar results for gradient descent with particular decreasing step sizes): (i) the risk converges to 0; (ii) the normalized i-th weight matrix asymptotically equals its rank-1 approximation u_iv_i^T; (iii) these rank-1 matrices are aligned across layers, meaning |v_{i+1}^T u_i| -> 1. In the case of the logistic loss (binary cross entropy), more can be said: the linear function induced by the network --- the product of its weight matrices --- converges to the same direction as the maximum margin solution. This last property was identified in prior work, but only under assumptions on gradient descent which here are implied by the alignment phenomenon.""","""This paper studies the behavior of weight parameters for linear networks when trained on separable data with strictly decreasing loss functions. For this setting the paper shows that the gradient descent solution converges to max margin solution and each layer converges to a rank 1 matrix with consequent layers aligned. All reviewers agree that the paper provides novel results for understanding implicit regularization effects of gradient descent for linear networks. Despite the limitations of this paper such as studying networks with linear activation, studying gradient descent not with practical step sizes, assuming data is linearly separable, reviewers find the results useful and a good addition to existing literature."""
812,"""Backdrop: Stochastic Backpropagation""","['stochastic optimization', 'multi-scale data analysis', 'non-decomposable loss', 'generalization', 'one-shot learning']","""We introduce backdrop, a flexible and simple-to-implement method, intuitively described as dropout acting only along the backpropagation pipeline. Backdrop is implemented via one or more masking layers which are inserted at specific points along the network. Each backdrop masking layer acts as the identity in the forward pass, but randomly masks parts of the backward gradient propagation. Intuitively, inserting a backdrop layer after any convolutional layer leads to stochastic gradients corresponding to features of that scale. Therefore, backdrop is well suited for problems in which the data have a multi-scale, hierarchical structure. Backdrop can also be applied to problems with non-decomposable loss functions where standard SGD methods are not well suited. We perform a number of experiments and demonstrate that backdrop leads to significant improvements in generalization.""","""Dear authors, All reviewers pointed out that the proximity with Dropout warranted special treatment and that the justification provided in the paper was not enough to understand why exactly the changes were important. In its current state, this work is not suitable for publication to ICLR. Should you decide to resubmit this work to another venue, please take the reviewers' comments into account."""
813,"""Learning To Solve Circuit-SAT: An Unsupervised Differentiable Approach""","['Neuro-Symbolic Methods', 'Circuit Satisfiability', 'Neural SAT Solver', 'Graph Neural Networks']","""Recent efforts to combine Representation Learning with Formal Methods, commonly known as the Neuro-Symbolic Methods, have given rise to a new trend of applying rich neural architectures to solve classical combinatorial optimization problems. In this paper, we propose a neural framework that can learn to solve the Circuit Satisfiability problem. Our framework is built upon two fundamental contributions: a rich embedding architecture that encodes the problem structure and an end-to-end differentiable training procedure that mimics Reinforcement Learning and trains the model directly toward solving the SAT problem. The experimental results show the superior out-of-sample generalization performance of our framework compared to the recently developed NeuroSAT method.""","""This paper introduces a new graph neural network architecture designed to learn to solve Circuit SAT problems, a fundamental problem in computer science. The key innovation is the ability to to use the DAG structure as an input, as opposed to typical undirected (factor graph style) representations of SAT problems. The reviewers appreciated the novelty of the approach as well as the empirical results provided that demonstrate the effectiveness of the approach. Writing is clear. While the comparison with NeuroSAT is interesting and useful, there is no comparison with existing SAT solvers which are not based on learning methods. So it is not clear how big the gap with state-of-the-art is. Overall, I recommend acceptance, as the results are promising and this could inspire other researchers working on neural-symbolic approaches to search and optimization problems."""
814,"""Dissecting an Adversarial framework for Information Retrieval""","['GAN', 'Deep Learning', 'Reinforcement Learning']","""Recent advances in Generative Adversarial Networks facilitated by improvements to the framework and successful application to various problems has resulted in extensions to multiple domains. IRGAN attempts to leverage the framework for Information-Retrieval (IR), a task that can be described as modeling the correct conditional probability distribution p(d|q) over the documents (d), given the query (q). The work that proposes IRGAN claims that optimizing their minimax loss function will result in a generator which can learn the distribution, but their setup and baseline term steer the model away from an exact adversarial formulation, and this work attempts to point out certain inaccuracies in their formulation. Analyzing their loss curves gives insight into possible mistakes in the loss functions and better performance can be obtained by using the co-training like setup we propose, where two models are trained in a co-operative rather than an adversarial fashion.""","""The manuscript centers on a critique of IRGAN, a recently proposed extension of GANs to the information retrieval setting, and introduces a competing procedure. Reviewers found the findings and the proposed alternative to be interesting and in one case described the findings as ""illuminating"", but were overall unsatisfied with the depth of the analysis, and in more than one case complained that too much of the manuscript is spent reviewing IRGAN, with not enough emphasis and detailed investigation of the paper's own contribution. Notational issues, certain gaps in the related work and experiments were addressed in a revision but the paper still reads as spending a bit too much time on background relative to the contributions. Two reviewers seemed to agree that IRGAN's significance made at least some of the focus on it justifiable, but one remarked that SIGIR may be a better venue for this line of work (the AC doesn't necessarily agree). Given the nature of the changes and the status of the manuscript following revision, it does seem like a more comprehensive rewrite and reframing would be necessary to truly satisfy all reviewer concerns. I therefore recommend against acceptance at this point in time."""
815,"""Woulda, Coulda, Shoulda: Counterfactually-Guided Policy Search""","['reinforcement learning', 'generative models', 'model-based reinforcement learning', 'causal inference']","""Learning policies on data synthesized by models can in principle quench the thirst of reinforcement learning algorithms for large amounts of real experience, which is often costly to acquire. However, simulating plausible experience de novo is a hard problem for many complex environments, often resulting in biases for model-based policy evaluation and search. Instead of de novo synthesis of data, here we assume logged, real experience and model alternative outcomes of this experience under counterfactual actions, i.e. actions that were not actually taken. Based on this, we propose the Counterfactually-Guided Policy Search (CF-GPS) algorithm for learning policies in POMDPs from off-policy experience. It leverages structural causal models for counterfactual evaluation of arbitrary policies on individual off-policy episodes. CF-GPS can improve on vanilla model-based RL algorithms by making use of available logged data to de-bias model predictions. In contrast to off-policy algorithms based on Importance Sampling which re-weight data, CF-GPS leverages a model to explicitly consider alternative outcomes, allowing the algorithm to make better use of experience data. We find empirically that these advantages translate into improved policy evaluation and search results on a non-trivial grid-world task. Finally, we show that CF-GPS generalizes the previously proposed Guided Policy Search and that reparameterization-based algorithms such Stochastic Value Gradient can be interpreted as counterfactual methods.""","""see my comment to the authors below"""
816,"""Reinforced Pipeline Optimization: Behaving Optimally with Non-Differentiabilities""","['Pipeline Optimization', 'Reinforcement Learning', 'Stochastic Computation Graph', 'Faster R-CNN']","""Many machine learning systems are implemented as pipelines. A pipeline is essentially a chain/network of information processing units. As information flows in and out and gradients vice versa, ideally, a pipeline can be trained end-to-end via backpropagation provided with the right supervision and loss function. However, this is usually impossible in practice, because either the loss function itself may be non-differentiable, or there may exist some non-differentiable units. One popular way to superficially resolve this issue is to separate a pipeline into a set of differentiable sub-pipelines and train them with isolated loss functions. Yet, from a decision-theoretical point of view, this is equivalent to making myopic decisions using ad hoc heuristics along the pipeline while ignoring the real utility, which prevents the pipeline from behaving optimally. In this paper, we show that by converting a pipeline into a stochastic counterpart, it can then be trained end-to-end in the presence of non-differentiable parts. Thus, the resulting pipeline is optimal under certain conditions with respect to any criterion attached to it. In experiments, we apply the proposed approach - reinforced pipeline optimization - to Faster R-CNN, a state-of-the-art object detection pipeline, and obtain empirically near-optimal object detectors consistent with its base design in terms of mean average precision.""","""The work proposes a method for smoothing a non-differentiable machine learning pipeline (such as the Faster-RCNN detector) using policy gradient. Unfortunately, the reviewers identified a number of critical issues, including no significant improvement beyond existing works. The authors did not provide a rebuttal for these critical issues. """
817,"""Learning Joint Wasserstein Auto-Encoders for Joint Distribution Matching""","['joint distribution matching', 'image-to-image translation', 'video-to-video synthesis', 'Wasserstein distance']","""We study the joint distribution matching problem which aims at learning bidirectional mappings to match the joint distribution of two domains. This problem occurs in unsupervised image-to-image translation and video-to-video synthesis tasks, which, however, has two critical challenges: (i) it is difficult to exploit sufficient information from the joint distribution; (ii) how to theoretically and experimentally evaluate the generalization performance remains an open question. To address the above challenges, we propose a new optimization problem and design a novel Joint Wasserstein Auto-Encoders (JWAE) to minimize the Wasserstein distance of the joint distributions in two domains. We theoretically prove that the generalization ability of the proposed method can be guaranteed by minimizing the Wasserstein distance of joint distributions. To verify the generalization ability, we apply our method to unsupervised video-to-video synthesis by performing video frame interpolation and producing visually smooth videos in two domains, simultaneously. Both qualitative and quantitative comparisons demonstrate the superiority of our method over several state-of-the-arts.""","""This paper proposes a new image to image translation technique, presenting a theoretical extension of Wasserstein GANs to the bidirectional mapping case. Although the work presents promise, the extent of miscommunication and errors of the original presentation was too great to confidently conclude about the contribution of this work. The authors have already included extensive edits and comments in response to the reviews to improve the clarity of method, experiments and statement of contribution. We encourage the authors to further incorporate the suggestions and seek to clarify points of confusion from other reviewers and submit a revised version to a future conference."""
818,"""Evaluation Methodology for Attacks Against Confidence Thresholding Models""",['adversarial examples'],"""Current machine learning algorithms can be easily fooled by adversarial examples. One possible solution path is to make models that use confidence thresholding to avoid making mistakes. Such models refuse to make a prediction when they are not confident of their answer. We propose to evaluate such models in terms of tradeoff curves with the goal of high success rate on clean examples and low failure rate on adversarial examples. Existing untargeted attacks developed for models that do not use confidence thresholding tend to underestimate such models' vulnerability. We propose the MaxConfidence family of attacks, which are optimal in a variety of theoretical settings, including one realistic setting: attacks against linear models. Experiments show the attack attains good results in practice. We show that simple defenses are able to perform well on MNIST but not on CIFAR, contributing further to previous calls that MNIST should be retired as a benchmarking dataset for adversarial robustness research. We release code for these evaluations as part of the cleverhans (Papernot et al 2018) library (ICLR reviewers should be careful not to look at who contributed these features to cleverhans to avoid de-anonymizing this submission).""","""The reviewers agree the paper is not ready for publication. """
819,"""Learning shared manifold representation of images and attributes for generalized zero-shot learning""","['zero-shot learning', 'variational autoencoders']","""Many of the zero-shot learning methods have realized predicting labels of unseen images by learning the relations between images and pre-defined class-attributes. However, recent studies show that, under the more realistic generalized zero-shot learning (GZSL) scenarios, these approaches severely suffer from the issue of biased prediction, i.e., their classifier tends to predict all the examples from both seen and unseen classes as one of the seen classes. The cause of this problem is that they cannot properly learn a mapping to the representation space generalized to the unseen classes since the training set does not include any unseen class information. To solve this, we propose a concept to learn a mapping that embeds both images and attributes to the shared representation space that can be generalized even for unseen classes by interpolating from the information of seen classes, which we refer to shared manifold learning. Furthermore, we propose modality invariant variational autoencoders, which can perform shared manifold learning by training variational autoencoders with both images and attributes as inputs. The empirical validation of well-known datasets in GZSL shows that our method achieves the significantly superior performances to the existing relation-based studies.""","""The paper addresses generalized zero shot learning (test data contains examples from both seen as well as unseen classes) and proposes to learn a shared representation of images and attributes via multimodal variational autoencoders. The reviewers and AC note the following potential weaknesses: (1) low technical contribution, i.e. the proposed multimodal VAE model is very similar to Vedantam et al (2017) as noted by R2, and to JMVAE model by Suzuki et al, 2016, as noted by R1. The authors clarified in their response that indeed VAE in Vedantam et al (2017) is similar, but it has been used for image synthesis and not classification/GZSL. (2) Empirical evaluations and setup are not convincing (R2) and not clear -- R3 has provided a very detailed review and a follow up discussion raising several important concerns such as (i) absence of a validation set to test generalization, (ii) the hyperparameters set up; (iii) not clear advantages of learning a joint model as opposed to unidirectional mappings (R1 also supports this claim). The authors partially addressed some of these concerns in their response, however more in-depth analysis and major revision is required to assess the benefits and feasibility of the proposed approach. """
820,"""Harmonic Unpaired Image-to-image Translation""","['unpaired image-to-image translation', 'cyclegan', 'smoothness constraint']","""The recent direction of unpaired image-to-image translation is on one hand very exciting as it alleviates the big burden in obtaining label-intensive pixel-to-pixel supervision, but it is on the other hand not fully satisfactory due to the presence of artifacts and degenerated transformations. In this paper, we take a manifold view of the problem by introducing a smoothness term over the sample graph to attain harmonic functions to enforce consistent mappings during the translation. We develop HarmonicGAN to learn bi-directional translations between the source and the target domains. With the help of similarity-consistency, the inherent self-consistency property of samples can be maintained. Distance metrics defined on two types of features including histogram and CNN are exploited. Under an identical problem setting as CycleGAN, without additional manual inputs and only at a small training-time cost, HarmonicGAN demonstrates a significant qualitative and quantitative improvement over the state of the art, as well as improved interpretability. We show experimental results in a number of applications including medical imaging, object transfiguration, and semantic labeling. We outperform the competing methods in all tasks, and for a medical imaging task in particular our method turns CycleGAN from a failure to a success, halving the mean-squared error, and generating images that radiologists prefer over competing methods in 95% of cases.""","""The proposed method introduces a method for unsupervised image-to-image mapping, using a new term into the objective function that enforces consistency in similarity between image patches across domains. Reviewers left constructive and detailed comments, which, the authors have made substantial efforts to address. Reviewers have ranked paper as borderline, and in Area Chair's opinion, most major issued have been addressed: - R3&R2: Novelty compared to DistanceGAN/CRF limited: authors have clarified contributions in reference to DistanceGAN/CRF and demonstrated improved performance relative to several datasets. - R3&R1: Evaluation on additional datasets required: authors added evaluation on 4 more tasks - R3&R1: Details missing: authors added details. """
821,"""Soft Q-Learning with Mutual-Information Regularization""","['reinforcement learning', 'regularization', 'entropy', 'mutual information']","""We propose a reinforcement learning (RL) algorithm that uses mutual-information regularization to optimize a prior action distribution for better performance and exploration. Entropy-based regularization has previously been shown to improve both exploration and robustness in challenging sequential decision-making tasks. It does so by encouraging policies to put probability mass on all actions. However, entropy regularization might be undesirable when actions have significantly different importance. In this paper, we propose a theoretically motivated framework that dynamically weights the importance of actions by using the mutual-information. In particular, we express the RL problem as an inference problem where the prior probability distribution over actions is subject to optimization. We show that the prior optimization introduces a mutual-information regularizer in the RL objective. This regularizer encourages the policy to be close to a non-uniform distribution that assigns higher probability mass to more important actions. We empirically demonstrate that our method significantly improves over entropy regularization methods and unregularized methods.""","""The paper proposes a new RL algorithm (MIRL) in the control-as-inference framework that learns a state-independent action prior. A connection is provided to mutual information regularization. Compared to entropic regularization, this approach is expected to work better when actions have significantly different importance. The algorithm is shown to beat baselines in 11 out of 19 Atari games. The paper is well written. The derivation is novel, and the resulting algorithm is interesting and has good empirical results. A few concerns were raised in initial reviews, including certain questions about experiments and potential negative impacts of the use of nonuniform action priors in MIRL. The author responses and the new version were quite helpful, and all reviewers agree the paper is an interesting contribution. In a revised version, the authors are encouraged to (1) include a discussion of when MIRL might fail, and (2) improve the related work section to compare the proposed method to other entropy regularized RL (sometimes under a different name in the literature), for example the following recent works and the references therein: pseudo-url pseudo-url pseudo-url pseudo-url"""
822,"""Model Comparison for Semantic Grouping""","['model comparison', 'semantic similarity', 'STS', 'von Mises-Fisher', 'Information Theoretic Criteria']","""We introduce a probabilistic framework for quantifying the semantic similarity between two groups of embeddings. We formulate the task of semantic similarity as a model comparison task in which we contrast a generative model which jointly models two sentences versus one that does not. We illustrate how this framework can be used for the Semantic Textual Similarity tasks using clear assumptions about how the embeddings of words are generated. We apply information criteria based model comparison to overcome the shortcomings of Bayesian model comparison, whilst still penalising model complexity. We achieve competitive results by applying the proposed framework with an appropriate choice of likelihood on the STS datasets.""","""This paper presents a novel family of probabilistic approaches to computing the similarities between two sentences using bag-of-embeddings representations, and presents evaluations on a standard benchmark to demonstrate the effectiveness of the approach. While there seem to be no substantial disputes about the soundness of the paper in its current form, the reviewers were not convinced by the broad motivation for the approach, and did not find the empirical results compelling enough to serve as a motivation on its own. Given that, no reviewer was willing to argue that this paper makes an important enough contribution to be accepted. It is unfortunate that one of the assigned reviewersby their own admissionwas not well qualified to review it and that a second reviewer did not submit a review at all, necessitating a late fill-in review (thank you, anonymous emergency reviewer!). However, the paper was considered seriously: I can attest that both of the two higher-confidence reviewers are well qualified to review work on problems and methods like these."""
823,"""Countdown Regression: Sharp and Calibrated Survival Predictions""",[],"""Personalized probabilistic forecasts of time to event (such as mortality) can be crucial in decision making, especially in the clinical setting. Inspired by ideas from the meteorology literature, we approach this problem through the paradigm of maximizing sharpness of prediction distributions, subject to calibration. In regression problems, it has been shown that optimizing the continuous ranked probability score (CRPS) instead of maximum likelihood leads to sharper prediction distributions while maintaining calibration. We introduce the Survival-CRPS, a generalization of the CRPS to the time to event setting, and present right-censored and interval-censored variants. To holistically evaluate the quality of predicted distributions over time to event, we present the scale agnostic Survival-AUPRC evaluation metric, an analog to area under the precision-recall curve. We apply these ideas by building a recurrent neural network for mortality prediction, using an Electronic Health Record dataset covering millions of patients. We demonstrate signicant benets in models trained by the Survival-CRPS objective instead of maximum likelihood.""","""All reviewers agree to reject. While there were many positive points to this work, reviewers believed that it was not yet ready for acceptance."""
824,"""Learning Grid Cells as Vector Representation of Self-Position Coupled with Matrix Representation of Self-Motion""",[],"""This paper proposes a representational model for grid cells. In this model, the 2D self-position of the agent is represented by a high-dimensional vector, and the 2D self-motion or displacement of the agent is represented by a matrix that transforms the vector. Each component of the vector is a unit or a cell. The model consists of the following three sub-models. (1) Vector-matrix multiplication. The movement from the current position to the next position is modeled by matrix-vector multi- plication, i.e., the vector of the next position is obtained by multiplying the matrix of the motion to the vector of the current position. (2) Magnified local isometry. The angle between two nearby vectors equals the Euclidean distance between the two corresponding positions multiplied by a magnifying factor. (3) Global adjacency kernel. The inner product between two vectors measures the adjacency between the two corresponding positions, which is defined by a kernel function of the Euclidean distance between the two positions. Our representational model has explicit algebra and geometry. It can learn hexagon patterns of grid cells, and it is capable of error correction, path integral and path planning.""","""The authors have presented a simple yet elegant model to learn grid-like responses to encode spatial position, relying only on relative Euclidean distances to train the model, and achieving a good path integration accuracy. The model is simpler than recent related work and uses a structure of 'disentangled blocks' to achieve multi-scale grids rather than requiring dropout or injected noise. The paper is clearly written and it is intriguing to get down to the fundamentals of the grid code. On the negative side, the section on planning does not hold up as well and makes unverifiable claims, and one reviewer suggests that this section be replaced altogether by additional analysis of the grid model. Another reviewer points out that the authors have missed an opportunity to give a theoretical perspective on their model. Although there are aspects of the work which could be improved, the AC and all reviewers are in favor of acceptance of this paper."""
825,"""Sample Efficient Imitation Learning for Continuous Control""","['Imitation Learning', 'Continuous Control', 'Reinforcement Learning', 'Inverse Reinforcement Learning', 'Conditional Generative Adversarial Network']","""The goal of imitation learning (IL) is to enable a learner to imitate expert behavior given expert demonstrations. Recently, generative adversarial imitation learning (GAIL) has shown significant progress on IL for complex continuous tasks. However, GAIL and its extensions require a large number of environment interactions during training. In real-world environments, the more an IL method requires the learner to interact with the environment for better imitation, the more training time it requires, and the more damage it causes to the environments and the learner itself. We believe that IL algorithms could be more applicable to real-world problems if the number of interactions could be reduced. In this paper, we propose a model-free IL algorithm for continuous control. Our algorithm is made up mainly three changes to the existing adversarial imitation learning (AIL) methods (a) adopting off-policy actor-critic (Off-PAC) algorithm to optimize the learner policy, (b) estimating the state-action value using off-policy samples without learning reward functions, and (c) representing the stochastic policy function so that its outputs are bounded. Experimental results show that our algorithm achieves competitive results with GAIL while significantly reducing the environment interactions.""","""The paper proposes a simple method for improving the sample efficiency of GAIL, essentially a way of turning inverse reinforcement learning into classification. As reviewers noted, the method is based on a simple idea with potentially broad applicability. Concerns were raised about the multiple components of the system and what each contributed, and missing pointers to the literature. A baseline wherein GAIL is initialized with behaviour cloning, although only suggested but not tried in previous works. The authors did, however, attempt this setting and found it to hurt, not help, performance. I find this surprising and would urge the authors to validate that this isn't merely an uninteresting artifact of the setup, however I commend the authors for trying it and don't believe that a surprising result in this regard is a barrier to publication. As several reviewers did not provide feedback on revisions addressing their concerns, this Area Chair was left to determine to a large degree whether or not reviewer concerns were in fact addressed. I thank AnonReviewer4 for revisiting their review towards the end of the period, and concur with them that many of the concerns raised by reviewers have indeed been adequately dealt with. """
826,"""Distilled Agent DQN for Provable Adversarial Robustness""","['reinforcement learning', 'dqn', 'adversarial examples', 'robustness analysis', 'adversarial defense', 'robust learning', 'robust rl']","""As deep neural networks have become the state of the art for solving complex reinforcement learning tasks, susceptibility to perceptual adversarial examples have become a concern. The transferability of adversarial examples is known to enable attacks capable of tricking the agent into bad states. In this work we demonstrate a simple poisoning attack able to keep deep RL from learning, and into fooling it when trained with defense methods commonly used for classification tasks. We then propose an algorithm called DadQN, based on deep Q-networks, which enables the use of stronger defenses, including defenses enabling the first ever on-line robustness certification of a deep RL agent.""","""Reviewers had several concerns about the paper, primary among them being limited novelty of the approach. The reviewers have offered suggestions for improving the work which we encourage the authors to read and consider."""
827,"""On the Ineffectiveness of Variance Reduced Optimization for Deep Learning""","['machine learning', 'optimization', 'variance reduction']","""The application of stochastic variance reduction to optimization has shown remarkable recent theoretical and practical success. The applicability of these techniques to the hard non-convex optimization problems encountered during training of modern deep neural networks is an open problem. We show that naive application of the SVRG technique and related approaches fail, and explore why.""","""All reviewers agreed that this paper addresses an important question in deep learning (why doesn't SVRG help for deep learning)? But the paper still has some issues that need to be addressed before publication, thus the AC recommends ""revise and resubmit""."""
828,"""Teaching to Teach by Structured Dark Knowledge""","['teaching to teach', 'dark knowledge', 'curriculum learning', 'teaching']","""To educate hyper deep learners, \emph{Curriculum Learnings} (CLs) require either human heuristic participation or self-deciding the difficulties of training instances. These coaching manners are blind to the coherent structures among examples, categories, and tasks, which are pregnant with more knowledgeable curriculum-routed teachers. In this paper, we propose a general methodology \emph{Teaching to Teach} (T2T). T2T is facilitated by \emph{Structured Dark Knowledge} (SDK) that constitutes a communication protocol between structured knowledge prior and teaching strategies. On one hand, SDK adaptively extracts structured knowledge by selecting a training subset consistent with the previous teaching decisions. On the other hand, SDK teaches curriculum-agnostic teachers by transferring this knowledge to update their teaching policy. This virtuous cycle can be flexibly-deployed in most existing CL platforms and more importantly, very generic across various structured knowledge characteristics, e.g., diversity, complementarity, and causality. We evaluate T2T across different learners, teachers, and tasks, which significantly demonstrates that structured knowledge can be inherited by the teachers to further benefit learners' training. ""","""This work proposes a ""teaching to teach"" (T2T) method to incorporate structured prior knowledge into the teaching model for machine learning tasks. This is an interesting and timely topic. Unfortunately, among many other issues, this paper is fairly poorly writing and can benefit from a significant rewriting. The authors did not provide a rebuttal and hence we recommend a rejection. """
829,"""Learning with Random Learning Rates.""","['step size', 'stochastic gradient descent', 'hyperparameter tuning']","""Hyperparameter tuning is a bothersome step in the training of deep learning mod- els. One of the most sensitive hyperparameters is the learning rate of the gradient descent. We present the All Learning Rates At Once (Alrao) optimization method for neural networks: each unit or feature in the network gets its own learning rate sampled from a random distribution spanning several orders of magnitude. This comes at practically no computational cost. Perhaps surprisingly, stochastic gra- dient descent (SGD) with Alrao performs close to SGD with an optimally tuned learning rate, for various architectures and problems. Alrao could save time when testing deep learning models: a range of models could be quickly assessed with Alrao, and the most promising models could then be trained more extensively. This text comes with a PyTorch implementation of the method, which can be plugged on an existing PyTorch model.""","""The paper proposes a new optimization approach for neural nets where, instead of a fixed learning rate (often hard to tune), there is one learning rate per unit, randomly sampled from a distribution. Reviewers think the idea is novel, original and simple. Overall, reviewers found the experiments unconvincing enough in practice. I found the paper really borderline, and decided to side with the reviewers in rejecting the paper."""
830,"""Structured Adversarial Attack: Towards General Implementation and Better Interpretability""",[],"""When generating adversarial examples to attack deep neural networks (DNNs), Lp norm of the added perturbation is usually used to measure the similarity between original image and adversarial example. However, such adversarial attacks perturbing the raw input spaces may fail to capture structural information hidden in the input. This work develops a more general attack model, i.e., the structured attack (StrAttack), which explores group sparsity in adversarial perturbation by sliding a mask through images aiming for extracting key spatial structures. An ADMM (alternating direction method of multipliers)-based framework is proposed that can split the original problem into a sequence of analytically solvable subproblems and can be generalized to implement other attacking methods. Strong group sparsity is achieved in adversarial perturbations even with the same level of Lp-norm distortion (p {1,2,}) as the state-of-the-art attacks. We demonstrate the effectiveness of StrAttack by extensive experimental results on MNIST, CIFAR-10 and ImageNet. We also show that StrAttack provides better interpretability (i.e., better correspondence with discriminative image regions) through adversarial saliency map (Paper-not et al., 2016b) and class activation map (Zhou et al., 2016).""","""This paper contributes a novel approach to evaluating the robustness of DNN based on structured sparsity to exploit the underlying structure of the image and introduces a method to solve it. The proposed approach is well evaluated and the authors answered the main concerns of the reviewers. """
831,"""Efficient Training on Very Large Corpora via Gramian Estimation""","['similarity learning', 'pairwise learning', 'matrix factorization', 'Gramian estimation', 'variance reduction', 'neural embedding models', 'recommender systems']","""We study the problem of learning similarity functions over very large corpora using neural network embedding models. These models are typically trained using SGD with random sampling of unobserved pairs, with a sample size that grows quadratically with the corpus size, making it expensive to scale. We propose new efficient methods to train these models without having to sample unobserved pairs. Inspired by matrix factorization, our approach relies on adding a global quadratic penalty and expressing this term as the inner-product of two generalized Gramians. We show that the gradient of this term can be efficiently computed by maintaining estimates of the Gramians, and develop variance reduction schemes to improve the quality of the estimates. We conduct large-scale experiments that show a significant improvement both in training time and generalization performance compared to sampling methods.""","""This paper presents methods to scale learning of embedding models estimated using neural networks. The main idea is to work with Gram matrices whose sizes depend on the length of the embedding. Building upon existing works like SAG algorithm, the paper proposes two new stochastic methods for learning using stochastic estimates of Gram matrices. Reviewers find the paper interesting and useful, although have given many suggestions to improve the presentation and experiments. For this reason, I recommend to accept this paper. A small note: SAG algorithm was originally proposed in 2013. The paper only cites the 2017 version. Please include the 2013 version as well. """
832,"""Rigorous Agent Evaluation: An Adversarial Approach to Uncover Catastrophic Failures""","['agent evaluation', 'adversarial examples', 'robustness', 'safety', 'reinforcement learning']","""This paper addresses the problem of evaluating learning systems in safety critical domains such as autonomous driving, where failures can have catastrophic consequences. We focus on two problems: searching for scenarios when learned agents fail and assessing their probability of failure. The standard method for agent evaluation in reinforcement learning, Vanilla Monte Carlo, can miss failures entirely, leading to the deployment of unsafe agents. We demonstrate this is an issue for current agents, where even matching the compute used for training is sometimes insufficient for evaluation. To address this shortcoming, we draw upon the rare event probability estimation literature and propose an adversarial evaluation approach. Our approach focuses evaluation on adversarially chosen situations, while still providing unbiased estimates of failure probabilities. The key difficulty is in identifying these adversarial situations -- since failures are rare there is little signal to drive optimization. To solve this we propose a continuation approach that learns failure modes in related but less robust agents. Our approach also allows reuse of data already collected for training the agent. We demonstrate the efficacy of adversarial evaluation on two standard domains: humanoid control and simulated driving. Experimental results show that our methods can find catastrophic failures and estimate failures rates of agents multiple orders of magnitude faster than standard evaluation schemes, in minutes to hours rather than days.""",""" * Strengths The paper addresses a timely topic, and reviewers generally agreed that the approach is reasonable and the experiments are convincing. Reviewers raised a number of specific concerns (which could be addressed in a revised version or future work), described below. * Weaknesses Some reviewers were concerned the baselines are weak. Several reviewers were concerned that relying on failures observed during training could create issues by narrowing the proposal distribution (Reviewer 3 characterizes this in a particularly precise manner). In addition, there was a general feeling that more steps are needed before the method can be used in practice (but this could be said of most research). * Recommendation All reviewers agreed that the paper should be accepted, although there was also consensus that the paper would benefit from stronger baselines and more close attention to issues that could be caused by an overly narrow proposal distribution. The authors should consider addressing or commenting on these issues in the final version."""
833,"""Post Selection Inference with Incomplete Maximum Mean Discrepancy Estimator""","['Maximum Mean Discrepancy', 'Selective Inference', 'Feature Selection', 'GAN']","""Measuring divergence between two distributions is essential in machine learning and statistics and has various applications including binary classification, change point detection, and two-sample test. Furthermore, in the era of big data, designing divergence measure that is interpretable and can handle high-dimensional and complex data becomes extremely important. In this paper, we propose a post selection inference (PSI) framework for divergence measure, which can select a set of statistically significant features that discriminate two distributions. Specifically, we employ an additive variant of maximum mean discrepancy (MMD) for features and introduce a general hypothesis test for PSI. A novel MMD estimator using the incomplete U-statistics, which has an asymptotically normal distribution (under mild assumptions) and gives high detection power in PSI, is also proposed and analyzed theoretically. Through synthetic and real-world feature selection experiments, we show that the proposed framework can successfully detect statistically significant features. Last, we propose a sample selection framework for analyzing different members in the Generative Adversarial Networks (GANs) family. ""","""The submission evaluates maximum mean discrepancy estimators for post selection inference. It combines two contributions: (i) it proposes an incomplete u-statistic estimator for MMD, (ii) it evaluates this and existing estimators in a post selection inference setting. The method extends the post selection inference approach of (Lee et al. 2016) to the current u-statistic approach for MMD. The top-k selection problem is phrased as a linear constraint reducing it to the problem of Lee et al. The approach is illustrated on toy examples and a GAN application. The main criticism of the problem is the novelty of the paper. R1 feels that it is largely just the combination of two known approaches (although it appears that the incomplete estimator is key), while R3 was significantly more impressed. Both are senior experts in the topic. On the balance, the reviewers were more positive than negative. R2 felt that the authors comments helped to address their concerns, while R3 gave detailed arguments in favor of the submission and championed the paper. The paper provides an additional interesting framework for evaluation of estimators, and considers their application in a broader context of post-selection inference."""
834,"""Learning models for visual 3D localization with implicit mapping""","['generative learning', 'generative models', 'generative query networks', 'camera re-localization']","""We consider learning based methods for visual localization that do not require the construction of explicit maps in the form of point clouds or voxels. The goal is to learn an implicit representation of the environment at a higher, more abstract level, for instance that of objects. We propose to use a generative approach based on Generative Query Networks (GQNs, Eslami et al. 2018), asking the following questions: 1) Can GQN capture more complex scenes than those it was originally demonstrated on? 2) Can GQN be used for localization in those scenes? To study this approach we consider procedurally generated Minecraft worlds, for which we can generate images of complex 3D scenes along with camera pose coordinates. We first show that GQNs, enhanced with a novel attention mechanism can capture the structure of 3D scenes in Minecraft, as evidenced by their samples. We then apply the models to the localization problem, comparing the results to a discriminative baseline, and comparing the ways each approach captures the task uncertainty. ""","""The paper proposes a method that learns mapping implicitly, by using a generative query network of Eslami et al. with an attention mechanism to learn to predict egomotion. The empirical findings is that training for egomotion estimation alongside the generative task of view prediction helps over a discriminative baseline, that does not consoder view prediction. The model is tested in Minecraft environments. A comparison to some baseline SLAM-like method, e.g., a method based on bundle adjustment, would be important to include despite beliefs of the authors that eventually learning-based methods would win over geometric methods. For example, potentially environments with changes can be considered, which will cause the geometric method to fail, but the proposed learning-based method to succeed. Moreover, there are currently learning based methods for the re-localization problem that the paper would be important to compare against (instead of just cite), such as ""MapNet: An Allocentric Spatial Memory for Mapping Environments"" of Henriques et al. and ""Active Neural Localization"" of Chaplot et al. . In particular, Mapnet has a generative interpretation by using cross-convolutions as part of its architecture, which generalize very well, and which consider the geometric formation process. The paper makes a big distinction between generative and discriminative, however the architectural details behind the egomotion estimation network are potentially more or equally important to the loss used. This means, different discriminative networks depending on their architecture may perform very differently. Thus, it would be important to present quantitative results against such methods that use cross-convolutions for egomotion estimation/re-localization. """
835,"""Gradient Descent Happens in a Tiny Subspace""","['Gradient Descent', 'Hessian', 'Deep Learning']","""We show that in a variety of large-scale deep learning scenarios the gradient dynamically converges to a very small subspace after a short period of training. The subspace is spanned by a few top eigenvectors of the Hessian (equal to the number of classes in the dataset), and is mostly preserved over long periods of training. A simple argument then suggests that gradient descent may happen mostly in this subspace. We give an example of this effect in a solvable model of classification, and we comment on possible implications for optimization and learning.""","""The paper is overally interesting and addresses an important problem, however reviewers ask for more rigorous empirical study and less restrictive settings."""
836,"""GenEval: A Benchmark Suite for Evaluating Generative Models""","['generative models', 'GAN', 'VAE', 'Real NVP']","""Generative models are important for several practical applications, from low level image processing tasks, to model-based planning in robotics. More generally, the study of generative models is motivated by the long-standing endeavor to model uncertainty and to discover structure by leveraging unlabeled data. Unfortunately, the lack of an ultimate task of interest has hindered progress in the field, as there is no established way to compare models and, often times, evaluation is based on mere visual inspection of samples drawn from such models. In this work, we aim at addressing this problem by introducing a new benchmark evaluation suite, dubbed \textit{GenEval}. GenEval hosts a large array of distributions capturing many important properties of real datasets, yet in a controlled setting, such as lower intrinsic dimensionality, multi-modality, compositionality, independence and causal structure. Any model can be easily plugged for evaluation, provided it can generate samples. Our extensive evaluation suggests that different models have different strenghts, and that GenEval is a great tool to gain insights about how models and metrics work. We offer GenEval to the community~\footnote{Available at: \it{coming soon}.} and believe that this benchmark will facilitate comparison and development of new generative models.""","""The paper introduces a benchmark suite providing a series of synthetic distributions and metrics for the evaluation of generative models. While providing such a tool-kit is interesting and helpful and it extends existing approaches for evaluating generative models on simple distributions, it seems not to allow for very different additional conclusions or insights.This limits the paper's significance. Adding more problems and metrics to the benchmark suite would make it more convincing."""
837,"""Stable Recurrent Models""","['stability', 'gradient descent', 'non-convex optimization', 'recurrent neural networks']","""Stability is a fundamental property of dynamical systems, yet to this date it has had little bearing on the practice of recurrent neural networks. In this work, we conduct a thorough investigation of stable recurrent models. Theoretically, we prove stable recurrent neural networks are well approximated by feed-forward networks for the purpose of both inference and training by gradient descent. Empirically, we demonstrate stable recurrent models often perform as well as their unstable counterparts on benchmark sequence tasks. Taken together, these findings shed light on the effective power of recurrent networks and suggest much of sequence learning happens, or can be made to happen, in the stable regime. Moreover, our results help to explain why in many cases practitioners succeed in replacing recurrent models by feed-forward models. ""","""The paper presents both theoretical analysis (based upon lambda-stability) and experimental evidence on stability of recurrent neural networks. The results are convincing but is concerns with a restricted definition of stability. Even with this restriction acceptance is recommended. """
838,"""Towards Decomposed Linguistic Representation with Holographic Reduced Representation""",[],"""The vast majority of neural models in Natural Language Processing adopt a form of structureless distributed representations. While these models are powerful at making predictions, the representational form is rather crude and does not provide insights into linguistic structures. In this paper we introduce novel language models with representations informed by the framework of Holographic Reduced Representation (HRR). This allows us to inject structures directly into our word-level and chunk-level representations. Our analyses show that by using HRR as a structured compositional representation, our models are able to discover crude linguistic roles, which roughly resembles a classic division between syntax and semantics.""","""This paper proposes the use of holographic reduced representations in language modeling, which allows for a cleaner decomposition of various linguistic traits in the representation. Results show improvements over baseline language models, and analysis shows that the representations are indeed decomposing as expected. The main reviewer concern was the lack of strength of the baseline, although the authors stress that they were using the default baseline from TensorFlow, which seems like it will be reasonable to me. Another concern is that there is other work on using HRR to disentangle syntax and semantics in representations for language (e.g. ""Distributed Tree Kernels"" ICML 2012, but also others), that has not been considered. Based on this, this seems like a very borderline case. Given that no reviewer is pushing strongly for the paper I'm leaning towards not recommending acceptance, but I could very easily see the paper being accepted as well."""
839,"""Identifying and Controlling Important Neurons in Neural Machine Translation""","['neural machine translation', 'individual neurons', 'unsupervised', 'analysis', 'correlation', 'translation control', 'distributivity', 'localization']","""Neural machine translation (NMT) models learn representations containing substantial linguistic information. However, it is not clear if such information is fully distributed or if some of it can be attributed to individual neurons. We develop unsupervised methods for discovering important neurons in NMT models. Our methods rely on the intuition that different models learn similar properties, and do not require any costly external supervision. We show experimentally that translation quality depends on the discovered neurons, and find that many of them capture common linguistic phenomena. Finally, we show how to control NMT translations in predictable ways, by modifying activations of individual neurons.""","""Strong points: -- Interesting, fairly systematic and novel analyses of recurrent NMT models, revealing individual neurons responsible for specific type of information (e.g., verb tense or gender) -- Interesting experiments showing how these neurons can be used to manipulate translations in specific ways (e.g., specifying the gender for a pronoun when the source sentence does not reveal it) -- The paper is well written Weak points -- Nothing serious (e.g., maybe interesting to test across multiple runs how stable these findings are). There is a consensus among the reviewers that this is a strong paper and should be accepted. """
840,"""Total Style Transfer with a Single Feed-Forward Network""","['Image Style Transfer', 'Deep Learning', 'Neural Network']","""Recent image style transferring methods achieved arbitrary stylization with input content and style images. To transfer the style of an arbitrary image to a content image, these methods used a feed-forward network with a lowest-scaled feature transformer or a cascade of the networks with a feature transformer of a corresponding scale. However, their approaches did not consider either multi-scaled style in their single-scale feature transformer or dependency between the transformed feature statistics across the cascade networks. This shortcoming resulted in generating partially and inexactly transferred style in the generated images. To overcome this limitation of partial style transfer, we propose a total style transferring method which transfers multi-scaled feature statistics through a single feed-forward process. First, our method transforms multi-scaled feature maps of a content image into those of a target style image by considering both inter-channel correlations in each single scaled feature map and inter-scale correlations between multi-scaled feature maps. Second, each transformed feature map is inserted into the decoder layer of the corresponding scale using skip-connection. Finally, the skip-connected multi-scaled feature maps are decoded into a stylized image through our trained decoder network.""","""The reviewers and this AC agree that the paper is not of acceptable form due to several issues: (1) limited novelty, (2) limited/unclear experimental validation, and (3) presentation issues."""
841,"""SOSELETO: A Unified Approach to Transfer Learning and Training with Noisy Labels""",['transfer learning'],"""We present SOSELETO (SOurce SELEction for Target Optimization), a new method for exploiting a source dataset to solve a classification problem on a target dataset. SOSELETO is based on the following simple intuition: some source examples are more informative than others for the target problem. To capture this intuition, source samples are each given weights; these weights are solved for jointly with the source and target classification problems via a bilevel optimization scheme. The target therefore gets to choose the source samples which are most informative for its own classification task. Furthermore, the bilevel nature of the optimization acts as a kind of regularization on the target, mitigating overfitting. SOSELETO may be applied to both classic transfer learning, as well as the problem of training on datasets with noisy labels; we show state of the art results on both of these problems.""","""The paper proposes an approach for transfer learning by assigning weights to source samples and learning these jointly with the network parameters. Reviewers had a few concerns about experiments, some of which have been addressed by the authors. The proposed approach is simple which is a positive but it is not evaluated on any of the regular transfer learning benchmarks (eg, the ones used in Kornblith et al., 2018 ""Do Better ImageNet Models Transfer Better?""). The tasks used in the paper, such as CIFAR noisy -> CIFAR and SVHN0-4 -> MNIST5-9, are artificially constructed, and the paper falls short of demonstrating the effectiveness of the approach on real settings. The paper is on the borderline with current scores and the lack of regular transfer learning benchmarks in the evaluations makes me lean towards not recommending acceptance. """
842,"""Unsupervised one-to-many image translation""","['Image-to-image', 'Translation', 'Unsupervised', 'Generation', 'Adversarial', 'Learning']","""We perform completely unsupervised one-sided image to image translation between a source domain pseudo-formula and a target domain pseudo-formula such that we preserve relevant underlying shared semantics (e.g., class, size, shape, etc). In particular, we are interested in a more difficult case than those typically addressed in the literature, where the source and target are ``far"" enough that reconstruction-style or pixel-wise approaches fail. We argue that transferring (i.e., \emph{translating}) said relevant information should involve both discarding source domain-specific information while incorporate target domain-specific information, the latter of which we model with a noisy prior distribution. In order to avoid the degenerate case where the generated samples are only explained by the prior distribution, we propose to minimize an estimate of the mutual information between the generated sample and the sample from the prior distribution. We discover that the architectural choices are an important factor to consider in order to preserve the shared semantic between pseudo-formula and pseudo-formula . We show state of the art results on the MNIST to SVHN task for unsupervised image to image translation.""","""The paper formulates the problem of unsupervised one-to-many image translation and addresses the problem by minimizing the mutual information. The reviewers and AC note the critical limitation of novelty and comparison of this paper to meet the high standard of ICLR. AC decided that the authors need more works to publish."""
843,"""Are adversarial examples inevitable?""","['adversarial examples', 'neural networks', 'security']","""A wide range of defenses have been proposed to harden neural networks against adversarial attacks. However, a pattern has emerged in which the majority of adversarial defenses are quickly broken by new attacks. Given the lack of success at generating robust defenses, we are led to ask a fundamental question: Are adversarial attacks inevitable? This paper analyzes adversarial examples from a theoretical perspective, and identifies fundamental bounds on the susceptibility of a classifier to adversarial attacks. We show that, for certain classes of problems, adversarial examples are inescapable. Using experiments, we explore the implications of theoretical guarantees for real-world problems and discuss how factors such as dimensionality and image complexity limit a classifier's robustness against adversarial examples. ""","""There's precious little work asking existential questions about adversarial examples, and so this work is most welcome. The work connects with deep results in probability to make simple and transparent claims about the inevitability of adversarial examples under some assumptions. The authors have addressed the key criticisms of the authors around clarity."""
844,"""Finding Mixed Nash Equilibria of Generative Adversarial Networks""","['GANs', 'mixed Nash equilibrium', 'mirror descent', 'sampling']","""We reconsider the training objective of Generative Adversarial Networks (GANs) from the mixed Nash Equilibria (NE) perspective. Inspired by the classical prox methods, we develop a novel algorithmic framework for GANs via an infinite-dimensional two-player game and prove rigorous convergence rates to the mixed NE. We then propose a principled procedure to reduce our novel prox methods to simple sampling routines, leading to practically efficient algorithms. Finally, we provide experimental evidence that our approach outperforms methods that seek pure strategy equilibria, such as SGD, Adam, and RMSProp, both in speed and quality.""","""While the authors made a strong rebuttal, none of the reviewers were particularly enthusiastic about the contributions of this paper and we unfortunately have to reject borderline papers. Concerns were expressed about the presentation, as well as the scalability of the approach. The AC encourages the authors to ""revise and resubmit""."""
845,"""End-to-End Multi-Lingual Multi-Speaker Speech Recognition""","['end-to-end ASR', 'multi-lingual ASR', 'multi-speaker ASR', 'code-switching', 'encoder-decoder', 'connectionist temporal classification']","""The expressive power of end-to-end automatic speech recognition (ASR) systems enables direct estimation of the character or word label sequence from a sequence of acoustic features. Direct optimization of the whole system is advantageous because it not only eliminates the internal linkage necessary for hybrid systems, but also extends the scope of potential application use cases by training the model for multiple objectives. Several multi-lingual ASR systems were recently proposed based on a monolithic neural network architecture without language-dependent modules, showing that modeling of multiple languages is well within the capabilities of an end-to-end framework. There has also been growing interest in multi-speaker speech recognition, which enables generation of multiple label sequences from single-channel mixed speech. In particular, a multi-speaker end-to-end ASR system that can directly model one-to-many mappings without additional auxiliary clues was recently proposed. In this paper, we propose an all-in-one end-to-end multi-lingual multi-speaker ASR system that integrates the capabilities of these two systems. The proposed model is evaluated using mixtures of two speakers generated by using 10 languages, including mixed-language utterances. ""","""The authors present a system for end-to-end multi-lingual and multi-speaker speech recognition. The presented method is based on multiple prior works that propose end-to-end models for multi-lingual ASR and multi-speaker ASR; the work combines these techniques and shows that a single system can do both with minimal changes. The main critique from the reviewers is that the paper lacks novelty. It builds heavily on existing work, and does not make any enough contributions to be accepted at ICLR. Furthermore, training and evaluations are all on simulated test sets that are not very realistic. So it is unclear how well the techniques would generalize to real use-cases. For these reasons, the recommendation is to reject the paper."""
846,"""Incremental Hierarchical Reinforcement Learning with Multitask LMDPs""","['Reinforcement learning', 'hierarchy', 'linear markov decision process', 'lmdl', 'subtask discovery', 'incremental']","""Exploration is a well known challenge in Reinforcement Learning. One principled way of overcoming this challenge is to find a hierarchical abstraction of the base problem and explore at these higher levels, rather than in the space of primitives. However, discovering a deep abstraction autonomously remains a largely unsolved problem, with practitioners typically hand-crafting these hierarchical control architectures. Recent work with multitask linear Markov decision processes, allows for the autonomous discovery of deep hierarchical abstractions, but operates exclusively in the offline setting. By extending this work, we develop an agent that is capable of incrementally growing a hierarchical representation, and using its experience to date to improve exploration.""","""The paper studies an interesting problem with a reasonable solution. However, reviewers feel that the technical contributions are somewhat incremental. Furthermore, the empirical study would have been stronger with more proper baselines (simple adaptation to the multitask setting), and on problems beyond the simple grid worlds. In addition, reviewers also find the presentation should be improved substantially."""
847,"""Learning Representations of Sets through Optimized Permutations""","['sets', 'representation learning', 'permutation invariance']","""Representations of sets are challenging to learn because operations on sets should be permutation-invariant. To this end, we propose a Permutation-Optimisation module that learns how to permute a set end-to-end. The permuted set can be further processed to learn a permutation-invariant representation of that set, avoiding a bottleneck in traditional set models. We demonstrate our model's ability to learn permutations and set representations with either explicit or implicit supervision on four datasets, on which we achieve state-of-the-art results: number sorting, image mosaics, classification from image mosaics, and visual question answering. ""","""The paper proposes an architecture to learn over sets, by proposing a way to have permutations differentiable end-to-end, hence learnable by gradient descent. Reviewers pointed out to the computational limitation (quadratic in the size of the set just to consider pairwise interactions, and cubic overall). One reviewer (with low confidence) though the approach was not novel but didn't appreciate the integration of learning-to-permute with a differentiable setting, so I decided to down-weight their score. Overall, I found the paper borderline but would propose to accept it if possible."""
848,"""NOODL: Provable Online Dictionary Learning and Sparse Coding""","['dictionary learning', 'provable dictionary learning', 'online dictionary learning', 'sparse coding', 'support recovery', 'iterative hard thresholding', 'matrix factorization', 'neural architectures', 'neural networks', 'noodl']","""We consider the dictionary learning problem, where the aim is to model the given data as a linear combination of a few columns of a matrix known as a dictionary, where the sparse weights forming the linear combination are known as coefficients. Since the dictionary and coefficients, parameterizing the linear model are unknown, the corresponding optimization is inherently non-convex. This was a major challenge until recently, when provable algorithms for dictionary learning were proposed. Yet, these provide guarantees only on the recovery of the dictionary, without explicit recovery guarantees on the coefficients. Moreover, any estimation error in the dictionary adversely impacts the ability to successfully localize and estimate the coefficients. This potentially limits the utility of existing provable dictionary learning methods in applications where coefficient recovery is of interest. To this end, we develop NOODL: a simple Neurally plausible alternating Optimization-based Online Dictionary Learning algorithm, which recovers both the dictionary and coefficients exactly at a geometric rate, when initialized appropriately. Our algorithm, NOODL, is also scalable and amenable for large scale distributed implementations in neural architectures, by which we mean that it only involves simple linear and non-linear operations. Finally, we corroborate these theoretical results via experimental evaluation of the proposed algorithm with the current state-of-the-art techniques.""","""Alternating minimization is surprisingly effective for low-rank matrix factorization and dictionary learning problems. Better theoretical characterization of these methods is well motivated. This paper fills up a gap by providing simultaneous guarantees for support recovery as well as coefficient estimates for linearly convergence to the true factors, in the online learning setting. The reviewers are largely in agreement that the paper is well written and makes a valuable contribution. The authors are advised to address some of the review comments around relationship to prior work highlighting novelties."""
849,"""The Forward-Backward Embedding of Directed Graphs""","['Graph embedding', 'SVD', 'random walk', 'co-clustering']","""We introduce a novel embedding of directed graphs derived from the singular value decomposition (SVD) of the normalized adjacency matrix. Specifically, we show that, after proper normalization of the singular vectors, the distances between vectors in the embedding space are proportional to the mean commute times between the corresponding nodes by a forward-backward random walk in the graph, which follows the edges alternately in forward and backward directions. In particular, two nodes having many common successors in the graph tend to be represented by close vectors in the embedding space. More formally, we prove that our representation of the graph is equivalent to the spectral embedding of some co-citation graph, where nodes are linked with respect to their common set of successors in the original graph. The interest of our approach is that it does not require to build this co-citation graph, which is typically much denser than the original graph. Experiments on real datasets show the efficiency of the approach. ""","""The reviewers are unanimous in their assessment that the paper lacks originality in its current form to be publishable at ICLR-2018."""
850,"""Robust Determinantal Generative Classifier for Noisy Labels and Adversarial Attacks""","['Noisy Labels', 'Adversarial Attacks', 'Generative Models']","""Large-scale datasets may contain significant proportions of noisy (incorrect) class labels, and it is well-known that modern deep neural networks poorly generalize from such noisy training datasets. In this paper, we propose a novel inference method, Deep Determinantal Generative Classifier (DDGC), which can obtain a more robust decision boundary under any softmax neural classifier pre-trained on noisy datasets. Our main idea is inducing a generative classifier on top of hidden feature spaces of the discriminative deep model. By estimating the parameters of generative classifier using the minimum covariance determinant estimator, we significantly improve the classification accuracy, with neither re-training of the deep model nor changing its architectures. In particular, we show that DDGC not only generalizes well from noisy labels, but also is robust against adversarial perturbations due to its large margin property. Finally, we propose the ensemble version ofDDGC to improve its performance, by investigating the layer-wise characteristics of generative classifier. Our extensive experimental results demonstrate the superiority of DDGC given different learning models optimized by various training techniques to handle noisy labels or adversarial samples. For instance, on CIFAR-10 dataset containing 45% noisy training labels, we improve the test accuracy of a deep model optimized by the state-of-the-art noise-handling training method from33.34% to 43.02%.""","""While the paper contains interesting ideas, the reviewers agree the experimental study can be improved. """
851,"""(Unconstrained) Beam Search is Sensitive to Large Search Discrepancies""","['beam search', 'sequence models', 'search', 'sequence to sequence']","""Beam search is the most popular inference algorithm for decoding neural sequence models. Unlike greedy search, beam search allows for a non-greedy local decisions that can potentially lead to a sequence with a higher overall probability. However, previous work found that the performance of beam search tends to degrade with large beam widths. In this work, we perform an empirical study of the behavior of the beam search algorithm across three sequence synthesis tasks. We find that increasing the beam width leads to sequences that are disproportionately based on early and highly non-greedy decisions. These sequences typically include a very low probability token that is followed by a sequence of tokens with higher (conditional) probability leading to an overall higher probability sequence. However, as beam width increases, such sequences are more likely to have a lower evaluation score. Based on our empirical analysis we propose to constrain the beam search from taking highly non-greedy decisions early in the search. We evaluate two methods to constrain the search and show that constrained beam search effectively eliminates the problem of beam search degradation and in some cases even leads to higher evaluation scores. Our results generalize and improve upon previous observations on copies and training set predictions.""","""This paper examines a concept (also coined by the paper) of ""search discrepancies"" where the search algorithm behaves differently with large beam sizes. It then proposes heuristics to help prevent the model from performing worse when the size of the beam is increased. I think there are some interesting insights in this paper with respect to how search works in modern neural models, but most reviewers (and me) were concerned by the heuristic approach taken to fix these errors. I still think that within a search paper, a clear separation between modeling errors and search errors is useful, and adding heuristics on top has a potential to making things more complicated down the road when, for example, we change our model or we change our training algorithm. It would be nice if the nice insights in the paper could be turned into a more theoretically clean framework that could be re-submitted to a future conference."""
852,"""Ordered Neurons: Integrating Tree Structures into Recurrent Neural Networks""","['Deep Learning', 'Natural Language Processing', 'Recurrent Neural Networks', 'Language Modeling']","""Natural language is hierarchically structured: smaller units (e.g., phrases) are nested within larger units (e.g., clauses). When a larger constituent ends, all of the smaller constituents that are nested within it must also be closed. While the standard LSTM architecture allows different neurons to track information at different time scales, it does not have an explicit bias towards modeling a hierarchy of constituents. This paper proposes to add such inductive bias by ordering the neurons; a vector of master input and forget gates ensures that when a given neuron is updated, all the neurons that follow it in the ordering are also updated. Our novel recurrent architecture, ordered neurons LSTM (ON-LSTM), achieves good performance on four different tasks: language modeling, unsupervised parsing, targeted syntactic evaluation, and logical inference.""","""This paper presents a substantially new way of introducing a syntax-oriented inductive bias into sentence-level models for NLP without explicitly injecting linguistic knowledge. This is a major topic of research in representation learning for NLP, so to see something genuinely original work well is significant. All three reviewers were impressed by the breadth of the experiments and by the results, and this will clearly be among the more ambitious papers presented at this conference. In preparing a final version of this paper, though, I'd urge the authors to put serious further effort into the writing and presentation. All three reviewers had concerns about confusing or misleading passages, including the title and the discussion of the performance of tree-structured models so far."""
853,"""Neural MMO: A massively multiplayer game environment for intelligent agents""","['MMO', 'Multiagent', 'Game', 'Reinforcement Learning', 'Platform', 'Framework', 'Niche Formation', 'Exploration']","""We present an artificial intelligence research platform inspired by the human game genre of MMORPGs (Massively Multiplayer Online Role-Playing Games, a.k.a. MMOs). We demonstrate how this platform can be used to study behavior and learning in large populations of neural agents. Unlike currently popular game environments, our platform supports persistent environments, with variable number of agents, and open-ended task descriptions. The emergence of complex life on Earth is often attributed to the arms race that ensued from a huge number of organisms all competing for finite resources. Our platform aims to simulate this setting in microcosm: we conduct a series of experiments to test how large-scale multiagent competition can incentivize the development of skillful behavior. We find that population size magnifies the complexity of the behaviors that emerge and results in agents that out-compete agents trained in smaller populations.""","""The reviewers raise a number of concerns including limited methodological novelty, limited experimental evaluation (comparisons), and poor readability. Although the authors did address some of the concerns, the paper as is needs a lot of polishing and rewriting. Hence, I cannot recommend this work for presentation at ICLR."""
854,"""Deterministic Variational Inference for Robust Bayesian Neural Networks""","['Bayesian neural network', 'variational inference', 'variational bayes', 'variance reduction', 'empirical bayes']","""Bayesian neural networks (BNNs) hold great promise as a flexible and principled solution to deal with uncertainty when learning from finite data. Among approaches to realize probabilistic inference in deep neural networks, variational Bayes (VB) is theoretically grounded, generally applicable, and computationally efficient. With wide recognition of potential advantages, why is it that variational Bayes has seen very limited practical use for BNNs in real applications? We argue that variational inference in neural networks is fragile: successful implementations require careful initialization and tuning of prior variances, as well as controlling the variance of Monte Carlo gradient estimates. We provide two innovations that aim to turn VB into a robust inference tool for Bayesian neural networks: first, we introduce a novel deterministic method to approximate moments in neural networks, eliminating gradient variance; second, we introduce a hierarchical prior for parameters and a novel Empirical Bayes procedure for automatically selecting prior variances. Combining these two innovations, the resulting method is highly efficient and robust. On the application of heteroscedastic regression we demonstrate good predictive performance over alternative approaches.""","""The manuscript proposes deterministic approximations for Bayesian neural networks as an alternative to the standard Monte-Carlo approach. The results suggest that the deterministic approximation can be more accurate than previous methods. Some explicit contributions include efficient moment estimates and empirical Bayes procedures. The reviewers and ACs note weakness in the breadth and complexity of models evaluated, particularly with regards to ablation studies. This issue seems to have been addressed to the reviewer's satisfaction by the rebuttal. The updated manuscript also improves references to related prior work. Overall, reviewers and AC agree that the general problem statement is timely and interesting, and well executed. We recommend acceptance."""
855,"""A Differentiable Self-disambiguated Sense Embedding Model via Scaled Gumbel Softmax""","['unsupervised representation learning', 'sense embedding', 'word sense disambiguation', 'human evaluation']","""We present a differentiable multi-prototype word representation model that disentangles senses of polysemous words and produces meaningful sense-specific embeddings without external resources. It jointly learns how to disambiguate senses given local context and how to represent senses using hard attention. Unlike previous multi-prototype models, our model approximates discrete sense selection in a differentiable manner via a modified Gumbel softmax. We also propose a novel human evaluation task that quantitatively measures (1) how meaningful the learned sense groups are to humans and (2) how well the model is able to disambiguate senses given a context sentence. Our model outperforms competing approaches on both human evaluations and multiple word similarity tasks.""",""" Pros: * High quality evaluation across different benchmarks, plus human eval * The paper is well written (though one could quibble about the motivation for the method, see Cons) Cons: * The approach is incremental, the main contribution is replacing marginalization or RL with G-S. G-S has already been studied in the context of VAEs with categorical latent variables, i.e. very similar models. * The main technical novelty is varying amount of added noise (i.e. downscaling Gumbel noise). In principle, the Gumbel relaxation is not needed here as exact marginalization can be done (as) effectively. Unlike the standard strategy used to make discrete r.v. tractable in complex models, samples from G-S are not used in this work to weight input to the 'decoder' (thus avoiding expensive marginalization) but to weight terms corresponding to reconstruction from individual latent states (in constract, e.g., to SkimRNN of Seo et al (ICLR 2018)). Presumably adding noise to softmax helps to force sharpness on the posteriors (~ argmax in previous work) and stochasticity may also help exploration. (Given the above, ""to preserve differentiability and circumvent the difficulties in training with reinforcement learning, we apply the reparameterization trick with Gumbel softmax"" seems slightly misleading) * With contextualized embeddings, which are sense-disambiguated given the context, learning discrete senses (which are anyway only coarse approximations of reality) is less practically important Two reviewers are somewhat lukewarm (weak accept) about the paper (limited novelty), whereas one reviewer is considerably more positive. I do not believe that the reviews diverge in any factual information though. """
856,"""Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids""","['Dynamics modeling', 'Control', 'Particle-Based Representation']","""Real-life control tasks involve matters of various substances---rigid or soft bodies, liquid, gas---each with distinct physical behaviors. This poses challenges to traditional rigid-body physics engines. Particle-based simulators have been developed to model the dynamics of these complex scenes; however, relying on approximation techniques, their simulation often deviates from real-world physics, especially in the long term. In this paper, we propose to learn a particle-based simulator for complex control tasks. Combining learning with particle-based systems brings in two major benefits: first, the learned simulator, just like other particle-based systems, acts widely on objects of different materials; second, the particle-based representation poses strong inductive bias for learning: particles of the same type have the same dynamics within. This enables the model to quickly adapt to new environments of unknown dynamics within a few observations. We demonstrate robots achieving complex manipulation tasks using the learned simulator, such as manipulating fluids and deformable foam, with experiments both in simulation and in the real world. Our study helps lay the foundation for robot learning of dynamic scenes with particle-based representations.""","""The paper proposes a particle based framework for learning object dynamics. A scene is represented by a hierarchical graph over particles, edges between particles are established dynamically based on Euclidean distance. The model is used for model predictive control, and there is also one experiment with a particle graph built from a real scene as opposed to simulation. All reviewers agree that the architectural changes over previous relational networks are worthwhile and merit publication. They also suggest to tone down the ``dynamic part of the graph construction by stating that edges are determined based on a radius. In particular, previous works also consider similar addition of edges during collisions, quoting Mrowca et al. ""Collisions between objects are handled by dynamically defining pairwise collision relations ... between leaf particles..."" which suggests that comparison against a baseline for Mrowca et al. that uses a static graph is not entirely fair. The authors are encouraged to repeat the experiment without disabling such dynamic addition of edges. """
857,"""Learning Internal Dense But External Sparse Structures of Deep Neural Network""","['Convolutional Neural Network', 'Hierarchical Neural Architecture', 'Structural Sparsity', 'Evolving Algorithm']","""Recent years have witnessed two seemingly opposite developments of deep convolutional neural networks (CNNs). On one hand, increasing the density of CNNs by adding cross-layer connections achieve higher accuracy. On the other hand, creating sparsity structures through regularization and pruning methods enjoys lower computational costs. In this paper, we bridge these two by proposing a new network structure with locally dense yet externally sparse connections. This new structure uses dense modules, as basic building blocks and then sparsely connects these modules via a novel algorithm during the training process. Experimental results demonstrate that the locally dense yet externally sparse structure could acquire competitive performance on benchmark tasks (CIFAR10, CIFAR100, and ImageNet) while keeping the network structure slim.""","""This paper proposes a genetic algorithm to search neural network architectures with locally dense and globally sparse connections. A population-based genetic algorithm is used to find the sparse, connections between dense module units. The local dense but global sparse architecture is an interesting idea, yet is not well studied in the current version, e.g. overfitting and connections with other similar architecture search methods. Based on reviewers ratings (5,5,6), the current version of paper is proposed as borderline lean reject. """
858,"""Adversarial Imitation via Variational Inverse Reinforcement Learning""","['Inverse Reinforcement Learning', 'Imitation learning', 'Variational lnference', 'Learning from demonstrations']","""We consider a problem of learning the reward and policy from expert examples under unknown dynamics. Our proposed method builds on the framework of generative adversarial networks and introduces the empowerment-regularized maximum-entropy inverse reinforcement learning to learn near-optimal rewards and policies. Empowerment-based regularization prevents the policy from overfitting to expert demonstrations, which advantageously leads to more generalized behaviors that result in learning near-optimal rewards. Our method simultaneously learns empowerment through variational information maximization along with the reward and policy under the adversarial learning formulation. We evaluate our approach on various high-dimensional complex control tasks. We also test our learned rewards in challenging transfer learning problems where training and testing environments are made to be different from each other in terms of dynamics or structure. The results show that our proposed method not only learns near-optimal rewards and policies that are matching expert behavior but also performs significantly better than state-of-the-art inverse reinforcement learning algorithms.""","""This paper proposes a regularization for IRL based on empowerment. The paper has some good results, and is generally well-written. The reviewers raised concerns about how the approach was motivated; these concerns have largely been addressed from the reframing of the algorithm from the perspective of regularization. Now, all reviewers agree that the paper is somewhat above the bar for acceptance. Hence, I also recommend accept. There are several changes that the authors are strongly encouraged to incorporate in the final version of the paper (based on discussion between the reviewers): - The claim that empowerment acts as a regularizer in the policy update is a fairly complicated interpretation of the effect of the algorithm. It relies on an approximation derived in the appendix that relates the proposed objective with an empowerment regularized IRL formulation. The new framing makes much more sense. However, the one sentence reference to this section of the appendix in the main paper is not appropriate given that it is central to the claims of the paper's contribution. More discussion in the main text should be included. - There are still some parts of the implemented algorithm that could introduce bias (using a target network in the shaping term which differs from the theory in Ng et al. 1999), but this concern could be remedied by a code release. The authors are strongly encouraged to link to the code in the final non-blind submission, especially since IRL implementations tend to be quite difficult to get right. - The authors said they would change the way they bold their best numbers in their rebuttal. The current paper does not make the promised change, and actually adopts different bolding conventions in different tables which is even more confusing. The numbers should be bolded in a consistent way, bolding the numbers with the best performance up to statistical significance."""
859,"""A Data-Driven and Distributed Approach to Sparse Signal Representation and Recovery""","['Sparsity', 'Compressive Sensing', 'Convolutional Network']","""In this paper, we focus on two challenges which offset the promise of sparse signal representation, sensing, and recovery. First, real-world signals can seldom be described as perfectly sparse vectors in a known basis, and traditionally used random measurement schemes are seldom optimal for sensing them. Second, existing signal recovery algorithms are usually not fast enough to make them applicable to real-time problems. In this paper, we address these two challenges by presenting a novel framework based on deep learning. For the first challenge, we cast the problem of finding informative measurements by using a maximum likelihood (ML) formulation and show how we can build a data-driven dimensionality reduction protocol for sensing signals using convolutional architectures. For the second challenge, we discuss and analyze a novel parallelization scheme and show it significantly speeds-up the signal recovery process. We demonstrate the significant improvement our method obtains over competing methods through a series of experiments. ""","""This paper studies deep convolutional architectures to perform compressive sensing of natural images, demonstrating improved empirical performance with an efficient pipeline. Reviewers reached a consensus that this is an interesting contribution that advances data-driven methods for compressed sensing, despite some doubts about the experimental setup and the scope of the theoretical insights. We thus recommend acceptance as poster. """
860,"""Connecting the Dots Between MLE and RL for Sequence Generation""","['sequence generation', 'maximum likelihood learning', 'reinforcement learning', 'policy optimization', 'text generation', 'reward augmented maximum likelihood', 'exposure bias']","""Sequence generation models such as recurrent networks can be trained with a diverse set of learning algorithms. For example, maximum likelihood learning is simple and efficient, yet suffers from the exposure bias problem. Reinforcement learning like policy gradient addresses the problem but can have prohibitively poor exploration efficiency. A variety of other algorithms such as RAML, SPG, and data noising, have also been developed in different perspectives. This paper establishes a formal connection between these algorithms. We present a generalized entropy regularized policy optimization formulation, and show that the apparently divergent algorithms can all be reformulated as special instances of the framework, with the only difference being the configurations of reward function and a couple of hyperparameters. The unified interpretation offers a systematic view of the varying properties of exploration and learning efficiency. Besides, based on the framework, we present a new algorithm that dynamically interpolates among the existing algorithms for improved learning. Experiments on machine translation and text summarization demonstrate the superiority of the proposed algorithm.""","""I enjoyed reading the paper myself and I appreciate the unifying framework connecting RAML and SPG. While I do not put a lot of weight on the experiments, I agree with the reviewers that the experimental results are not very strong, and I am not convinced that the theoretical contribution meets the bar at ICLR. In the interpolation algorithm, there seems to be an additional annealing parameter and two tuning parameters. It is important to describe how the parameters are tuned. Given the additional hyper-parameters, one may consider giving all of the algorithms the same budget of hyper-parameter tuning. I also agree with reviewers that the policy gradient baseline seems to underperform typical results. One possible way to strengthen the experiments is to try to replicate the results of SPG or RAML and discuss the behavior of each algorithm as a function of hyper-parameters. """
861,"""A Closer Look at Deep Learning Heuristics: Learning rate restarts, Warmup and Distillation""","['deep learning heuristics', 'learning rate restarts', 'learning rate warmup', 'knowledge distillation', 'mode connectivity', 'SVCCA']","""The convergence rate and final performance of common deep learning models have significantly benefited from recently proposed heuristics such as learning rate schedules, knowledge distillation, skip connections and normalization layers. In the absence of theoretical underpinnings, controlled experiments aimed at explaining the efficacy of these strategies can aid our understanding of deep learning landscapes and the training dynamics. Existing approaches for empirical analysis rely on tools of linear interpolation and visualizations with dimensionality reduction, each with their limitations. Instead, we revisit the empirical analysis of heuristics through the lens of recently proposed methods for loss surface and representation analysis, viz. mode connectivity and canonical correlation analysis (CCA), and hypothesize reasons why the heuristics succeed. In particular, we explore knowledge distillation and learning rate heuristics of (cosine) restarts and warmup using mode connectivity and CCA. Our empirical analysis suggests that: (a) the reasons often quoted for the success of cosine annealing are not evidenced in practice; (b) that the effect of learning rate warmup is to prevent the deeper layers from creating training instability; and (c) that the latent knowledge shared by the teacher is primarily disbursed in the deeper layers.""","""The presented method uses mode connectivity to help illustrate the surfaces of parameter space between various selections of models (either through changes of parameters, learning methods, or epochs), and canonical correlation analysis (CCA) to visualize the similarity of model layers across two different selected models. These analyses are then used to study 3 forms of learning heuristics: stochastic gradient descent with restart (SGDR), warmup, and distillation. Reviews tend to be leaning toward acceptance. Pros: + R1: Well-written + R1: Papers that analyze learning strategies are generally informative to the larger community. These experiments haven't been previously performed. + R1: Thorough experiments + R3: Results brought into context of prior hypotheses Cons: - R3: Batch normalization not studied, but authors have added experiments in response. - R3 & R2: Practical implications not clear, but authors have added a discussion. """
862,"""GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding""","['natural language understanding', 'multi-task learning', 'evaluation']","""For natural language understanding (NLU) technology to be maximally useful, it must be able to process language in a way that is not exclusive to a single task, genre, or dataset. In pursuit of this objective, we introduce the General Language Understanding Evaluation (GLUE) benchmark, a collection of tools for evaluating the performance of models across a diverse set of existing NLU tasks. By including tasks with limited training data, GLUE is designed to favor and encourage models that share general linguistic knowledge across tasks. GLUE also includes a hand-crafted diagnostic test suite that enables detailed linguistic analysis of models. We evaluate baselines based on current methods for transfer and representation learning and find that multi-task training on all tasks performs better than training a separate model per task. However, the low absolute performance of our best model indicates the need for improved general NLU systems.""","""This paper provides an interesting benchmark for multitask learning in NLP. I wish the dataset included language generation tasks instead of just classification but it's still a step in the right direction. """
863,"""HyperGAN: Exploring the Manifold of Neural Networks""","['hypernetworks', 'generative adversarial networks', 'anomaly detection']","""We introduce HyperGAN, a generative network that learns to generate all the weight parameters of deep neural networks. HyperGAN first transforms low dimensional noise into a latent space, which can be sampled from to obtain diverse, performant sets of parameters for a target architecture. We utilize an architecture that bears resemblance to generative adversarial networks, but we evaluate the likelihood of samples with a classification loss. This is equivalent to minimizing the KL-divergence between the generated network parameter distribution and an unknown true parameter distribution. We apply HyperGAN to classification, showing that HyperGAN can learn to generate parameters which solve the MNIST and CIFAR-10 datasets with competitive performance to fully supervised learning while learning a rich distribution of effective parameters. We also show that HyperGAN can also provide better uncertainty than standard ensembles. This is evaluated by the ability of HyperGAN-generated ensembles to detect out of distribution data as well as adversarial examples. We see that in addition to being highly accurate on inlier data, HyperGAN can provide reasonable uncertainty estimates. ""","""All of the reviewers find this paper to contain interesting ideas. Originally, clarity was a major issue, although a few issues remain (see the comments of reviewer 3). The reviewers believe that the paper has been substantially improved from its original form, however there is still room for improvement: more comprehensive comparisons to existing work (reviewer 1), careful ablations (reviewer 3), etc. With a little bit of polish, this paper is likely to be accepted at another venue. I am certainly not penalizing you for anonymously sharing your code on Github, as this was specifically requested by reviewer 1. """
864,"""Improving Differentiable Neural Computers Through Memory Masking, De-allocation, and Link Distribution Sharpness Control""","['rnn', 'dnc', 'memory augmented neural networks', 'mann']","""The Differentiable Neural Computer (DNC) can learn algorithmic and question answering tasks. An analysis of its internal activation patterns reveals three problems: Most importantly, the lack of key-value separation makes the address distribution resulting from content-based look-up noisy and flat, since the value influences the score calculation, although only the key should. Second, DNC's de-allocation of memory results in aliasing, which is a problem for content-based look-up. Thirdly, chaining memory reads with the temporal linkage matrix exponentially degrades the quality of the address distribution. Our proposed fixes of these problems yield improved performance on arithmetic tasks, and also improve the mean error rate on the bAbI question answering dataset by 43%.""",""" pros: - Identification of several interesting problems with the original DNC model: masked attention, erasion of de-allocated elements, and sharpened temporal links - An improved architecture which addresses the issues and shows improved performance on synthetic memory tasks and bAbI over the original model - Clear writing cons: - Does not really show this modified DNC can solve a task that the original DNC could not and the bAbI tasks are effectively solved anyway. It is still not clear whether the DNC even with these improvements will have much impact beyond these toy tasks. Overall the reviewers found this to be a solid paper with a useful analysis and I agree. I recommend acceptance. """
865,"""Surprising Negative Results for Generative Adversarial Tree Search ""","['Deep Reinforcement Learning', 'Generative Adversarial Nets']","""While many recent advances in deep reinforcement learning rely on model-free methods, model-based approaches remain an alluring prospect for their potential to exploit unsupervised data to learn environment dynamics. One prospect is to pursue hybrid approaches, as in AlphaGo, which combines Monte-Carlo Tree Search (MCTS)a model-based methodwith deep-Q networks (DQNs)a model-free method. MCTS requires generating rollouts, which is computationally expensive. In this paper, we propose to simulate roll-outs, exploiting the latest breakthroughs in image-to-image transduction, namely Pix2Pix GANs, to predict the dynamics of the environment. Our proposed algorithm, generative adversarial tree search (GATS), simulates rollouts up to a specified depth using both a GAN- based dynamics model and a reward predictor. GATS employs MCTS for planning over the simulated samples and uses DQN to estimate the Q-function at the leaf states. Our theoretical analysis establishes some favorable properties of GATS vis-a-vis the bias-variance trade-off and empirical results show that on 5 popular Atari games, the dynamics and reward predictors converge quickly to accurate solutions. However, GATS fails to outperform DQNs in 4 out of 5 games. Notably, in these experiments, MCTS has only short rollouts (up to tree depth 4), while previous successes of MCTS have involved tree depth in the hundreds. We present a hypothesis for why tree search with short rollouts can fail even given perfect modeling.""","""The paper addresses questions on the relationship between model-free and model-based reinforcement learning, in particular focusing on planning using learned generative models. The proposed approach, GATS, uses learned generative models for rollouts in MCTS, and provide theoretical insights that show a favorable bias-variance tradeoff. Despite this theoretical advantage, and high-quality models, the proposed approach fails to perform well empirically. This surprising negative results motivates the paper and providing insights on it is the main contribution. Based on the initial submitted version, the reviewers positively emphasized the need to understand and publish important negative results. All reviewers and the AC appreciate the import role that such a contribution can bring to the research community. Reviewers also note the careful discussion of modeling choices for the generative models. The reviewers also noted several potential weaknesses. Central were the need to better motivate and investigate the hypothesis proposed to explain the negative results. Several avenues towards a better understanding were proposed, and many of these were picked up by the authors in the revision and rebuttal. A novel toy domain ""goldfish and gold bucket"" was introduced for empirical analysis, and experiments there show that GATS can outperform DQN when a longer planning horizon is used. The introduced toy domain provides additional insights into the relationship between planning horizon and GATS / MCTS performance. However, it does not address key questions around why the negative result is maintained. The authors hypothesize that the Q-value is less accurate in the GATS setting - this is something that can be empirically evaluated, but specific evidence for this hypothesis is not clearly shown. Other forms of analysis that could shed further light on why the specific negative result occurs could be to inspect model errors. For example, if generated frames are sorted by the magnitude of prediction errors - what are the largest mistakes? Could these cause learning performance to deteriorate? The reviewers also raised several issues around the theoretical analysis, clarity (especially of captions) and structure - these were largely addressed by the revision. The concern that most strongly affected the final evaluation is the limited insight (and evidence) of the factors that influence performance of the proposed approach. Due to this, the consensus is to not accept the paper for publication at ICLR at this stage."""
866,"""Adversarial Audio Synthesis""","['audio', 'waveform', 'spectrogram', 'GAN', 'adversarial', 'WaveGAN', 'SpecGAN']","""Audio signals are sampled at high temporal resolutions, and learning to synthesize audio requires capturing structure across a range of timescales. Generative adversarial networks (GANs) have seen wide success at generating images that are both locally and globally coherent, but they have seen little application to audio generation. In this paper we introduce WaveGAN, a first attempt at applying GANs to unsupervised synthesis of raw-waveform audio. WaveGAN is capable of synthesizing one second slices of audio waveforms with global coherence, suitable for sound effect generation. Our experiments demonstrate thatwithout labelsWaveGAN learns to produce intelligible words when trained on a small-vocabulary speech dataset, and can also synthesize audio from other domains such as drums, bird vocalizations, and piano. We compare WaveGAN to a method which applies GANs designed for image generation on image-like audio feature representations, finding both approaches to be promising.""","""This paper proposes a GAN model to synthesize raw-waveform audio by adapting the popular DC-GAN architecture to handle audio signals. Experimental results are reported on several datasets, including speech and instruments. Unfortunately this paper received two low-quality reviews, with little signal. The only substantial review was mildly positive, highlighting the clarity, accessibility and reproducibility of the work, and expressing concerns about the relative lack of novelty. The AC shares this assessment. The paper claims to be the first successful GAN application operating directly on wave-forms. Whereas this is certainly an important contribution, it is less clear to the AC whether this contribution belongs to a venue such as ICLR, as opposed to ICASSP or Ismir. This is a borderline paper, and the decision is ultimately relative to other submissions with similar scores. In this context, given the mainstream popularity of GANs for image modeling, the AC feels this paper can help spark significant further research in adversarial training for audio modeling, and therefore recommends acceptance. I also encourage the authors to address the issues raised by R1. """
867,"""Neural Network Cost Landscapes as Quantum States""","['quantum', 'neural networks', 'meta-learning']","""Quantum computers promise significant advantages over classical computers for a number of different applications. We show that the complete loss function landscape of a neural network can be represented as the quantum state output by a quantum computer. We demonstrate this explicitly for a binary neural network and, further, show how a quantum computer can train the network by manipulating this state using a well-known algorithm known as quantum amplitude amplification. We further show that with minor adaptation, this method can also represent the meta-loss landscape of a number of neural network architectures simultaneously. We search this meta-loss landscape with the same method to simultaneously train and design a binary neural network. ""","""This paper studies the problem of training binary neural networks using quantum amplitude amplification method. Reviewers agree that the problem considered is novel and interesting. However the consensus is that there are only few experiments in the current paper and the paper needs more experiments on different datasets with comparisons to proper baselines. Reviewers opined that the paper was not so easy to follow initially, though later revisions may have somewhat alleviated this problem."""
868,"""Provable Defenses against Spatially Transformed Adversarial Inputs: Impossibility and Possibility Results""",[],"""One intriguing property of neural networks is their inherent vulnerability to adversarial inputs, which are maliciously crafted samples to trigger target networks to misbehave. The state-of-the-art attacks generate adversarial inputs using either pixel perturbation or spatial transformation. Thus far, several provable defenses have been proposed against pixel perturbation-based attacks; yet, little is known about whether such solutions exist for spatial transformation-based attacks. This paper bridges this striking gap by conducting the first systematic study on provable defenses against spatially transformed adversarial inputs. Our findings convey mixed messages. On the impossibility side, we show that such defenses may not exist in practice: for any given networks, it is possible to find legitimate inputs and imperceptible transformations to generate adversarial inputs that force arbitrarily large errors. On the possibility side, we show that it is still feasible to construct adversarial training methods to significantly improve the resilience of networks against adversarial inputs over empirical datasets. We believe our findings provide insights for designing more effective defenses against spatially transformed adversarial inputs.""","""This paper conducts a study on provable defenses to spatially transformed adversarial examples. In general, the paper pursues an interesting direction, but reviewers had many concerns regarding the clarity of the presentation and the depth of the experimental results, which the authors did not address in a rebuttal. """
869,"""Learning Recurrent Binary/Ternary Weights""","['Quantized Recurrent Neural Network', 'Hardware Implementation', 'Deep Learning']","""Recurrent neural networks (RNNs) have shown excellent performance in processing sequence data. However, they are both complex and memory intensive due to their recursive nature. These limitations make RNNs difficult to embed on mobile devices requiring real-time processes with limited hardware resources. To address the above issues, we introduce a method that can learn binary and ternary weights during the training phase to facilitate hardware implementations of RNNs. As a result, using this approach replaces all multiply-accumulate operations by simple accumulations, bringing significant benefits to custom hardware in terms of silicon area and power consumption. On the software side, we evaluate the performance (in terms of accuracy) of our method using long short-term memories (LSTMs) and gated recurrent units (GRUs) on various sequential models including sequence classification and language modeling. We demonstrate that our method achieves competitive results on the aforementioned tasks while using binary/ternary weights during the runtime. On the hardware side, we present custom hardware for accelerating the recurrent computations of LSTMs with binary/ternary weights. Ultimately, we show that LSTMs with binary/ternary weights can achieve up to 12x memory saving and 10x inference speedup compared to the full-precision hardware implementation design.""","""This work proposes a simple but useful way to train RNN with binary / ternary weights for improving memory and power efficiency. The paper presented a sequence of experiments on various benchmarks and demonstrated significant improvement on memory size with only minor decrease of accuracy. Authors' rebuttal addressed the reviewers' concern nicely. """
870,"""Graph Transformer ""","['Graph neural networks', 'transformer', 'attention']","""Graph neural networks (GNN) have gained increasing research interests as a mean to the challenging goal of robust and universal graph learning. Previous GNNs have assumed single pre-fixed graph structure and permitted only local context encoding. This paper proposes a novel Graph Transformer (GTR) architecture that captures long-range dependency with global attention, and enables dynamic graph structures. In particular, GTR propagates features within the same graph structure via an intra-graph message passing, and transforms dynamic semantics across multi-domain graph-structured data (e.g. images, sequences, knowledge graphs) for multi-modal learning via an inter-graph message passing. Furthermore, GTR enables effective incorporation of any prior graph structure by weighted averaging of the prior and learned edges, which can be crucially useful for scenarios where prior knowledge is desired. The proposed GTR achieves new state-of-the-arts across three benchmark tasks, including few-shot learning, medical abnormality and disease classification, and graph classification. Experiments show that GTR is superior in learning robust graph representations, transforming high-level semantics across domains, and bridging between prior graph structure with automatic structure learning. ""","""The reviewers all agree that the work is interesting, but none have stood out and championed the paper as exceptional. The reviewers note that the paper is well-written, contributes a methodological innovation, and provides compelling experiments. However, given the reviewers' positive but unenthusiastic scores, and after discussion with PCs, this paper does not meet the bar for acceptance into ICLR."""
871,"""Discriminator-Actor-Critic: Addressing Sample Inefficiency and Reward Bias in Adversarial Imitation Learning""","['deep learning', 'reinforcement learning', 'imitation learning', 'adversarial learning']","""We identify two issues with the family of algorithms based on the Adversarial Imitation Learning framework. The first problem is implicit bias present in the reward functions used in these algorithms. While these biases might work well for some environments, they can also lead to sub-optimal behavior in others. Secondly, even though these algorithms can learn from few expert demonstrations, they require a prohibitively large number of interactions with the environment in order to imitate the expert for many real-world applications. In order to address these issues, we propose a new algorithm called Discriminator-Actor-Critic that uses off-policy Reinforcement Learning to reduce policy-environment interaction sample complexity by an average factor of 10. Furthermore, since our reward function is designed to be unbiased, we can apply our algorithm to many problems without making any task-specific adjustments. ""","""This work highlights the problem of biased rewards present in common adversarial imitation learning implementations, and proposes adding absorbing states to to fix the issue. This is combined with an off-policy training algorithm, yielding significantly improved sample efficiency, whose benefits are convincingly shown empirically. The paper is well written and clearly presents the contributions. Questions were satisfactorily answered during discussion, and resulted in an improved submission, a paper that all reviewers now agree is worth presenting at ICLR. """
872,"""A bird's eye view on coherence, and a worm's eye view on cohesion""","['text generation', 'natural language processing', 'neural language model']","""Generating coherent and cohesive long-form texts is a challenging problem in natural language generation. Previous works relied on a large amount of human-generated texts to train neural language models, however, few attempted to explicitly model the desired linguistic properties of natural language text, such as coherence and cohesion using neural networks. In this work, we train two expert discriminators for coherence and cohesion to provide hierarchical feedback for text generation. We also propose a simple variant of policy gradient, called 'negative-critical sequence training' in which the reward 'baseline' is constructed from randomly generated negative samples. We demonstrate the effectiveness of our approach through empirical studies, showing improvements over the strong baseline -- attention-based bidirectional MLE-trained neural language model -- in a number of automated metrics. The proposed model can serve as baseline architectures to promote further research in modeling additional linguistic properties for downstream NLP tasks.""","""This paper attempts at modeling coherence of generated text, and proposes two kinds of discriminators that tries to measure whether a piece of text is coherent or not. However, the paper misses several related critical references, and also lacks extensive evaluation (especially manual evaluation). There is consensus between the reviewers that this paper needs more work before it is accepted to a conference such as ICLR. """
873,"""Unsupervised Video-to-Video Translation""","['Generative Adversarial Networks', 'Computer Vision', 'Deep Learning']","""Unsupervised image-to-image translation is a recently proposed task of translating an image to a different style or domain given only unpaired image examples at training time. In this paper, we formulate a new task of unsupervised video-to-video translation, which poses its own unique challenges. Translating video implies learning not only the appearance of objects and scenes but also realistic motion and transitions between consecutive frames. We investigate the performance of per-frame video-to-video translation using existing image-to-image translation networks, and propose a spatio-temporal 3D translator as an alternative solution to this problem. We evaluate our 3D method on multiple synthetic datasets, such as moving colorized digits, as well as the realistic segmentation-to-video GTA dataset and a new CT-to-MRI volumetric images translation dataset. Our results show that frame-wise translation produces realistic results on a single frame level but underperforms significantly on the scale of the whole video compared to our three-dimensional translation approach, which is better able to learn the complex structure of video and motion and continuity of object appearance. ""","""In this work, a central idea introduced by CycleGAN is extended from 2D convolutions to 3D convolutions to ensure better consistency of style transfer across time. Authors demonstrate improvements on a variety of datasets in comparison to frame-by-frame style transfer. Reviewer Pros: + Seems to be effective at enforcing improved consistency over time + Proposed medical dataset may be good contribution to community. + Good quality evaluation Reviewer Cons: - All reviewers felt the technical novelty was low. - Some questions arose around quantitative results, left unanswered by authors. - Experiments missing some baseline approaches - Architecture limited to fixed length video segments Reviewer consensus is to reject. Authors are encouraged to continue their work and take into account suggestions made by reviewers, including adding additional comparison baselines """
874,"""Towards the first adversarially robust neural network model on MNIST""","['adversarial examples', 'MNIST', 'robustness', 'deep learning', 'security']","""Despite much effort, deep neural networks remain highly susceptible to tiny input perturbations and even for MNIST, one of the most common toy datasets in computer vision, no neural network model exists for which adversarial perturbations are large and make semantic sense to humans. We show that even the widely recognized and by far most successful L-inf defense by Madry et~al. (1) has lower L0 robustness than undefended networks and still highly susceptible to L2 perturbations, (2) classifies unrecognizable images with high certainty, (3) performs not much better than simple input binarization and (4) features adversarial perturbations that make little sense to humans. These results suggest that MNIST is far from being solved in terms of adversarial robustness. We present a novel robust classification model that performs analysis by synthesis using learned class-conditional data distributions. We derive bounds on the robustness and go to great length to empirically evaluate our model using maximally effective adversarial attacks by (a) applying decision-based, score-based, gradient-based and transfer-based attacks for several different Lp norms, (b) by designing a new attack that exploits the structure of our defended model and (c) by devising a novel decision-based attack that seeks to minimize the number of perturbed pixels (L0). The results suggest that our approach yields state-of-the-art robustness on MNIST against L0, L2 and L-inf perturbations and we demonstrate that most adversarial examples are strongly perturbed towards the perceptual boundary between the original and the adversarial class.""","""The paper presents a technique of training robust classification models that uses the input distribution within each class to achieve high accuracy and robustness against adversarial perturbations. Strengths: - The resulting model offers good robustness guarantees for a wide range of norm-bounded perturbations - The authors put a lot of care into the robustness evaluation Weaknesses: - Some of the ""shortcomings"" attributed to the previous work seem confusing, as the reported vulnerability corresponds to threat models that the previous work did not made claims about Overall, this looks like a valuable and interesting contribution. """
875,"""Adversarial Attacks on Graph Neural Networks via Meta Learning""","['graph mining', 'adversarial attacks', 'meta learning', 'graph neural networks', 'node classification']","""Deep learning models for graphs have advanced the state of the art on many tasks. Despite their recent success, little is known about their robustness. We investigate training time attacks on graph neural networks for node classification that perturb the discrete graph structure. Our core principle is to use meta-gradients to solve the bilevel problem underlying training-time attacks, essentially treating the graph as a hyperparameter to optimize. Our experiments show that small graph perturbations consistently lead to a strong decrease in performance for graph convolutional networks, and even transfer to unsupervised embeddings. Remarkably, the perturbations created by our algorithm can misguide the graph neural networks such that they perform worse than a simple baseline that ignores all relational information. Our attacks do not assume any knowledge about or access to the target classifiers.""",""" The paper proposes an method for investigating robustness of graph neural nets for node classification problem; training-time attacks for perturbing graph structure are generated using meta-learning approach. Reviewers agree that the contribution is novel and empirical results support the validity of the approach. """
876,"""Deep Convolutional Networks as shallow Gaussian Processes""","['Gaussian process', 'CNN', 'ResNet', 'Bayesian']","""We show that the output of a (residual) CNN with an appropriate prior over the weights and biases is a GP in the limit of infinitely many convolutional filters, extending similar results for dense networks. For a CNN, the equivalent kernel can be computed exactly and, unlike ""deep kernels"", has very few parameters: only the hyperparameters of the original CNN. Further, we show that this kernel has two properties that allow it to be computed efficiently; the cost of evaluating the kernel for a pair of images is similar to a single forward pass through the original CNN with only one filter per layer. The kernel equivalent to a 32-layer ResNet obtains 0.84% classification error on MNIST, a new record for GP with a comparable number of parameters.""","""This paper builds on a promising line of literature developing connections between Gaussian processes and deep neural networks. Viewing one model under the lens of (the infinite limit of) another can lead to neat new insights and algorithms. In this case the authors develop a connection between convolutional networks and Gaussian processes with a particular kind of kernel. The reviews were quite mixed with one champion and two just below borderline. The reviewers all believed the paper had contributions which would be interesting to the community (such as R1: ""the paper presents a novel efficient way to compute the convolutional kernel, which I believe has merits on its own"" and R2: ""I really like the idea of authors that kernels based on convolutional networks might be more practical compared to the ones based on fully connected networks""). All the reviewers found the contribution of the covariance function to be novel and exciting. Some cited weaknesses of the paper were that the authors didn't analyze the uncertainty from the model (arguably the reasoning for adopting a Bayesian treatment), novelty in appealing to the central limit theorem to arrive at the connection, and scalability of the model. In the review process it also became apparent that there was another paper with a substantially similar contribution. The decision for this paper was calibrated accordingly with that work. Weighing the strengths and weaknesses of the paper and taking into account a reviewer willing to champion the work it seems there is enough novel contribution and interest in the work to justify acceptance. The authors provided responses to the reviewer concerns including calibration plots and timing experiments in the discussion period and it would be appreciated if these can be incorporated into the camera ready version."""
877,"""LSH Microbatches for Stochastic Gradients: Value in Rearrangement""","['Stochastic Gradient Descent', 'Metric Embeddings', 'Locality  Sensitive Hashing', 'Microbatches', 'Sample coordination']",""" Metric embeddings are immensely useful representations of associations between entities (images, users, search queries, words, and more). Embeddings are learned by optimizing a loss objective of the general form of a sum over example associations. Typically, the optimization uses stochastic gradient updates over minibatches of examples that are arranged independently at random. In this work, we propose the use of {\em structured arrangements} through randomized {\em microbatches} of examples that are more likely to include similar ones. We make a principled argument for the properties of our arrangements that accelerate the training and present efficient algorithms to generate microbatches that respect the marginal distribution of training examples. Finally, we observe experimentally that our structured arrangements accelerate training by 3-20\%. Structured arrangements emerge as a powerful and novel performance knob for SGD that is independent and complementary to other SGD hyperparameters and thus is a candidate for wide deployment.""","""Following the unanimous vote of the four submitted reviews, this paper is not ready for publication at ICLR. Among other concerns raised, the experiments need significant work."""
878,"""PPO-CMA: Proximal Policy Optimization with Covariance Matrix Adaptation""","['Continuous Control', 'Reinforcement Learning', 'Policy Optimization', 'Policy Gradient', 'Evolution Strategies', 'CMA-ES', 'PPO']","""Proximal Policy Optimization (PPO) is a highly popular model-free reinforcement learning (RL) approach. However, in continuous state and actions spaces and a Gaussian policy -- common in computer animation and robotics -- PPO is prone to getting stuck in local optima. In this paper, we observe a tendency of PPO to prematurely shrink the exploration variance, which naturally leads to slow progress. Motivated by this, we borrow ideas from CMA-ES, a black-box optimization method designed for intelligent adaptive Gaussian exploration, to derive PPO-CMA, a novel proximal policy optimization approach that expands the exploration variance on objective function slopes and only shrinks the variance when close to the optimum. This is implemented by using separate neural networks for policy mean and variance and training the mean and variance in separate passes. Our experiments demonstrate a clear improvement over vanilla PPO in many difficult OpenAI Gym MuJoCo tasks.""","""This paper proposes to improve the exploration in the PPO algorithm by applying CMA-ES. Major concerns of the paper include: paper editing can be improved; the choices of baselines used in the paper may be not reasonable; flaws in comparisons with SOTA. It is also not quite clear why CMA can improve exploration, further justification required. Overall, this paper cannot be published in its current form."""
879,"""DELTA: DEEP LEARNING TRANSFER USING FEATURE MAP WITH ATTENTION FOR CONVOLUTIONAL NETWORKS""","['transfer learning', 'deep learning', 'regularization', 'attention', 'cnn']","""Transfer learning through fine-tuning a pre-trained neural network with an extremely large dataset, such as ImageNet, can significantly accelerate training while the accuracy is frequently bottlenecked by the limited dataset size of the new target task. To solve the problem, some regularization methods, constraining the outer layer weights of the target network using the starting point as references (SPAR), have been studied. In this paper, we propose a novel regularized transfer learning framework DELTA, namely DEep Learning Transfer using Feature Map with Attention. Instead of constraining the weights of neural network, DELTA aims to preserve the outer layer outputs of the target network. Specifically, in addition to minimizing the empirical loss, DELTA intends to align the outer layer outputs of two networks, through constraining a subset of feature maps that are precisely selected by attention that has been learned in an supervised learning manner. We evaluate DELTA with the state-of-the-art algorithms, including L2 and L2-SP. The experiment results show that our proposed method outperforms these baselines with higher accuracy for new tasks.""","""This paper argues that each layer of a network may have some channels useful for and some not useful for transfer learning. The main contribution is an approach which identifies the useful channels through an attention based mechanism. The reviewers agree that this work offers a valuable new approach that offers modest improvements over prior work. The authors should take care to refine their definition of behavior regularization, including/expanding on the discussion from the rebuttal phase. The authors are also encouraged to experiment with other architecture backbones and report both overall performance as well as run time for learning with the larger models. """
880,"""Generalized Tensor Models for Recurrent Neural Networks""","['expressive power', 'recurrent neural networks', 'Tensor-Train decomposition']","""Recurrent Neural Networks (RNNs) are very successful at solving challenging problems with sequential data. However, this observed efficiency is not yet entirely explained by theory. It is known that a certain class of multiplicative RNNs enjoys the property of depth efficiency --- a shallow network of exponentially large width is necessary to realize the same score function as computed by such an RNN. Such networks, however, are not very often applied to real life tasks. In this work, we attempt to reduce the gap between theory and practice by extending the theoretical analysis to RNNs which employ various nonlinearities, such as Rectified Linear Unit (ReLU), and show that they also benefit from properties of universality and depth efficiency. Our theoretical results are verified by a series of extensive computational experiments.""","""AR1 finds that extension of the previously presented ICLR'18 paper are interesting and sufficient due to the provided analysis of universality and depth efficiency. AR2 is concerned with the lack of any concrete toy example between the proposed architecture and RNNs. Kindly make an effort to add such a basic step-by-step illustration for a simple chosen architecture e.g. in the supplementary material. AR3 is the most critical (the analysis TT-RNN based on the product non-linearity done before, particular case of rectifier non-linearity is used, etc.) Despite the authors cannot guarantee the existence of corresponding weight tensor W in less trivial cases, the overall analysis is very interesting and it is the starting point for further modeling. Thus, AC advocates acceptance of this paper. The review scores do not indicate this can be an oral paper, e.g. it currently is unlikely to be in top few percent of accepted papers. Nonetheless, this is a valuable and solid work. Moreover, for the camera-ready paper, kindly refresh your list of citations as a mere 1 page of citations feels rather too conservative. This makes the background of the paper and related work obscure to average reader unfamiliar with this topic, tensors, tensor outer products etc. There are numerous works on tensor decompositions that can be acknowledged: - Multilinear Analysis of Image Ensembles: TensorFaces by Vasilescou et al. - Multilinear Projection for Face Recognition via Canonical Decomposition by Vasilescou et al. - Tensor decompositions for learning latent variable models by Anandkumar et al. - Fast and guaranteed tensor decomposition via sketching by Anandkumar et al. One good example of the use of the outer product (sums over rank one outer products of higher-order) is paper from 2013. They perform higher-order pooling on encoded feature vectors (although this seems to be the shallow setting) similar to Eq. 2 and 3 (this submission): - Higher-order occurrence pooling on mid-and low-level features: Visual concept detection by Koniusz et al. (e.g. equations equations 49 and 50 or 1, 16 and 17 realize Eq. 3 and 13 in this submission) - Higher-Order Occurrence Pooling for Bags-of-Words: Visual Concept Detection (similar follow-up work) Other related papers include: - Long-term Forecasting using Tensor-Train RNNs by Anandkumar et al. - Tensor Regression Networks with various Low-Rank Tensor Approximations by Cao et al. Of course, the authors are encouraged to cite even more related works."""
881,"""Sufficient Conditions for Robustness to Adversarial Examples: a Theoretical and Empirical Study with Bayesian Neural Networks""","['Bayesian deep learning', 'Bayesian neural networks', 'adversarial examples']","""We prove, under two sufficient conditions, that idealised models can have no adversarial examples. We discuss which idealised models satisfy our conditions, and show that idealised Bayesian neural networks (BNNs) satisfy these. We continue by studying near-idealised BNNs using HMC inference, demonstrating the theoretical ideas in practice. We experiment with HMC on synthetic data derived from MNIST for which we know the ground-truth image density, showing that near-perfect epistemic uncertainty correlates to density under image manifold, and that adversarial images lie off the manifold in our setting. This suggests why MC dropout, which can be seen as performing approximate inference, has been observed to be an effective defence against adversarial examples in practice; We highlight failure-cases of non-idealised BNNs relying on dropout, suggesting a new attack for dropout models and a new defence as well. Lastly, we demonstrate the defence on a cats-vs-dogs image classification task with a VGG13 variant.""","""This paper conducts a study of the adversarial robustness of Bayesian Neural Network models. The reviewers all agree that the paper presents an interesting direction, with sound theoretical backing. However, there are important concerns regarding the significance and clarity of the work. In particular, the paper would greatly benefit from more demonstrated empirical significance, and more polished definitions and theoretical results. """
882,"""Optimistic mirror descent in saddle-point problems: Going the extra (gradient) mile""","['Mirror descent', 'extra-gradient', 'generative adversarial networks', 'saddle-point problems']","""Owing to their connection with generative adversarial networks (GANs), saddle-point problems have recently attracted considerable interest in machine learning and beyond. By necessity, most theoretical guarantees revolve around convex-concave (or even linear) problems; however, making theoretical inroads towards efficient GAN training depends crucially on moving beyond this classic framework. To make piecemeal progress along these lines, we analyze the behavior of mirror descent (MD) in a class of non-monotone problems whose solutions coincide with those of a naturally associated variational inequality a property which we call coherence. We first show that ordinary, vanilla MD converges under a strict version of this condition, but not otherwise; in particular, it may fail to converge even in bilinear models with a unique solution. We then show that this deficiency is mitigated by optimism: by taking an extra-gradient step, optimistic mirror descent (OMD) converges in all coherent problems. Our analysis generalizes and extends the results of Daskalakis et al. [2018] for optimistic gradient descent (OGD) in bilinear problems, and makes concrete headway for provable convergence beyond convex-concave games. We also provide stochastic analogues of these results, and we validate our analysis by numerical experiments in a wide array of GAN models (including Gaussian mixture models, and the CelebA and CIFAR-10 datasets).""","""This paper investigates the usage of the extragradient step for solving saddle-point problems with non-monotone stochastic variational inequalities, motivated by GANs. The authors propose an assumption weaker/diffrerent than the pseudo-monotonicity of the variational inequality for their convergence analysis (that they call ""coherence""). Interestingly, they are able to show the (asympotic) last iterate convergence for the extragradient algorithm in this case (in contrast to standard results which normally requires averaging of the iterates for the stochastic *and* mototone variational inequality such as the cited work by Gidel et al.). The authors also describe an interesting difference between the gradient method without the extragradient step (mirror descent) vs. with (that they called optimistic mirror descent). R2 thought the coherence condition was too related to the notion of pseudo-monoticity for which one could easily extend previous known convergence results for stochastic variational inequality. The AC thinks that this point was well answered by the authors rebuttal and in their revision: the conditions are sufficiently different, and while there is still much to do to analyze non variational inequalities or having realistic assumptions, this paper makes some non-trivial and interesting steps in this direction. The AC thus sides with expert reviewer R1 and recommends acceptance."""
883,"""Expanding the Reach of Federated Learning by Reducing Client Resource Requirements""",[],"""Communication on heterogeneous edge networks is a fundamental bottleneck in Federated Learning (FL), restricting both model capacity and user participation. To address this issue, we introduce two novel strategies to reduce communication costs: (1) the use of lossy compression on the global model sent server-to-client; and (2) Federated Dropout, which allows users to efficiently train locally on smaller subsets of the global model and also provides a reduction in both client-to-server communication and local computation. We empirically show that these strategies, combined with existing compression approaches for client-to-server communication, collectively provide up to a 9.6x reduction in server-to-client communication, a 1.5x reduction in local computation, and a 24x reduction in upload communication, all without degrading the quality of the final model. We thus comprehensively reduce FL's impact on client device resources, allowing higher capacity models to be trained, and a more diverse set of users to be reached.""","""This paper focuses on communication efficient Federated Learning (FL) and proposes an approach for training large models on heterogeneous edge devices. The paper is well-written and the approach is promising, but all reviewers pointed out that both novelty of the approach and empirical evaluation, including comparison with state-of-art, are somewhat limited. We hope that suggestions provided by the reviewers will be helpful for extending and improving this work."""
884,"""There Are Many Consistent Explanations of Unlabeled Data: Why You Should Average""","['semi-supervised learning', 'computer vision', 'classification', 'consistency regularization', 'flatness', 'weight averaging', 'stochastic weight averaging']","""Presently the most successful approaches to semi-supervised learning are based on consistency regularization, whereby a model is trained to be robust to small perturbations of its inputs and parameters. To understand consistency regularization, we conceptually explore how loss geometry interacts with training procedures. The consistency loss dramatically improves generalization performance over supervised-only training; however, we show that SGD struggles to converge on the consistency loss and continues to make large steps that lead to changes in predictions on the test data. Motivated by these observations, we propose to train consistency-based methods with Stochastic Weight Averaging (SWA), a recent approach which averages weights along the trajectory of SGD with a modified learning rate schedule. We also propose fast-SWA, which further accelerates convergence by averaging multiple points within each cycle of a cyclical learning rate schedule. With weight averaging, we achieve the best known semi-supervised results on CIFAR-10 and CIFAR-100, over many different quantities of labeled training data. For example, we achieve 5.0% error on CIFAR-10 with only 4000 labels, compared to the previous best result in the literature of 6.3%.""","""All reviewers appreciate the empirical analysis and insights provided in the paper. The paper also reports impressive results on SSL. It will be a good addition to the ICLR program. """
885,"""Hierarchical Visuomotor Control of Humanoids""","['hierarchical reinforcement learning', 'motor control', 'motion capture']","""We aim to build complex humanoid agents that integrate perception, motor control, and memory. In this work, we partly factor this problem into low-level motor control from proprioception and high-level coordination of the low-level skills informed by vision. We develop an architecture capable of surprisingly flexible, task-directed motor control of a relatively high-DoF humanoid body by combining pre-training of low-level motor controllers with a high-level, task-focused controller that switches among low-level sub-policies. The resulting system is able to control a physically-simulated humanoid body to solve tasks that require coupling visual perception from an unstabilized egocentric RGB camera during locomotion in the environment. Supplementary video link: pseudo-url""","""A hierarchical method is presented for developing humanoid motion control, using low-level control fragments, egocentric visual input, recurrent high-level control. It is likely the first demonstration of 3D humanoids learning to do memory-enabled tasks using only proprioceptive and head-based ego-centric vision. The use of control fragments as opposed to mocapclip-based skills allows for finer-grained repurposing of pieces of motion, while still allowing for mocap-based learning Weaknesses: It is largely a mashup up of previously known results (R2). Caveat: this can be said for all research at some sufficient level of abstraction. The motions are jerky when transitions happen between control fragments (R2,R3). There are some concerns as to whether the method compares against other methods; the authors note that they are either not directly comparable, i.e., solving a different problem, or are implicitly contained in some of the comparisons that are performed in the paper. Overall, the reviewers and AC are in broad agreement regarding the strengths and weaknesses of the paper. The AC believes that the work will be of broad interest. Demonstrating memory-enabled, vision-driven, mocap-imitating skills is a broad step forward. The paper also provides a further datapoint as to which combinations of method work well, and some of the specific features required to make them work. The paper could acknowledge motion quality artifacts, as noted by the reviewers and in the online discussion. Suggest to include [Peng et al 2017] as some of the most relevant related HRL humanoid control work, as per the reviews & discussion. """
886,"""Discovery of Natural Language Concepts in Individual Units of CNNs""","['interpretability of deep neural networks', 'natural language representation']","""Although deep convolutional networks have achieved improved performance in many natural language tasks, they have been treated as black boxes because they are difficult to interpret. Especially, little is known about how they represent language in their intermediate layers. In an attempt to understand the representations of deep convolutional networks trained on language tasks, we show that individual units are selectively responsive to specific morphemes, words, and phrases, rather than responding to arbitrary and uninterpretable patterns. In order to quantitatively analyze such intriguing phenomenon, we propose a concept alignment method based on how units respond to replicated text. We conduct analyses with different architectures on multiple datasets for classification and translation tasks and provide new insights into how deep models understand natural language.""","""Important problem (making NN more transparent); reasonable approach for identifying which linguistic concepts different neurons are sensitive to; rigorous experiments. Paper was reviewed by three experts. Initially there were some concerns but after the author response and reviewer discussion, all three unanimously recommend acceptance."""
887,"""Spherical CNNs on Unstructured Grids""","['Spherical CNN', 'unstructured grid', 'panoramic', 'semantic segmentation', 'parameter efficiency']","""We present an efficient convolution kernel for Convolutional Neural Networks (CNNs) on unstructured grids using parameterized differential operators while focusing on spherical signals such as panorama images or planetary signals. To this end, we replace conventional convolution kernels with linear combinations of differential operators that are weighted by learnable parameters. Differential operators can be efficiently estimated on unstructured grids using one-ring neighbors, and learnable parameters can be optimized through standard back-propagation. As a result, we obtain extremely efficient neural networks that match or outperform state-of-the-art network architectures in terms of performance but with a significantly lower number of network parameters. We evaluate our algorithm in an extensive series of experiments on a variety of computer vision and climate science tasks, including shape classification, climate pattern segmentation, and omnidirectional image semantic segmentation. Overall, we present (1) a novel CNN approach on unstructured grids using parameterized differential operators for spherical signals, and (2) we show that our unique kernel parameterization allows our model to achieve the same or higher accuracy with significantly fewer network parameters.""","""The paper presents a simple and effective convolution kernel for CNNs on spherical data (convolution by a linear combination of differential operators). The proposed method is efficient in the number of parameters and achieves strong classification and segmentation performance in several benchmarks. The paper is generally well written but the authors should clarify the details and address reviewer comments (for example, clarity/notations of equations) in the revision. """
888,"""Marginal Policy Gradients: A Unified Family of Estimators for Bounded Action Spaces with Applications""","['reinforcement learning', 'policy gradient', 'MOBA games']","""Many complex domains, such as robotics control and real-time strategy (RTS) games, require an agent to learn a continuous control. In the former, an agent learns a policy over R^d and in the latter, over a discrete set of actions each of which is parametrized by a continuous parameter. Such problems are naturally solved using policy based reinforcement learning (RL) methods, but unfortunately these often suffer from high variance leading to instability and slow convergence. Unnecessary variance is introduced whenever policies over bounded action spaces are modeled using distributions with unbounded support by applying a transformation T to the sampled action before execution in the environment. Recently, the variance reduced clipped action policy gradient (CAPG) was introduced for actions in bounded intervals, but to date no variance reduced methods exist when the action is a direction, something often seen in RTS games. To this end we introduce the angular policy gradient (APG), a stochastic policy gradient method for directional control. With the marginal policy gradients family of estimators we present a unified analysis of the variance reduction properties of APG and CAPG; our results provide a stronger guarantee than existing analyses for CAPG. Experimental results on a popular RTS game and a navigation task show that the APG estimator offers a substantial improvement over the standard policy gradient.""","""The paper introduces a new variance reduced policy gradient method, for directional and clipped action spaces, with provable guarantees that the gradient is lower variance. The paper is clearly written and the theory an important contribution. The experiments provide some preliminary insights that the algorithm could be beneficial in practice. """
889,"""Generative model based on minimizing exact empirical Wasserstein distance""","['Generative modeling', 'Generative Adversarial Networks (GANs)', 'Wasserstein GAN', 'Optimal transport']","""Generative Adversarial Networks (GANs) are a very powerful framework for generative modeling. However, they are often hard to train, and learning of GANs often becomes unstable. Wasserstein GAN (WGAN) is a promising framework to deal with the instability problem as it has a good convergence property. One drawback of the WGAN is that it evaluates the Wasserstein distance in the dual domain, which requires some approximation, so that it may fail to optimize the true Wasserstein distance. In this paper, we propose evaluating the exact empirical optimal transport cost efficiently in the primal domain and performing gradient descent with respect to its derivative to train the generator network. Experiments on the MNIST dataset show that our method is significantly stable to converge, and achieves the lowest Wasserstein distance among the WGAN variants at the cost of some sharpness of generated images. Experiments on the 8-Gaussian toy dataset show that better gradients for the generator are obtained in our method. In addition, the proposed method enables more flexible generative modeling than WGAN.""","""This method proposes a primal approach to minimizing Wasserstein distance for generative models. It estimates WD by computing the exact WD between empirical distributions. As the reviewers point out, the primal approach has been studied by other papers (which this submission doesn't cite, even in the revision), and suffers from a well-known problem of high variance. The authors have not responded to key criticisms of the reviewers. I don't think this work is ready for publication in ICLR. """
890,"""FAVAE: SEQUENCE DISENTANGLEMENT USING IN- FORMATION BOTTLENECK PRINCIPLE""",['disentangled representation learning'],"""A state-of-the-art generative model, a factorized action variational autoencoder (FAVAE), is presented for learning disentangled and interpretable representations from sequential data via the information bottleneck without supervision. The purpose of disentangled representation learning is to obtain interpretable and transferable representations from data. We focused on the disentangled representation of sequential data because there is a wide range of potential applications if disentanglement representation is extended to sequential data such as video, speech, and stock price data. Sequential data is characterized by dynamic factors and static factors: dynamic factors are time-dependent, and static factors are independent of time. Previous works succeed in disentangling static factors and dynamic factors by explicitly modeling the priors of latent variables to distinguish between static and dynamic factors. However, this model can not disentangle representations between dynamic factors, such as disentangling picking and throwing in robotic tasks. In this paper, we propose new model that can disentangle multiple dynamic factors. Since our method does not require modeling priors, it is capable of disentangling between dynamic factors. In experiments, we show that FAVAE can extract the disentangled dynamic factors.""","""This paper introduces an autoencoder architecture that can handle sequences of data, and attempts to automatically disentangle multiple static and dynamic factors. Quality: The main idea is relatively well-motivated. However the motivation for the particular technical choices made seems a little lacking, and the complexity of the proposed model put a lot of strain on the experiments. A lot of important updates were made by the authors in the rebuttal period, however I feel the number of changes are a lot to ask the reviewers to re-evaluate. Clarity: The English of the paper isn't great, including the title (should be ""Using an ..."" or ""Using the ...""). The intro is clear enough, but belabors a relatively simple point about how an image model can't model factors in video. There were some concerning parts where major issues seemed to be glossed over. E.g. ""FHVAE model uses label information to disentangle time series data, which is different setup with our FAVAE model."" As far as I understand, they both are trained from unsupervised data. Originality: This paper does a good job of citing related work, but seems incremental in relation to the FHVAE. But the main problem is that the proposed method makes a lot of changes from a standard time-series VAE, and the limited number of experiments means it's hard to say what the important factor in this model's performance is. Significance: Ultimately it's hard to say what the takeaway from this paper is. The authors motivated and evaluated a new model, but the work wasn't done in a systematic enough way to make an strong conclusions. What conclusion were asserted seem specious and overly general, e.g. "" Since dynamic factors have the same time dependency, these models cannot disentangle dynamic factors."". Why not? Why can't a dynamic model learn the time-scales of each of its factors automatically? """
891,"""Selective Convolutional Units: Improving CNNs via Channel Selectivity""","['convolutional neural networks', 'channel-selectivity', 'channel re-wiring', 'bottleneck architectures', 'deep learning']","""Bottleneck structures with identity (e.g., residual) connection are now emerging popular paradigms for designing deep convolutional neural networks (CNN), for processing large-scale features efficiently. In this paper, we focus on the information-preserving nature of identity connection and utilize this to enable a convolutional layer to have a new functionality of channel-selectivity, i.e., re-distributing its computations to important channels. In particular, we propose Selective Convolutional Unit (SCU), a widely-applicable architectural unit that improves parameter efficiency of various modern CNNs with bottlenecks. During training, SCU gradually learns the channel-selectivity on-the-fly via the alternative usage of (a) pruning unimportant channels, and (b) rewiring the pruned parameters to important channels. The rewired parameters emphasize the target channel in a way that selectively enlarges the convolutional kernels corresponding to it. Our experimental results demonstrate that the SCU-based models without any postprocessing generally achieve both model compression and accuracy improvement compared to the baselines, consistently for all tested architectures.""","""This paper proposed Selective Convolutional Unit (SCU) for improving the 1x1 convolutions used in the bottleneck of a ResNet block. The main idea is to remove channels of low importance and replace them by other ones which are in a similar fashion found to be important. To this end the authors propose the so-called expected channel damage score (ECDS) which is used for channel selection. The authors also show the effectiveness of SCU on CIFAR-10, CIFAR-100 and Imagenet. The major concerns from various reviewers are that the design seems the over-complicated as well as the experiments are not state-of-the-art. In response, the authors add some explanations on the design idea and new experiments of DenseNet-BC-190 on CIFAR10/100. But the reviewers major concerns are still there and did not change their ratings (6,5,5). Based on current results, the paper is proposed for borderline lean reject. """
892,"""RANDOM MASK: Towards Robust Convolutional Neural Networks""","['adversarial examples', 'robust machine learning', 'cnn structure', 'metric', 'deep feature representations']","""Robustness of neural networks has recently been highlighted by the adversarial examples, i.e., inputs added with well-designed perturbations which are imperceptible to humans but can cause the network to give incorrect outputs. In this paper, we design a new CNN architecture that by itself has good robustness. We introduce a simple but powerful technique, Random Mask, to modify existing CNN structures. We show that CNN with Random Mask achieves state-of-the-art performance against black-box adversarial attacks without applying any adversarial training. We next investigate the adversarial examples which fool a CNN with Random Mask. Surprisingly, we find that these adversarial examples often fool humans as well. This raises fundamental questions on how to define adversarial examples and robustness properly.""","""This paper presents a new technique for modifying neural network structure, and suggest that this structure provides improved robustness to black-box attacks, as compared to standard architectures. The paper is very thorough in its experimentation, and the method is simple and quite easy to understand. It also raises some important questions about adversarial examples. However, there are serious concerns regarding the evaluation methodology. In particular, the authors claim ""black-box robustness"" but do not test against any query-based attacks, which are known to perform better against gradient masking-based adversarial defenses. Furthermore, it is not clear why one would expect adversarial examples to transfer between models representing two completely different functions (i.e. from a standard model to a random mask model). So, the gray-box evaluation is much more informative and, unfortunately, random-mask seems to provide little to no robustness in this setting. Given how fundamental sound and convincing evaluation is for proposed defense methods, the submission is not ready for publication yet. In particular, the authors are urged to (a) evaluate on stronger black-box attacks, and (b) compare to a baseline that is known to be non-robust, (e.g. JPEG encoding or SAP), to verify that these results are actually due to black-box robustness and not simply obfuscation."""
893,"""LanczosNet: Multi-Scale Deep Graph Convolutional Networks""","['Lanczos Network', 'Graph Neural Networks', 'Deep Graph Convolutional Networks', 'Deep Learning on Graph Structured Data', 'QM8 Quantum Chemistry Benchmark']","""We propose Lanczos network (LanczosNet) which uses the Lanczos algorithm to construct low rank approximations of the graph Laplacian for graph convolution. Relying on the tridiagonal decomposition of the Lanczos algorithm, we not only efficiently exploit multi-scale information via fast approximated computation of matrix power but also design learnable spectral filters. Being fully differentiable, LanczosNet facilitates both graph kernel learning as well as learning node embeddings. We show the connection between our LanczosNet and graph based manifold learning, especially diffusion maps. We benchmark our model against pseudo-formula recent deep graph networks on citation datasets and QM8 quantum chemistry dataset. Experimental results show that our model achieves the state-of-the-art performance in most tasks.""","""The reviewers unanimously agreed that the paper was a significant advance in the field of machine learning on graph-structured inputs. They commented particularly on the quality of the research idea, and its depth of development. The results shared by the researchers are compelling, and they also report optimal hyperparameters, a welcome practice when describing experiments and results. A small drawback the reviewers highlighted is the breadth of the content in the paper, which gave the impression of a slight lack of focus. Overall, the paper is a clear advance, and I recommend it for acceptance. """
894,"""Offline Deep models calibration with bayesian neural networks""","['calibration', 'deep models', 'bayesian neural networks']","""We apply Bayesian Neural Networks to improve calibration of state-of-the-art deep neural networks. We show that, even with the most basic amortized approximate posterior distribution, and fast fully connected neural network for the likelihood, the Bayesian framework clearly outperforms other simple maximum likelihood based solutions that have recently shown very good performance, as temperature scaling. As an example, we reduce the Expected Calibration Error (ECE) from 0.52 to 0.24 on CIFAR-10 and from 4.28 to 2.456 on CIFAR-100 on two Wide ResNet with 96.13% and 80.39% accuracy respectively, which are among the best results published for this task. We demonstrate our robustness and performance with experiments on a wide set of state-of-the-art computer vision models. Moreover, our approach acts off-line, and thus can be applied to any probabilistic model regardless of the limitations that the model may present during training. This make it suitable to calibrate systems that make use of pre-trained deep neural networks that are expensive to train for a specific task, or to directly train a calibrated deep convolutional model with Monte Carlo Dropout approximations, among others. However, our method is still complementary with any Bayesian Neural Network for further improvement.""","""Reviewers are in a consensus and recommended to reject after engaging with the authors. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit. """
895,"""COMPOSITION AND DECOMPOSITION OF GANS""",[],"""In this work, we propose a composition/decomposition framework for adversarially training generative models on composed data - data where each sample can be thought of as being constructed from a fixed number of components. In our framework, samples are generated by sampling components from component generators and feeding these components to a composition function which combines them into a composed sample. This compositional training approach improves the modularity, extensibility and interpretability of Generative Adversarial Networks (GANs) - providing a principled way to incrementally construct complex models out of simpler component models, and allowing for explicit division of responsibility between these components. Using this framework, we define a family of learning tasks and evaluate their feasibility on two datasets in two different data modalities (image and text). Lastly, we derive sufficient conditions such that these compositional generative models are identifiable. Our work provides a principled approach to building on pretrained generative models or for exploiting the compositional nature of data distributions to train extensible and interpretable models. ""","""This paper investigates composition and decomposition for adversarially training generative models that work on composed data. Components that are sampled from component generators are then fed into a composition function to generate composed samples, aiming to improve modularity, extensibility, and interpretability of GANs. The paper is written very clearly and is easy to follow. Experiments considered application to both images (MNIST) and text (yelp reviews). The original version of the paper lacks any qualitative analysis, even though experiments were described. Authors revised the paper to include some experimental results, however, they are still not sufficient. State-of-the-art baselines, from previous work suggested by the reviewers should be included for comparison."""
896,"""SOM-VAE: Interpretable Discrete Representation Learning on Time Series""","['deep learning', 'self-organizing map', 'variational autoencoder', 'representation learning', 'time series', 'machine learning', 'interpretability']","""High-dimensional time series are common in many domains. Since human cognition is not optimized to work well in high-dimensional spaces, these areas could benefit from interpretable low-dimensional representations. However, most representation learning algorithms for time series data are difficult to interpret. This is due to non-intuitive mappings from data features to salient properties of the representation and non-smoothness over time. To address this problem, we propose a new representation learning framework building on ideas from interpretable discrete dimensionality reduction and deep generative modeling. This framework allows us to learn discrete representations of time series, which give rise to smooth and interpretable embeddings with superior clustering performance. We introduce a new way to overcome the non-differentiability in discrete representation learning and present a gradient-based version of the traditional self-organizing map algorithm that is more performant than the original. Furthermore, to allow for a probabilistic interpretation of our method, we integrate a Markov model in the representation space. This model uncovers the temporal transition structure, improves clustering performance even further and provides additional explanatory insights as well as a natural representation of uncertainty. We evaluate our model in terms of clustering performance and interpretability on static (Fashion-)MNIST data, a time series of linearly interpolated (Fashion-)MNIST images, a chaotic Lorenz attractor system with two macro states, as well as on a challenging real world medical time series application on the eICU data set. Our learned representations compare favorably with competitor methods and facilitate downstream tasks on the real world data.""","""This paper combines probabilistic models, VAEs, and self-organizing maps to learn interpretable representations on time series. The proposed contributions are a novel and interesting combination of existing ideas, in particular, the extension to time-series data by modeling the cluster dynamics. The empirical results show improved unsupervised clustering performance, on both synthetic and real datasets, compared to a number of baselines. The resulting 2D embedding also provides an interpretable visualization. The reviewers and the AC identified a number of potential weaknesses in the presentation in the original submission: (1) there was insufficient background on SOMs, leaving the readers unable to comprehend the contributions, (2) some of the details about the experiments were missing, such as how the baselines were constructed, (3) additional experiments were needed in regards to the hyper-parameters, such as number of clusters and the weighting in the loss, and (4) Figure 4d required a description of the results. The revision and the comments by the authors addressed most of these comments, and the reviewers felt that their concerns had been alleviated. Thus, the reviewers felt the paper should be accepted."""
897,"""Activity Regularization for Continual Learning""","['continual learning', 'regularization']","""While deep neural networks have achieved remarkable successes, they suffer the well-known catastrophic forgetting issue when switching from existing tasks to tackle a new one. In this paper, we study continual learning with deep neural networks that learn from tasks arriving sequentially. We first propose an approximated multi-task learning framework that unifies a family of popular regularization based continual learning methods. We then analyze the weakness of existing approaches, and propose a novel regularization method named Activity Regularization (AR), which alleviates forgetting meanwhile keeping models plasticity to acquire new knowledge. Extensive experiments show that our method outperform state-of-the-art methods and effectively overcomes catastrophic forgetting. ""","""There is no author response for this paper. The paper presents a multi-task learning framework as a unified view on the previous methods for tackling catastrophic forgetting in continual learning. In light of this framework, the authors propose to minimize the KL-divergence between the predictions of the previous optimal model and the current model using some stored samples from the previous tasks. The consensus among all three reviewers and AC is that the paper lacks (1) novelty, as the proposed approach is similar if not identical to Learning without forgetting (LwF)[Li&Hoiem 2017] with the difference that the KL-divergence is computed on samples kept from the previous tasks (and LwF uses samples from the current task). Methodological and experimental comparison to LwF is crucial to assess the benefits and novelty of the proposed approach. Also the reviewers address other potential weaknesses and give suggestions for improvement: (2) empirical evaluations can be substantially improved with sensitivity analysis of the hyper-parameters on the validation data (R3), indicating errors and error bars for all results (R3 and R2), using more challenging and realistic experimental setting where the data comes from different domains (R1), justifying the results better -- see R2s questions; (3) lack of clarity and motivation in Section 3.1 -- see R2s and R1s suggestions for how to improve clarity and potentially take advantage of the current task to probably correct the previous models prediction when it was wrong. AC suggests, in its current state the manuscript is not ready for a publication. We hope the reviews are useful for improving and revising the paper. """
898,"""The Anisotropic Noise in Stochastic Gradient Descent: Its Behavior of Escaping from Minima and Regularization Effects""","['Stochastic gradient descent', 'anisotropic noise', 'regularization']","""Understanding the behavior of stochastic gradient descent (SGD) in the context of deep neural networks has raised lots of concerns recently. Along this line, we theoretically study a general form of gradient based optimization dynamics with unbiased noise, which unifies SGD and standard Langevin dynamics. Through investigating this general optimization dynamics, we analyze the behavior of SGD on escaping from minima and its regularization effects. A novel indicator is derived to characterize the efficiency of escaping from minima through measuring the alignment of noise covariance and the curvature of loss function. Based on this indicator, two conditions are established to show which type of noise structure is superior to isotropic noise in term of escaping efficiency. We further show that the anisotropic noise in SGD satisfies the two conditions, and thus helps to escape from sharp and poor minima effectively, towards more stable and flat minima that typically generalize well. We verify our understanding through comparing this anisotropic diffusion with full gradient descent plus isotropic diffusion (i.e. Langevin dynamics) and other types of position-dependent noise.""","""The reviewers point our concerns regarding paper's novelty, theoretical soundness, and empirical strength. The authors provided to clarifications to the reviewers."""
899,"""Convergence Guarantees for RMSProp and ADAM in Non-Convex Optimization and an Empirical Comparison to Nesterov Acceleration""","['adaptive gradient descent', 'deeplearning', 'ADAM', 'RMSProp', 'autoencoders']","""RMSProp and ADAM continue to be extremely popular algorithms for training neural nets but their theoretical convergence properties have remained unclear. Further, recent work has seemed to suggest that these algorithms have worse generalization properties when compared to carefully tuned stochastic gradient descent or its momentum variants. In this work, we make progress towards a deeper understanding of ADAM and RMSProp in two ways. First, we provide proofs that these adaptive gradient algorithms are guaranteed to reach criticality for smooth non-convex objectives, and we give bounds on the running time. Next we design experiments to empirically study the convergence and generalization properties of RMSProp and ADAM against Nesterov's Accelerated Gradient method on a variety of common autoencoder setups and on VGG-9 with CIFAR-10. Through these experiments we demonstrate the interesting sensitivity that ADAM has to its momentum parameter \beta_1. We show that at very high values of the momentum parameter (\beta_1 = 0.99) ADAM outperforms a carefully tuned NAG on most of our experiments, in terms of getting lower training and test losses. On the other hand, NAG can sometimes do better when ADAM's \beta_1 is set to the most commonly used value: \beta_1 = 0.9, indicating the importance of tuning the hyperparameters of ADAM to get better generalization performance. We also report experiments on different autoencoders to demonstrate that NAG has better abilities in terms of reducing the gradient norms, and it also produces iterates which exhibit an increasing trend for the minimum eigenvalue of the Hessian of the loss function at the iterates. ""","""The reviewers and ACs acknowledge that the paper has a solid theoretical contribution because it give a convergence (to critical points) of the ADAM and RMSprop algorithms, and also shows that NAG can be tuned to match or outperform SGD in test errors. However, reviewers and the AC also note that potential improvements for the paper a) the exposition/notations can be improved; b) better comparison to the prior work could be made; c) the theoretical and empirical parts of the paper are somewhat disconnected; d) the proof has an error (that is fixed by the authors with additional assumptions.) Therefore, the paper is not quite ready for publications right now but the AC encourages the authors to submit revisions to other top ML venues. """
900,"""Conditional Network Embeddings""","['Network embedding', 'graph embedding', 'learning node representations', 'link prediction', 'multi-label classification of nodes']","""Network Embeddings (NEs) map the nodes of a given network into pseudo-formula -dimensional Euclidean space pseudo-formula . Ideally, this mapping is such that 'similar' nodes are mapped onto nearby points, such that the NE can be used for purposes such as link prediction (if 'similar' means being 'more likely to be connected') or classification (if 'similar' means 'being more likely to have the same label'). In recent years various methods for NE have been introduced, all following a similar strategy: defining a notion of similarity between nodes (typically some distance measure within the network), a distance measure in the embedding space, and a loss function that penalizes large distances for similar nodes and small distances for dissimilar nodes. A difficulty faced by existing methods is that certain networks are fundamentally hard to embed due to their structural properties: (approximate) multipartiteness, certain degree distributions, assortativity, etc. To overcome this, we introduce a conceptual innovation to the NE literature and propose to create \emph{Conditional Network Embeddings} (CNEs); embeddings that maximally add information with respect to given structural properties (e.g. node degrees, block densities, etc.). We use a simple Bayesian approach to achieve this, and propose a block stochastic gradient descent algorithm for fitting it efficiently. We demonstrate that CNEs are superior for link prediction and multi-label classification when compared to state-of-the-art methods, and this without adding significant mathematical or computational complexity. Finally, we illustrate the potential of CNE for network visualization.""","""The conditional network embedding approach proposed in the paper seems nice and novel, and consistently outperforms state-of-art on variety of datasets; scalability demonstration was added during rebuttals, as well as multiple other improvements; although the reviewers did not respond by changing the scores, this paper with augmentations provided during the rebuttal appears to be a useful contribution worthy of publishing at ICLR. """
901,"""Initialized Equilibrium Propagation for Backprop-Free Training""","['credit assignment', 'energy-based models', 'biologically plausible learning']","""Deep neural networks are almost universally trained with reverse-mode automatic differentiation (a.k.a. backpropagation). Biological networks, on the other hand, appear to lack any mechanism for sending gradients back to their input neurons, and thus cannot be learning in this way. In response to this, Scellier & Bengio (2017) proposed Equilibrium Propagation - a method for gradient-based train- ing of neural networks which uses only local learning rules and, crucially, does not rely on neurons having a mechanism for back-propagating an error gradient. Equilibrium propagation, however, has a major practical limitation: inference involves doing an iterative optimization of neural activations to find a fixed-point, and the number of steps required to closely approximate this fixed point scales poorly with the depth of the network. In response to this problem, we propose Initialized Equilibrium Propagation, which trains a feedforward network to initialize the iterative inference procedure for Equilibrium propagation. This feed-forward network learns to approximate the state of the fixed-point using a local learning rule. After training, we can simply use this initializing network for inference, resulting in a learned feedforward network. Our experiments show that this network appears to work as well or better than the original version of Equilibrium propagation. This shows how we might go about training deep networks without using backpropagation.""","""The paper investigates a novel initialisation method to improve Equilibrium Propagation. In particular, the results are convincing, but the reviewers remain with small issues here and there. An issue with the paper is the biological plausibility of the approach. Nonetheless publication is recommended. """
902,"""Adversarial Examples Are a Natural Consequence of Test Error in Noise""","['Adversarial examples', 'generalization']",""" Over the last few years, the phenomenon of adversarial examples --- maliciously constructed inputs that fool trained machine learning models --- has captured the attention of the research community, especially when the adversary is restricted to making small modifications of a correctly handled input. At the same time, less surprisingly, image classifiers lack human-level performance on randomly corrupted images, such as images with additive Gaussian noise. In this work, we show that these are two manifestations of the same underlying phenomenon. We establish this connection in several ways. First, we find that adversarial examples exist at the same distance scales we would expect from a linear model with the same performance on corrupted images. Next, we show that Gaussian data augmentation during training improves robustness to small adversarial perturbations and that adversarial training improves robustness to several types of image corruptions. Finally, we present a model-independent upper bound on the distance from a corrupted image to its nearest error given test performance and show that in practice we already come close to achieving the bound, so that improving robustness further for the corrupted image distribution requires significantly reducing test error. All of this suggests that improving adversarial robustness should go hand in hand with improving performance in the presence of more general and realistic image corruptions. This yields a computationally tractable evaluation metric for defenses to consider: test error in noisy image distributions.""","""In light of the reviews and the rebuttal, it seems that the paper needs to be rewritten to head off some of the confusions and criticisms that the reviewers have made. That said, the main argument seems to contradict some of the lower bounds recently established by Madry and colleagues, showing the existence of distributions where the sample complexity for finding robust classifiers is arbitrarily larger than that for finding low-risk classifiers. I recommend the authors take a closer look at this apparent contradiction when revising."""
903,"""Probabilistic Binary Neural Networks""","['binary neural Network', 'efficient deep learning', 'stochastic training', 'discrete neural network', 'efficient inference']","""Low bit-width weights and activations are an effective way of combating the increasing need for both memory and compute power of Deep Neural Networks. In this work, we present a probabilistic training method for Neural Network with both binary weights and activations, called PBNet. By embracing stochasticity during training, we circumvent the need to approximate the gradient of functions for which the derivative is zero almost always, such as pseudo-formula , while still obtaining a fully Binary Neural Network at test time. Moreover, it allows for anytime ensemble predictions for improved performance and uncertainty estimates by sampling from the weight distribution. Since all operations in a layer of the PBNet operate on random variables, we introduce stochastic versions of Batch Normalization and max pooling, which transfer well to a deterministic network at test time. We evaluate two related training methods for the PBNet: one in which activation distributions are propagated throughout the network, and one in which binary activations are sampled in each layer. Our experiments indicate that sampling the binary activations is an important element for stochastic training of binary Neural Networks. ""","""The paper proposes a probabilistic training method for binary Neural Network with stochastic versions of Batch Normalization and max pooling. The reviewers and AC note the following potential weaknesses: (1) limited novelty and (2) preliminary experimental results. AC thinks the proposed method has potential and is interesting, but decided that the authors need more works to publish."""
904,"""MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders""","['VAE', 'regularization', 'auto-regressive']","""Variational Autoencoder (VAE), a simple and effective deep generative model, has led to a number of impressive empirical successes and spawned many advanced variants and theoretical investigations. However, recent studies demonstrate that, when equipped with expressive generative distributions (aka. decoders), VAE suffers from learning uninformative latent representations with the observation called KL Varnishing, in which case VAE collapses into an unconditional generative model. In this work, we introduce mutual posterior-divergence regularization, a novel regularization that is able to control the geometry of the latent space to accomplish meaningful representation learning, while achieving comparable or superior capability of density estimation.Experiments on three image benchmark datasets demonstrate that, when equipped with powerful decoders, our model performs well both on density estimation and representation learning.""","""This paper proposes a solution for the well-known problem of posterior collapse in VAEs: a phenomenon where the posteriors fail to diverge from the prior, which tends to happen in situations where the decoder is overly flexible. A downside of the proposed method is the introduction of hyper-parameters controlling the degree of regularization. The empirical results show improvements on various baselines. The paper proposes the addition of a regularization term that penalizes pairwise similarity of posteriors in latent space. The reviewers agree that the paper is clearly written and that the method is reasonably motivated. The experiments are also sufficiently convincing."""
905,"""A Universal Music Translation Network""",[],"""We present a method for translating music across musical instruments and styles. This method is based on unsupervised training of a multi-domain wavenet autoencoder, with a shared encoder and a domain-independent latent space that is trained end-to-end on waveforms. Employing a diverse training dataset and large net capacity, the single encoder allows us to translate also from musical domains that were not seen during training. We evaluate our method on a dataset collected from professional musicians, and achieve convincing translations. We also study the properties of the obtained translation and demonstrate translating even from a whistle, potentially enabling the creation of instrumental music by untrained humans. ""","""The paper describes a method which, given a music waveform, generates another recording of the same music which should sound as if it was performed by different instruments. The model is an auto-encoder with a WaveNet-like domain-specific decoder and a shared encoder, trained with an adversarial ""domain confusion loss"". Even though the method is constructed mostly from existing components, the reviewers found the results interesting and convincing, and recommended the paper for acceptance."""
906,"""Supervised Policy Update for Deep Reinforcement Learning""",['Deep Reinforcement Learning'],"""We propose a new sample-efficient methodology, called Supervised Policy Update (SPU), for deep reinforcement learning. Starting with data generated by the current policy, SPU formulates and solves a constrained optimization problem in the non-parameterized proximal policy space. Using supervised regression, it then converts the optimal non-parameterized policy to a parameterized policy, from which it draws new samples. The methodology is general in that it applies to both discrete and continuous action spaces, and can handle a wide variety of proximity constraints for the non-parameterized optimization problem. We show how the Natural Policy Gradient and Trust Region Policy Optimization (NPG/TRPO) problems, and the Proximal Policy Optimization (PPO) problem can be addressed by this methodology. The SPU implementation is much simpler than TRPO. In terms of sample efficiency, our extensive experiments show SPU outperforms TRPO in Mujoco simulated robotic tasks and outperforms PPO in Atari video game tasks.""","""The paper presents an interesting technique for constrained policy optimization, which is applicable to existing RL algorithms such as TRPO and PPO. All of the reviewers agree that the paper is above the bar and the authors have improved the exposition during the review process. I encourage the authors to address all of the comments in the final version."""
907,"""A Variational Autoencoder for Probabilistic Non-Negative Matrix Factorisation""","['Non-negative matrix factorisation', 'Variational autoencoder', 'Probabilistic']","""We introduce and demonstrate the variational autoencoder (VAE) for probabilistic non-negative matrix factorisation (PAE-NMF). We design a network which can perform non-negative matrix factorisation (NMF) and add in aspects of a VAE to make the coefficients of the latent space probabilistic. By restricting the weights in the final layer of the network to be non-negative and using the non-negative Weibull distribution we produce a probabilistic form of NMF which allows us to generate new data and find a probability distribution that effectively links the latent and input variables. We demonstrate the effectiveness of PAE-NMF on three heterogeneous datasets: images, financial time series and genomic.""","""The paper introduces a variant of the variational autoencoder (VAE) for probabilistic non-negative matrix factorization. The main idea is to use a Weibull distribution in the latent space. There is agreement among the reviewers that the paper is technically sound and well written, but that it lacks in motivation and demonstration of utility of the proposed method. All the reviewers think the approach is not particularly novel and somewhat incremental. The main issue is that the empirical evaluation of the algorithm is also quite limited. Specifically, it should have been compared with Bayesian NMF. Many papers have addressed Bayesian NMF with variational inference (Cemgil; Fevotte & Dikmen; Hoffman, Blei & Cook) like in VAE. Experimentally, Bayesian NMF and the proposed PAE-NMF could easily be quantitatively compared on matrix completion tasks. Overall, there was consensus among the reviewers that the paper is not ready for publication. """
908,"""Latent Convolutional Models""","['latent models', 'convolutional networks', 'unsupervised learning', 'deep learning', 'modeling natural images', 'image restoration']","""We present a new latent model of natural images that can be learned on large-scale datasets. The learning process provides a latent embedding for every image in the training dataset, as well as a deep convolutional network that maps the latent space to the image space. After training, the new model provides a strong and universal image prior for a variety of image restoration tasks such as large-hole inpainting, superresolution, and colorization. To model high-resolution natural images, our approach uses latent spaces of very high dimensionality (one to two orders of magnitude higher than previous latent image models). To tackle this high dimensionality, we use latent spaces with a special manifold structure (convolutional manifolds) parameterized by a ConvNet of a certain architecture. In the experiments, we compare the learned latent models with latent models learned by autoencoders, advanced variants of generative adversarial networks, and a strong baseline system using simpler parameterization of the latent space. Our model outperforms the competing approaches over a range of restoration tasks.""","""The reviewers are in general impressed by the results and like the idea but they also express some uncertainty about how the proposed actually is set up. The authors have made a good attempt to address the reviewers' concerns. """
909,"""Alignment Based Mathching Networks for One-Shot Classification and Open-Set Recognition""",[],"""Deep learning for object classification relies heavily on convolutional models. While effective, CNNs are rarely interpretable after the fact. An attention mechanism can be used to highlight the area of the image that the model focuses on thus offering a narrow view into the mechanism of classification. We expand on this idea by forcing the method to explicitly align images to be classified to reference images representing the classes. The mechanism of alignment is learned and therefore does not require that the reference objects are anything like those being classified. Beyond explanation, our exemplar based cross-alignment method enables classification with only a single example per category (one-shot). Our model cuts the 5-way, 1-shot error rate in Omniglot from 2.1\% to 1.4\% and in MiniImageNet from 53.5\% to 46.5\% while simultaneously providing point-wise alignment information providing some understanding on what the network is capturing. This method of alignment also enables the recognition of an unsupported class (open-set) in the one-shot setting while maintaining an F1-score of above 0.5 for Omniglot even with 19 other distracting classes while baselines completely fail to separate the open-set class in the one-shot setting.""","""The reviewers are polarized on this paper and the overall feeling is that it is not quite ready for publication. There is also an interesting interpretability aspect that, while given as a motivation for the approach, is never really explored beyond showing some figures of alignments. One of the main concerns of the methods effectiveness in practice is the computational cost. There is also concern from one of the reviewers that the formulation could result in creating sparse matching maps where only a few pixels get matched. The authors provide some justification for why this wouldnt happen, and this should be put in a future draft. Even better would be to show statistics to demonstrate empirically that this doesnt happen. There were a number of clarifications that were brought up during the discussion, and the authors should go over this carefully and update the draft to resolve these issues. There is also a typo in the title that should be fixed. """
910,"""A Deep Learning Approach for Dynamic Survival Analysis with Competing Risks""","['dynamic survival analysis', 'survival analysis', 'longitudinal measurements', 'competing risks']","""Currently available survival analysis methods are limited in their ability to deal with complex, heterogeneous, and longitudinal data such as that available in primary care records, or in their ability to deal with multiple competing risks. This paper develops a novel deep learning architecture that flexibly incorporates the available longitudinal data comprising various repeated measurements (rather than only the last available measurements) in order to issue dynamically updated survival predictions for one or multiple competing risk(s). Unlike existing works in the survival analysis on the basis of longitudinal data, the proposed method learns the time-to-event distributions without specifying underlying stochastic assumptions of the longitudinal or the time-to-event processes. Thus, our method is able to learn associations between the longitudinal data and the various associated risks in a fully data-driven fashion. We demonstrate the power of our method by applying it to real-world longitudinal datasets and show a drastic improvement over state-of-the-art methods in discriminative performance. Furthermore, our analysis of the variable importance and dynamic survival predictions will yield a better understanding of the predicted risks which will result in more effective health care.""","""While there was disagreement on this paper, reviewers remained unconvinced about the scalability and novelty of the presented work. While it was universally agreed that many positive points exist in this paper, it is not yet ready for publication. """
911,"""Seq2Slate: Re-ranking and Slate Optimization with RNNs""","['Recurrent neural networks', 'learning to rank', 'pointer networks']","""Ranking is a central task in machine learning and information retrieval. In this task, it is especially important to present the user with a slate of items that is appealing as a whole. This in turn requires taking into account interactions between items, since intuitively, placing an item on the slate affects the decision of which other items should be chosen alongside it. In this work, we propose a sequence-to-sequence model for ranking called seq2slate. At each step, the model predicts the next item to place on the slate given the items already chosen. The recurrent nature of the model allows complex dependencies between items to be captured directly in a flexible and scalable way. We show how to learn the model end-to-end from weak supervision in the form of easily obtained click-through data. We further demonstrate the usefulness of our approach in experiments on standard ranking benchmarks as well as in a real-world recommendation system.""","""The paper addresses the problem of learning to (re)rank slates of search results while optimizing some performance metric across the entire list of results (the slate). The work builds on a wealth of prior work on slate optimization from the information retrieval community, and proposes a novel approach to this problem, an extension of pointer networks, previously used in sequence learning tasks. The paper is motivated by an important real world application, and has potential for significant practical impact. Reviewers noted in particular the valuable evaluation in an A/B test against a strong production system - showing that the work has practical impact. Reviewers positively noted the discussion of practical issues related to applying the work at scale. The paper was found to be clearly written, and demonstrating a thorough understanding of related work. The authors and AC also note several potential weaknesses. Several of these were addressed by the authors, as follows. R3 asked for more breadth on metrics, and additional clarifications - the authors provided the requested information. Several questions were raised regarding the diverse-clicks setting and choice of hyperparameter \eta - both were discussed in the rebuttal. Further analysis / discussion of computational and performance trade-offs are requested and discussed. Overall, the main drawback of the paper, raised by all three reviewers, is the size of the contribution. The paper extends an approach called ""pointer networks"" to the model application setting considered here. The reviewers and AC agree that, while practically relevant and interesting, the research contribution of the resulting approach limited. As a result, the recommendation is to not accept the paper for publication at ICLR in its current form. """
912,"""When Will Gradient Methods Converge to Max-margin Classifier under ReLU Models?""","['gradient method', 'max-margin', 'ReLU model']","""We study the implicit bias of gradient descent methods in solving a binary classification problem over a linearly separable dataset. The classifier is described by a nonlinear ReLU model and the objective function adopts the exponential loss function. We first characterize the landscape of the loss function and show that there can exist spurious asymptotic local minima besides asymptotic global minima. We then show that gradient descent (GD) can converge to either a global or a local max-margin direction, or may diverge from the desired max-margin direction in a general context. For stochastic gradient descent (SGD), we show that it converges in expectation to either the global or the local max-margin direction if SGD converges. We further explore the implicit bias of these algorithms in learning a multi-neuron network under certain stationary conditions, and show that the learned classifier maximizes the margins of each sample pattern partition under the ReLU activation.""","""The reviewers and AC note the following potential weaknesses: 1) the proof techniques largley follow from previous work on linear models 2) its not clear how signficant it is to analyze a one-neuron ReLU model for linearly separable data. """
913,"""IEA: Inner Ensemble Average within a convolutional neural network""",['Ensemble Convolutional Neural Networks'],"""Ensemble learning is a method of combining multiple trained models to improve model accuracy. We propose the usage of such methods, specifically ensemble average, inside Convolutional Neural Network (CNN) architectures by replacing the single convolutional layers with Inner Average Ensembles (IEA) of multiple convolutional layers. Empirical results on different benchmarking datasets show that CNN models using IEA outperform those with regular convolutional layers and advances the state of art. A visual and a similarity score analysis of the features generated from IEA explains why it boosts the model performance.""","""The method under consideration uses parallel convolutional filter groups per layer, where activations are averaged between the groups, forming ""inner ensembles"". Reviewers raised a number of concerns, including the increased computational cost for apparently little performance gain, the choice of base architecture (later addressed with additional experiments using WideResNet and ResNeXt), issues of clarity of presentation (some of which were addressed). One reviewer was unconvinced without direct comparison to full ensembles. Another reviewer raised the issue of a missing direct comparison to the most similar method in the literature, maxout (Goodfellow et al, 2013). Authors rebutted this by claiming that maxout is difficult to implement and offering vague arguments for its inferiority to their method. The AC agrees that a maxout baseline is important here, as it is extremely close to the proposed method and also trivially implemented, and that in light of maxout (and other related methods) the degree of novelty is limited. The AC also concurs that a full ensemble baseline would strengthen the paper's claims. In the absence of either of these the AC concurs with the reviewers that this work is not suitable for publication at this time."""
914,"""Stochastic Optimization of Sorting Networks via Continuous Relaxations""","['continuous relaxations', 'sorting', 'permutation', 'stochastic computation graphs', 'Plackett-Luce']","""Sorting input objects is an important step in many machine learning pipelines. However, the sorting operator is non-differentiable with respect to its inputs, which prohibits end-to-end gradient-based optimization. In this work, we propose NeuralSort, a general-purpose continuous relaxation of the output of the sorting operator from permutation matrices to the set of unimodal row-stochastic matrices, where every row sums to one and has a distinct argmax. This relaxation permits straight-through optimization of any computational graph involve a sorting operation. Further, we use this relaxation to enable gradient-based stochastic optimization over the combinatorially large space of permutations by deriving a reparameterized gradient estimator for the Plackett-Luce family of distributions over permutations. We demonstrate the usefulness of our framework on three tasks that require learning semantic orderings of high-dimensional objects, including a fully differentiable, parameterized extension of the k-nearest neighbors algorithm""","""This paper proposes a general-purpose continuous relaxation of the output of the sorting operator. This enables end-to-end training to enable more efficient stochastic optimization over the combinatorially large space of permutations. In the submitted versions, two of the reviewers had difficulty in understanding the writing. After the rebuttal and the revised version, one of the reviewers is satisfied. I personally went through the paper and found that it could be tricky to read certain parts of the paper. For example, I am personally very familiar with the Placket-Luce model but the writing in Section 2.1 does not do a good job in explaining the model (particularly Eq 1 is not very easy to read, same with Eq. 3 for the key identity used in the paper). I encourage authors to improve writing and make it a bit more intuitive to read. Overall, this is a good paper and I recommend to accept it. """
915,"""Entropic GANs meet VAEs: A Statistical Approach to Compute Sample Likelihoods in GANs""","['GAN', 'VAE', 'likelihood estimation', 'statistical inference']","""Building on the success of deep learning, two modern approaches to learn a probability model of the observed data are Generative Adversarial Networks (GANs) and Variational AutoEncoders (VAEs). VAEs consider an explicit probability model for the data and compute a generative distribution by maximizing a variational lower-bound on the log-likelihood function. GANs, however, compute a generative model by minimizing a distance between observed and generated probability distributions without considering an explicit model for the observed data. The lack of having explicit probability models in GANs prohibits computation of sample likelihoods in their frameworks and limits their use in statistical inference problems. In this work, we show that an optimal transport GAN with the entropy regularization can be viewed as a generative model that maximizes a lower-bound on average sample likelihoods, an approach that VAEs are based on. In particular, our proof constructs an explicit probability model for GANs that can be used to compute likelihood statistics within GAN's framework. Our numerical results on several datasets demonstrate consistent trends with the proposed theory. ""","""The paper's strength is in that it shows the log likelihood objective is lower bounded by a GAN objective plus an entropy term. The theory is novel (but it seems to relate closely to the work pseudo-url.) The main drawback the reviewer raised includes a) it's not clear how tight the lower bound is; b) the theory only applies to a particular subcase of GANs --- it seems that the only reasonable instance that allows efficient generator is the case where Y = G(x)+\xi where \xi is Gaussian noise. The authors addressed the issue a) with some new experiments with linear generators and quadratic loss, but it lacks experiments with deep models which seems to be necessary since this is a critical issue. Based on this, the AC decided to recommend reject and would encourage the authors to add more experiments on the tightness of the lower bound with bigger models and submit to other top venues. """
916,"""NADPEx: An on-policy temporally consistent exploration method for deep reinforcement learning""","['Reinforcement learning', 'exploration']","""Reinforcement learning agents need exploratory behaviors to escape from local optima. These behaviors may include both immediate dithering perturbation and temporally consistent exploration. To achieve these, a stochastic policy model that is inherently consistent through a period of time is in desire, especially for tasks with either sparse rewards or long term information. In this work, we introduce a novel on-policy temporally consistent exploration strategy - Neural Adaptive Dropout Policy Exploration (NADPEx) - for deep reinforcement learning agents. Modeled as a global random variable for conditional distribution, dropout is incorporated to reinforcement learning policies, equipping them with inherent temporal consistency, even when the reward signals are sparse. Two factors, gradients' alignment with the objective and KL constraint in policy space, are discussed to guarantee NADPEx policy's stable improvement. Our experiments demonstrate that NADPEx solves tasks with sparse reward while naive exploration and parameter noise fail. It yields as well or even faster convergence in the standard mujoco benchmark for continuous control. ""","""The authors have proposed a new method for exploration that is related to parameter noise, but instead uses Gaussian dropout across entire episodes, thus allowing for temporally consistent exploration. The method is evaluated in sparsely rewarded continuous control domains such as half-cheetah and humanoid, and compared against PPO and other variants. The method is novel and does seem to work stably across the tested tasks, and simple exploration methods are important for the RL field. However, the paper is poorly and confusingly written and really really needs to be thoroughly edited before the camera ready deadline. There are many approaches which are referred to without any summary or description, which makes it difficult to read the paper. The three reviewers all had low confidence in their understanding of the paper, which makes this a very borderline submission even though the reviewers gave relatively high scores. """
917,"""Analysis of Memory Organization for Dynamic Neural Networks""","['memory analysis', 'recurrent neural network', 'LSTM', 'neural Turing machine', 'neural stack', 'differentiable neural computers']","""An increasing number of neural memory networks have been developed, leading to the need for a systematic approach to analyze and compare their underlying memory capabilities. Thus, in this paper, we propose a taxonomy for four popular dynamic models: vanilla recurrent neural network, long short-term memory, neural stack and neural RAM and their variants. Based on this taxonomy, we create a framework to analyze memory organization and then compare these network architectures. This analysis elucidates how different mapping functions capture the information in the past of the input, and helps to open the dynamic neural network black box from the perspective of memory usage. Four representative tasks that would fit optimally the characteristics of each memory network are carefully selected to show each network's expressive power. We also discuss how to use this taxonomy to help users select the most parsimonious type of memory network for a specific task. Two natural language processing applications are used to evaluate the methodology in a realistic setting. ""","""This paper presents a taxonomic study of neural network architectures, focussing on those which seek to map onto different part of the hierarchy of models of computation (DFAs, PDAs, etc). The paper splits between defining the taxonomy and comparing its elements on synthetic and ""NLP"" tasks (in fact, babi, which is also synthetic). I'm a fairly biased assessor of this sort of paper, as I generally like this topical area and think there is a need for more work of this nature in our field. I welcome, and believe the CFP calls for, papers like this (""learning representations of outputs or [structured] states"", ""theoretical issues in deep learning"")). However, despite my personal enthusiasm, the reviews tell a different story. The scores for this paper are all over the place, and that's after some attempt at harmonisation! I am satisfied that the authors have had a fair shot at defending their paper and that the reviewers have engaged with the discussion process. I'm afraid the emerging consensus still seems to be in favour of rejection. Despite my own views, I'm not comfortable bumping it up into acceptance territory on the basis of this assessment. Reviewer 1 is the only enthusiastic proponent of the paper, but their statement of support for the paper has done little to sway the others. The arguments by reviewer 3 specifically are quite salient: it is important to seek informative and useful taxonomies of the sort presented in this work, but they must have practical utility. From reading the paper, I share some of this reviewer's concerns: while it is clear to me what use there is the production of studies of the sort presented in this paper, it is not immediately clear what the utility of *this* study is. Would I, practically speaking, be able to make an informed choice as to what model class to attempt for a problem that wouldn't be indistinguishable from common approaches (e.g. ""start simple, add complexity""). I am afraid I agree with this reviewer that I would not. My conclusion is that there is not a strong consensus for accepting the paper. While I wouldn't mind seeing this work presented at the conference, but due to the competitive nature of the paper selection process, I'm afraid the line must be drawn somewhere. I do look forward to re-reading this paper after the authors have had a chance to improve and expand upon it."""
918,"""Non-Synergistic Variational Autoencoders""","['vae', 'unsupervised learning']","""Learning disentangling representations of the independent factors of variations that explain the data in an unsupervised setting is still a major challenge. In the following paper we address the task of disentanglement and introduce a new state-of-the-art approach called Non-synergistic variational Autoencoder (Non-Syn VAE). Our model draws inspiration from population coding, where the notion of synergy arises when we describe the encoded information by neurons in the form of responses from the stimuli. If those responses convey more information together than separate as independent sources of encoding information, they are acting synergetically. By penalizing the synergistic mutual information within the latents we encourage information independence and by doing that disentangle the latent factors. Notably, our approach could be added to the VAE framework easily, where the new ELBO function is still a lower bound on the log likelihood. In addition, we qualitatively compare our model with Factor VAE and show that this one implicitly minimises the synergy of the latents.""","""The paper introduces a form of variational auto encoder for learning disentangled representations. The idea is to penalise synergistic mutual information. The introduction of concepts from synergy to the community is appreciated. Although the approach appears interesting and forward looking in understanding complex models, at this point the paper does not convince on the theoretical nor on the experimental side. The main concepts used in the paper are developed elsewhere, the potential value of synergy is not properly examined. The reviewers agree on a not so positive view on this paper, with ratings either ok, but not good enough, or clear rejection. There is a consensus that the paper needs more work. """
919,"""Making Convolutional Networks Shift-Invariant Again""","['convolutional networks', 'signal processing', 'shift', 'translation', 'invariance', 'equivariance']","""Modern convolutional networks are not shift-invariant, despite their convolutional nature: small shifts in the input can cause drastic changes in the internal feature maps and output. In this paper, we isolate the cause -- the downsampling operation in convolutional and pooling layers -- and apply the appropriate signal processing fix -- low-pass filtering before downsampling. This simple architectural modification boosts the shift-equivariance of the internal representations and consequently, shift-invariance of the output. Importantly, this is achieved while maintaining downstream classification performance. In addition, incorporating the inductive bias of shift-invariance largely removes the need for shift-based data augmentation. Lastly, we observe that the modification induces spatially-smoother learned convolutional kernels. Our results suggest that this classical signal processing technique has a place in modern deep networks.""","""The reviewers are reasonably positive about this submission although two of them feel the paper is below acceptance threshold. AR1 advocates large scale experiments on ILSVRC2012/Cifar10/Cifar100 and so on. AR3 would like to see more comparisons to similar works and feels that the idea is not that significant. AR2 finds evaluations flawed. On balance, the reviewers find numerous flaws in experimentation that need to be improved. Additionally, AC is aware that approaches such as 'Convolutional Kernel Networks' by J. Mairal et al. derive a pooling layer which, by its motivation and design, obeys the sampling theorem to attain anti-aliasing. Essentially, for pooling, they obtain a convolution of feature maps with an appropriate Gaussian prior to sampling. Thus, on balance, the idea proposed in this ICLR submission may sound novel but it is not. Ideas such as 'blurring before downsampling' or 'low-pass filter kernels' applied here are simply special cases of anti-aliasing. The authors may also want to read about aliasing in 'Invariance, Stability, and Complexity of Deep Convolutional Representations' to see how to prevent aliasing. On balance, the theory behind this problem is mostly solved even if standard networks overlook this mechanism. Note also that there exist a fundamental trade-off between shift-invariance plus anti-aliasing (stability) and performance; this being a reason why max-pooling is still preferred over anti-aliasing (better performance versus stability). Though, this is nothing new for those who delve into more theoretical papers on CNNs: this is an invite for the authors to go thoroughly first through the relevant literature/numerous prior works on this topic."""
920,"""The Effectiveness of Pre-Trained Code Embeddings""","['machine learning', 'deep learning', 'summarization', 'embeddings', 'word embeddings', 'source code', 'programming languages', 'programming language processing']","""Word embeddings are widely used in machine learning based natural language processing systems. It is common to use pre-trained word embeddings which provide benefits such as reduced training time and improved overall performance. There has been a recent interest in applying natural language processing techniques to programming languages. However, none of this recent work uses pre-trained embeddings on code tokens. Using extreme summarization as the downstream task, we show that using pre-trained embeddings on code tokens provides the same benefits as it does to natural languages, achieving: over 1.9x speedup, 5\% improvement in test loss, 4\% improvement in F1 scores, and resistance to over-fitting. We also show that the choice of language used for the embeddings does not have to match that of the task to achieve these benefits and that even embeddings pre-trained on human languages provide these benefits to programming languages. ""","""All three reviewers agree that the research questionshould pretrained embeddings be used in code understanding tasksis a reasonable one. However, there were some early issues with the way in which the paper reported results (involving both metrics and baselines). After some discussion with the reviewers, it seems that the paper now presents a clear picture of the results, but that these results are not sufficiently strong to warrant acceptance. I'm wary to turn down a paper over what are basically negative results, but for results like this to be useful to the community, they'd have to come from a very thorough experiment, and they'd have to be accompanied by a frank and detailed discussion. Neither of the two more confident authors are convinced that this paper meets that bar."""
921,"""Tree-Structured Recurrent Switching Linear Dynamical Systems for Multi-Scale Modeling""","['machine learning', 'bayesian statistics', 'dynamical systems']","""Many real-world systems studied are governed by complex, nonlinear dynamics. By modeling these dynamics, we can gain insight into how these systems work, make predictions about how they will behave, and develop strategies for controlling them. While there are many methods for modeling nonlinear dynamical systems, existing techniques face a trade off between offering interpretable descriptions and making accurate predictions. Here, we develop a class of models that aims to achieve both simultaneously, smoothly interpolating between simple descriptions and more complex, yet also more accurate models. Our probabilistic model achieves this multi-scale property through of a hierarchy of locally linear dynamics that jointly approximate global nonlinear dynamics. We call it the tree-structured recurrent switching linear dynamical system. To fit this model, we present a fully-Bayesian sampling procedure using Polya-Gamma data augmentation to allow for fast and conjugate Gibbs sampling. Through a variety of synthetic and real examples, we show how these models outperform existing methods in both interpretability and predictive capability.""","""This paper presents a recurrent tree-structured linear dynamical system to model the dynamics of a complex nonlinear dynamical system. All reviewers agree that the paper is interesting and useful, and is likely to have an impact in the community. Some of the doubts that reviewers had were resolved after the rebuttal period. Overall, this is a good paper, and I recommend an acceptance."""
922,"""Learning Exploration Policies for Navigation""","['Exploration', 'navigation', 'reinforcement learning']","""Numerous past works have tackled the problem of task-driven navigation. But, how to effectively explore a new environment to enable a variety of down-stream tasks has received much less attention. In this work, we study how agents can autonomously explore realistic and complex 3D environments without the context of task-rewards. We propose a learning-based approach and investigate different policy architectures, reward functions, and training paradigms. We find that use of policies with spatial memory that are bootstrapped with imitation learning and finally finetuned with coverage rewards derived purely from on-board sensors can be effective at exploring novel environments. We show that our learned exploration policies can explore better than classical approaches based on geometry alone and generic learning-based exploration techniques. Finally, we also show how such task-agnostic exploration can be used for down-stream tasks. Videos are available at pseudo-url.""","""The authors have proposed an approach for directly learning a spatial exploration policy which is effective in unseen environments. Rather than use external task rewards, the proposed approach uses an internally computed coverage reward derived from on-board sensors. The authors use imitation learning to bootstrap the training and then fine-tune using the intrinsic coverage reward. Multiple experiments and ablations are given to support and understand the approach. The paper is well-written and interesting. The experiments are appropriate, although further evaluations in real-world settings really ought to be done to fully explore the significance of the approach. The reviewers were divided, with one reviewer finding fault with the paper in terms of the claims made, the positioning against prior art, and the chosen baselines. The other two reviewers supported publication even after considering the opposition of R1, noting that they believe that the baselines are sufficient, and the contribution is novel. After reviewing the long exchange and discussion, the AC sides with accepting the paper. Although R1 raises some valid concerns, the authors defend themselves convincingly and the arguments do not, in any case, detract substantially from what is a solid submission."""
923,"""A Variational Dirichlet Framework for Out-of-Distribution Detection""","['out-of-distribution detection', 'variational inference', 'Dirichlet distribution', 'deep learning', 'uncertainty measure']","""With the recently rapid development in deep learning, deep neural networks have been widely adopted in many real-life applications. However, deep neural networks are also known to have very little control over its uncertainty for test examples, which potentially causes very harmful and annoying consequences in practical scenarios. In this paper, we are particularly interested in designing a higher-order uncertainty metric for deep neural networks and investigate its performance on the out-of-distribution detection task proposed by~\cite{hendrycks2016baseline}. Our method first assumes there exists a underlying higher-order distribution pseudo-formula , which generated label-wise distribution pseudo-formula over classes on the K-dimension simplex, and then approximate such higher-order distribution via parameterized posterior function pseudo-formula under variational inference framework, finally we use the entropy of learned posterior distribution pseudo-formula as uncertainty measure to detect out-of-distribution examples. However, we identify the overwhelming over-concentration issue in such a framework, which greatly hinders the detection performance. Therefore, we further design a log-smoothing function to alleviate such issue to greatly increase the robustness of the proposed entropy-based uncertainty measure. Through comprehensive experiments on various datasets and architectures, our proposed variational Dirichlet framework with entropy-based uncertainty measure is consistently observed to yield significant improvements over many baseline systems.""","""The paper proposes a new framework for out-of-distribution detection, based on variational inference and a prior Dirichlet distribution. The reviewers and AC note the following potential weaknesses: (1) arguable and not well justified choices of parameters and (2) the performance degradation under many classes (e.g., CIFAR-100). For (2), the authors mentioned that this is because ""there are more than 20% of misclassified test images"". But, AC rather views it as a limitation of the proposed approach. The out-of-detection detection problem is a one or two classification task, independent of how many classes exist in the neural classifier. In overall, the proposed idea is interesting and makes sense but AC decided that the authors need more significant works to publish the work."""
924,"""Overcoming catastrophic forgetting through weight consolidation and long-term memory""","['Catastrophic Forgetting', 'Life-Long Learning', 'adversarial examples']","""Sequential learning of multiple tasks in artificial neural networks using gradient descent leads to catastrophic forgetting, whereby previously learned knowledge is erased during learning of new, disjoint knowledge. Here, we propose a new approach to sequential learning which leverages the recent discovery of adversarial examples. We use adversarial subspaces from previous tasks to enable learning of new tasks with less interference. We apply our method to sequentially learning to classify digits 0, 1, 2 (task 1), 4, 5, 6, (task 2), and 7, 8, 9 (task 3) in MNIST (disjoint MNIST task). We compare and combine our Adversarial Direction (AD) method with the recently proposed Elastic Weight Consolidation (EWC) method for sequential learning. We train each task for 20 epochs, which yields good initial performance (99.24% correct task 1 performance). After training task 2, and then task 3, both plain gradient descent (PGD) and EWC largely forget task 1 (task 1 accuracy 32.95% for PGD and 41.02% for EWC), while our combined approach (AD+EWC) still achieves 94.53% correct on task 1. We obtain similar results with a much more difficult disjoint CIFAR10 task (70.10% initial task 1 performance, 67.73% after learning tasks 2 and 3 for AD+EWC, while PGD and EWC both fall to chance level). We confirm qualitatively similar results for EMNIST with 5 tasks and under 3 variants of our approach. Our results suggest that AD+EWC can provide better sequential learning performance than either PGD or EWC.""","""The authors propose an approach for continual learning of a sequence of tasks which augments the network with task-specific neurons which encode 'adversarial subspaces' and prevent interference and forgetting when new tasks are being learnt. The approach is novel and seems to work relatively well on a simple sequence of MNIST or CIFAR10 classes, and has certain advantages, such as not requiring any stored data. However, the reviewers agreed that the presentation of the method is quite confusing and that the paper does not provide adequate intuition, visualisation, or explanation of the claim that they are preventing forgetting through the intersection of adversarial subspaces. Moreover, there was a concern that the baselines were not strong enough to validate the approach."""
925,"""Generative Question Answering: Learning to Answer the Whole Question""","['Question answering', 'question generation', 'reasoning', 'squad', 'clevr']","""Discriminative question answering models can overfit to superficial biases in datasets, because their loss function saturates when any clue makes the answer likely. We introduce generative models of the joint distribution of questions and answers, which are trained to explain the whole question, not just to answer it.Our question answering (QA) model is implemented by learning a prior over answers, and a conditional language model to generate the question given the answerallowing scalable and interpretable many-hop reasoning as the question is generated word-by-word. Our model achieves competitive performance with specialised discriminative models on the SQUAD and CLEVR benchmarks, indicating that it is a more general architecture for language understanding and reasoning than previous work. The model greatly improves generalisation both from biased training data and to adversarial testing data, achieving a new state-of-the-art on ADVERSARIAL SQUAD. We will release our code.""","""All reviewers recommend accept. Discussion can be consulted below."""
926,"""Excessive Invariance Causes Adversarial Vulnerability""","['Generalization', 'Adversarial Examples', 'Invariance', 'Information Theory', 'Invertible Networks']","""Despite their impressive performance, deep neural networks exhibit striking failures on out-of-distribution inputs. One core idea of adversarial example research is to reveal neural network errors under such distribution shifts. We decompose these errors into two complementary sources: sensitivity and invariance. We show deep networks are not only too sensitive to task-irrelevant changes of their input, as is well-known from epsilon-adversarial examples, but are also too invariant to a wide range of task-relevant changes, thus making vast regions in input space vulnerable to adversarial attacks. We show such excessive invariance occurs across various tasks and architecture types. On MNIST and ImageNet one can manipulate the class-specific content of almost any image without changing the hidden activations. We identify an insufficiency of the standard cross-entropy loss as a reason for these failures. Further, we extend this objective based on an information-theoretic analysis so it encourages the model to consider all task-dependent features in its decision. This provides the first approach tailored explicitly to overcome excessive invariance and resulting vulnerabilities.""","""This paper studies the roots of the existence of adversarial perspective from a new perspective. This perspective is quite interesting and thought-provoking. However, some of the contributions rely on fairly restrictive assumptions and/or are not properly evaluated. Still, overall, this paper should be a valuable addition to the program. """
927,"""Predict then Propagate: Graph Neural Networks meet Personalized PageRank""","['Graph', 'GCN', 'GNN', 'Neural network', 'Graph neural network', 'Message passing neural network', 'Semi-supervised classification', 'Semi-supervised learning', 'PageRank', 'Personalized PageRank']","""Neural message passing algorithms for semi-supervised classification on graphs have recently achieved great success. However, for classifying a node these methods only consider nodes that are a few propagation steps away and the size of this utilized neighborhood is hard to extend. In this paper, we use the relationship between graph convolutional networks (GCN) and PageRank to derive an improved propagation scheme based on personalized PageRank. We utilize this propagation procedure to construct a simple model, personalized propagation of neural predictions (PPNP), and its fast approximation, APPNP. Our model's training time is on par or faster and its number of parameters on par or lower than previous models. It leverages a large, adjustable neighborhood for classification and can be easily combined with any neural network. We show that this model outperforms several recently proposed methods for semi-supervised classification in the most thorough study done so far for GCN-like models. Our implementation is available online.""","""There were several ambivalent reviews for this submission and one favorable one. Although this is a difficult case, I am recommending accepting the paper. There were two main questions in my mind. 1. Did the authors justify that the limited neighborhood problem they try to fix with their method is a real problem and that they fixed it? If so, accept. Here I believe evidence has been presented, but the case remains undecided. 2. If they have not, is the method/experiments sufficiently useful to be interesting anyway? This question I would lean towards answering in the affirmative. I believe the paper as a whole is sufficiently interesting and executed sufficiently well to be accepted, although I was not convinced of the first point (1) above. One review voting to reject did not find the conceptual contribution very valuable but still thought the paper was not severely flawed. I am partly down-weighting the conceptual criticism they made. I am more concerned with experimental issues. However, I did not see sufficiently severe issues raised by the reviewers to justify rejection. Ultimately, I could go either way on this case, but I think some members of the community will benefit from reading this work enough that it should be accepted."""
928,"""Stochastic Gradient Push for Distributed Deep Learning""","['optimization', 'distributed', 'large scale', 'deep learning']","""Large mini-batch parallel SGD is commonly used for distributed training of deep networks. Approaches that use tightly-coupled exact distributed averaging based on AllReduce are sensitive to slow nodes and high-latency communication. In this work we show the applicability of Stochastic Gradient Push (SGP) for distributed training. SGP uses a gossip algorithm called PushSum for approximate distributed averaging, allowing for much more loosely coupled communications which can be beneficial in high-latency or high-variability scenarios. The tradeoff is that approximate distributed averaging injects additional noise in the gradient which can affect the train and test accuracies. We prove that SGP converges to a stationary point of smooth, non-convex objective functions. Furthermore, we validate empirically the potential of SGP. For example, using 32 nodes with 8 GPUs per node to train ResNet-50 on ImageNet, where nodes communicate over 10Gbps Ethernet, SGP completes 90 epochs in around 1.5 hours while AllReduce SGD takes over 5 hours, and the top-1 validation accuracy of SGP remains within 1.2% of that obtained using AllReduce SGD.""","""The reviewers liked the paper in general but the empirical evaluation lacks studies on a wider range of different data sets."""
929,"""Neural Causal Discovery with Learnable Input Noise""","['neural causal learning', 'learnable noise']","""Learning causal relations from observational time series with nonlinear interactions and complex causal structures is a key component of human intelligence, and has a wide range of applications. Although neural nets have demonstrated their effectiveness in a variety of fields, their application in learning causal relations has been scarce. This is due to both a lack of theoretical results connecting risk minimization and causality (enabling function approximators like neural nets to apply), and a lack of scalability in prior causal measures to allow for expressive function approximators like neural nets to apply. In this work, we propose a novel causal measure and algorithm using risk minimization to infer causal relations from time series. We demonstrate the effectiveness and scalability of our algorithms to learn nonlinear causal models in synthetic datasets as comparing to other methods, and its effectiveness in inferring causal relations in a video game environment and real-world heart-rate vs. breath-rate and rat brain EEG datasets.""","""Granger Causality is a beautiful operational definition of causality, that reduces causal modeling to the past-to-future predictive strength. The combination of classical granger causality with deep learning is very well motivated as a research problem. As such the continuation of the effort in this paper is strongly encouraged. However, the review process did uncover possible flaws in some of the main, original results of this paper. The reviewers also expressed concerns that the experiments were unconvincing due to very small data sizes. The paper will benefit from a revision and resubmission to another venue, and is not ready for acceptance at ICLR-2019."""
930,"""Padam: Closing the Generalization Gap of Adaptive Gradient Methods in Training Deep Neural Networks""",[],"""Adaptive gradient methods, which adopt historical gradient information to automatically adjust the learning rate, despite the nice property of fast convergence, have been observed to generalize worse than stochastic gradient descent (SGD) with momentum in training deep neural networks. This leaves how to close the generalization gap of adaptive gradient methods an open problem. In this work, we show that adaptive gradient methods such as Adam, Amsgrad, are sometimes ""over adapted"". We design a new algorithm, called Partially adaptive momentum estimation method (Padam), which unifies the Adam/Amsgrad with SGD by introducing a partial adaptive parameter p, to achieve the best from both worlds. Experiments on standard benchmarks show that Padam can maintain fast convergence rate as Adam/Amsgrad while generalizing as well as SGD in training deep neural networks. These results would suggest practitioners pick up adaptive gradient methods once again for faster training of deep neural networks.""","""This paper proposes a simple modification of the Adam optimizer, introducing a hyper-parameter 'p' (with value in the range [0,1/2]) parameterizing the parameter update: theta_new = theta_old + m/v^p where p=1/2 falls back to the standard Adam/Amsgrad optimizer, and p=0 falls back to a variant of SGD with momentum. The authors motivate the method by pointing out that: - Through the value of 'p', one can interpolate between SGD with momentum and Adam/Amsgrad. By choosing a value of 'p' smaller than 0.5, one can therefore use perform optimization that is 'partially adaptive'. - The method shows good empirical performance. The paper contains an inaccuracy, which we hope will be solved before the final version. The authors argue that the 1/sqrt(v) term in Adam results in a lower learning rate, and the authors argue that the effective learning rate ""easily explodes"" (section 3) because of this term, and that a ""more aggressive"" learning rate is more appropriate. This last point is false; the value of 1/sqrt(v) can be smaller or larger than 1 depending on the value of 'v', and that a decrease in value of 'p' can result in either an increase or decrease in effective learning rate, depending on the value of v. The value of 'v' is a function of the scale of loss function, which can really be arbitrary. (In case of very high-dimensional predictions, for example, the scale of the loss function is often proportional with the dimensionality of variable to be modeled, which can be arbitrarily large, e.g. in image or video modeling the loss function tends to be of a much larger scale than with classification.) The authors promise to include a comparison to AdamW [Loshchilov, 2017] that includes tuning of the weight decay parameter. The lack of this experiments makes it more difficult to make a conclusion regarding the performance relative to AdamW. However, the methods offer potentially orthogonal (and combinable) advantages. [Loshchilov, 2017] pseudo-url """
931,"""Analyzing Inverse Problems with Invertible Neural Networks""","['Inverse problems', 'Neural Networks', 'Uncertainty', 'Invertible Neural Networks']","""For many applications, in particular in natural science, the task is to determine hidden system parameters from a set of measurements. Often, the forward process from parameter- to measurement-space is well-defined, whereas the inverse problem is ambiguous: multiple parameter sets can result in the same measurement. To fully characterize this ambiguity, the full posterior parameter distribution, conditioned on an observed measurement, has to be determined. We argue that a particular class of neural networks is well suited for this task so-called Invertible Neural Networks (INNs). Unlike classical neural networks, which attempt to solve the ambiguous inverse problem directly, INNs focus on learning the forward process, using additional latent output variables to capture the information otherwise lost. Due to invertibility, a model of the corresponding inverse process is learned implicitly. Given a specific measurement and the distribution of the latent variables, the inverse pass of the INN provides the full posterior over parameter space. We prove theoretically and verify experimentally, on artificial data and real-world problems from medicine and astrophysics, that INNs are a powerful analysis tool to find multi-modalities in parameter space, uncover parameter correlations, and identify unrecoverable parameters.""","""This paper proposes a framework for using invertible neural networks to study inverse problems, e.g., recover hidden states or parameters of a system from measurements. This is an important and well-motivated topic, and the solution proposed is novel although somewhat incremental. The paper is generally well written. Some theoretical analysis is provided, giving conditions under which the proposed approach recovers the true posterior. Empirically, the approach is tested on synthetic data and real world problems from medicine and astronomy, where it is shown to compared favorably to ABC and conditional VAEs. Adding additional baselines (Bayesian MCMC and Stein methods) would be good. There are some potential issues regarding MMD scalability to high dimensional spaces, but overall the paper makes a solid contribution and all the reviewers agree it should be accepted for publication."""
932,"""Learning to Coordinate Multiple Reinforcement Learning Agents for Diverse Query Reformulation""","['Reinforcement Learning', 'Multi-agent', 'Information Retrieval', 'Question-Answering', 'Query Reformulation', 'Query Expansion']","""We propose a method to efficiently learn diverse strategies in reinforcement learning for query reformulation in the tasks of document retrieval and question answering. In the proposed framework an agent consists of multiple specialized sub-agents and a meta-agent that learns to aggregate the answers from sub-agents to produce a final answer. Sub-agents are trained on disjoint partitions of the training data, while the meta-agent is trained on the full training set. Our method makes learning faster, because it is highly parallelizable, and has better generalization performance than strong baselines, such as an ensemble of agents trained on the full data. We show that the improved performance is due to the increased diversity of reformulation strategies. """,""" pros: - The paper is clear and easy to read - Both Reviewer 1 and Reviewer 2 found the empirical evaluation to be good cons: - Some of the reviewers felt that the proposed approach lacked novelty (e.g. with respect to Nogueira and Cho) - Some of the architecture choices seem complicated and it was not fully clear to the reviewers (even after the rebuttal) how and why things were working better in this approach than in other similar ones. I think this is a good paper but it doesn't quite meet the bar for acceptance at this time. """
933,"""Learning a SAT Solver from Single-Bit Supervision""","['sat', 'search', 'graph neural network', 'theorem proving', 'proof']","""We present NeuroSAT, a message passing neural network that learns to solve SAT problems after only being trained as a classifier to predict satisfiability. Although it is not competitive with state-of-the-art SAT solvers, NeuroSAT can solve problems that are substantially larger and more difficult than it ever saw during training by simply running for more iterations. Moreover, NeuroSAT generalizes to novel distributions; after training only on random SAT problems, at test time it can solve SAT problems encoding graph coloring, clique detection, dominating set, and vertex cover problems, all on a range of distributions over small random graphs.""","""The submission proposes a machine learning approach to directly train a prediction system for whether a boolean sentence is satisfiable. The strengths of the paper seem to be largely in proposing an architecture for SAT problems and the analysis of the generalization performance of the resulting classifier on classes of problems not directly seen during training. Although the resulting system cannot be claimed to be a state of the art system, and it does not have a correctness guarantee like DPLL based approaches, the paper is a nice re-introduction of SAT in a machine learning context using deep networks. It may be nice to mention e.g. (W. Ruml. Adaptive Tree Search. PhD thesis, Harvard University, 2002) which applied reinforcement learning techniques to SAT problems. The empirical validation on variable sized problems, etc. is a nice contribution showing interesting generalization properties of the proposed approach. The reviewers were unanimous in their recommendation that the paper be accepted, and the review process attracted a number of additional comments showing the broader interest of the setting."""
934,"""Why do deep convolutional networks generalize so poorly to small image transformations?""","['Convolutional neural networks', 'The sampling theorem', 'Sensitivity to small image transformations', 'Dataset bias', 'Shiftability']","""Deep convolutional network architectures are often assumed to guarantee generalization for small image translations and deformations. In this paper we show that modern CNNs (VGG16, ResNet50, and InceptionResNetV2) can drastically change their output when an image is translated in the image plane by a few pixels, and that this failure of generalization also happens with other realistic small image transformations. Furthermore, we see these failures to generalize more frequently in more modern networks. We show that these failures are related to the fact that the architecture of modern CNNs ignores the classical sampling theorem so that generalization is not guaranteed. We also show that biases in the statistics of commonly used image datasets makes it unlikely that CNNs will learn to be invariant to these transformations. Taken together our results suggest that the performance of CNNs in object recognition falls far short of the generalization capabilities of humans.""","""This paper attempts to answer its suggestive title by arguing that this generic lack of invariance in large CNN architectures is due to aliasing introduced during the downsampling stages. This paper received mixed reviews. Positive aspects include the clarity and exhaustive empirical setups, whereas negative aspects focused on the lack of substance behind some of the claims. Ultimately, the AC took these considerations into account and made his/her own assessment, summarized here. The main claim of this paper implies the following: modern CNNs are unable to build invariance to small shifts, but somehow are able to learn far more complex invariances involving lighting, pose, texture, etc. This must be empirically verified beyond reasonable doubt, and the AC thinks that the current experimental setup does not achieve this threshold. As mentioned by reviewers and by public comments, the preprocessing pipeline is a key factor that may be confounding the analysis, and this should be better analysed. For example, as mentioned in the reviews below, the shift in the image can be either done by inpainting, cropping, or using a fixed background. The authors claim that there are no qualitative differences between those preprocessing choices, but by inspecting Figures 2B and 8C, the AC notices a severe change in 'jaggedness'; in other words, the choice of preprocessing *does* affect the quantitative measures of (un)stability, even though the qualitative assessment (unstable in all setups) is the same. In particular, using non-centered crops should be the default setup, since it requires no preprocessing. It is confusing that it appears in the appendix instead of the inpainting version of figure 2b. This is important, since it implies that the analysis is mixing two perturbations: the actual action of the translation group and the choice of preprocessing, and that the latter is by no means negligible. I would suggest the authors to perform the following experiment to disentangle the effect of translation by the effect of preprocessing. Since the translation forms a group, for any shift applied to the image, one can 'undo' it by applying the inverse shift. Say one applies a shift to image x of d pixels and obtains x'=T(x,+d) as a result (by using whatever border handling procedure). If border effects were negligible, then x''=T(x',-d) should give us back x, so a good measure of how unstable the network is is to measure the difference in prediction between x,x' and x''. If predicting x'' is as unstable as predicting x', it follows that the network is actually unstable to the border effect introduced by T. Given this, the AC recommends rejection at this time, and encourages the authors to resubmit their work by addressing the above point. """
935,"""SEGEN: SAMPLE-ENSEMBLE GENETIC EVOLUTIONARY NETWORK MODEL""","['Genetic Evolutionary Network', 'Deep Learning', 'Genetic Algorithm', 'Ensemble Learning', 'Representation Learning']","""Deep learning, a rebranding of deep neural network research works, has achieved a remarkable success in recent years. With multiple hidden layers, deep learning models aim at computing the hierarchical feature representations of the observational data. Meanwhile, due to its severe disadvantages in data consumption, computational resources, parameter tuning costs and the lack of result explainability, deep learning has also suffered from lots of criticism. In this paper, we will introduce a new representation learning model, namely Sample-Ensemble Genetic Evolutionary Network (SEGEN), which can serve as an alternative approach to deep learning models. Instead of building one single deep model, based on a set of sampled sub-instances, SEGEN adopts a genetic-evolutionary learning strategy to build a group of unit models generations by generations. The unit models incorporated in SEGEN can be either traditional machine learning models or the recent deep learning models with a much narrower and shallower architecture. The learning results of each instance at the final generation will be effectively combined from each unit model via diffusive propagation and ensemble learning strategies. From the computational perspective, SEGEN requires far less data, fewer computational resources and parameter tuning efforts, but has sound theoretic interpretability of the learning process and results. Extensive experiments have been done on several different real-world benchmark datasets, and the experimental results obtained by SEGEN have demonstrated its advantages over the state-of-the-art representation learning models.""","""This paper endeavors to combine genetic evolutionary algorithms with subsampling techniques. As noted by reviewers, this is an interesting topic and the paper is intriguing, but more work is required to make it convincing (fairer baselines, more detailed / clearer presentation, ablation studies to justify the claims made int he paper). Authors are encouraged to strengthen the paper by following reviewers' suggestions."""
936,"""Mol-CycleGAN - a generative model for molecular optimization""","['generative adversarial networks', 'drug design', 'deep learning', 'molecule optimization']","""Designing a molecule with desired properties is one of the biggest challenges in drug development, as it requires optimization of chemical compound structures with respect to many complex properties. To augment the compound design process we introduce Mol-CycleGAN -- a CycleGAN-based model that generates optimized compounds with a chemical scaffold of interest. Namely, given a molecule our model generates a structurally similar one with an optimized value of the considered property. We evaluate the performance of the model on selected optimization objectives related to structural properties (presence of halogen groups, number of aromatic rings) and to a physicochemical property (penalized logP). In the task of optimization of penalized logP of drug-like molecules our model significantly outperforms previous results. ""","""This paper introduces a variant of the CycleGAN designed to optimize molecular graphs to achieve a desired quality. The work is reasonably clear and sensible, however it's of limited technical novelty, since it's mainly just combining two existing techniques. Overall its specificity and incrementalness make it not meet the bar."""
937,"""Understanding Opportunities for Efficiency in Single-image Super Resolution Networks""","['Super-Resolution', 'Resource-Efficiency']","""A successful application of convolutional architectures is to increase the resolution of single low-resolution images -- a image restoration task called super-resolution (SR). Naturally, SR is of value to resource constrained devices like mobile phones, electronic photograph frames and televisions to enhance image quality. However, SR demands perhaps the most extreme amounts of memory and compute operations of any mainstream vision task known today, preventing SR from being deployed to devices that require them. In this paper, we perform a early systematic study of system resource efficiency for SR, within the context of a variety of architectural and low-precision approaches originally developed for discriminative neural networks. We present a rich set of insights, representative SR architectures, and efficiency trade-offs; for example, the prioritization of ways to compress models to reach a specific memory and computation target and techniques to compact SR models so that they are suitable for DSPs and FPGAs. As a result of doing so, we manage to achieve better and comparable performance with previous models in the existing literature, highlighting the practicality of using existing efficiency techniques in SR tasks. Collectively, we believe these results provides the foundation for further research into the little explored area of resource efficiency for SR. ""","""This paper targets improving the computation efficiency of super resolution task. Reviewers have a consensus that this paper lacks technical contribution, therefore not recommend acceptance. """
938,"""Diversity and Depth in Per-Example Routing Models""","['conditional computation', 'routing models', 'depth']","""Routing models, a form of conditional computation where examples are routed through a subset of components in a larger network, have shown promising results in recent works. Surprisingly, routing models to date have lacked important properties, such as architectural diversity and large numbers of routing decisions. Both architectural diversity and routing depth can increase the representational power of a routing network. In this work, we address both of these deficiencies. We discuss the significance of architectural diversity in routing models, and explain the tradeoffs between capacity and optimization when increasing routing depth. In our experiments, we find that adding architectural diversity to routing models significantly improves performance, cutting the error rates of a strong baseline by 35% on an Omniglot setup. However, when scaling up routing depth, we find that modern routing techniques struggle with optimization. We conclude by discussing both the positive and negative results, and suggest directions for future research.""",""" pros: - good, clear writing - interesting analysis - very important research area - nice results on multi-task omniglot cons: - somewhat limited experimental evaluation The reviewers I think all agree that the work is interesting and the paper well-written. I think there is still a need for more thorough experiments (which it sounds like the authors are undertaking). I recommend acceptance. """
939,"""DEEP GRAPH TRANSLATION""",[],"""The tremendous success of deep generative models on generating continuous data like image and audio has been achieved; however, few deep graph generative models have been proposed to generate discrete data such as graphs. The recently proposed approaches are typically unconditioned generative models which have no control over modes of the graphs being generated. Differently, in this paper, we are interested in a new problem named Deep Graph Translation: given an input graph, the goal is to infer a target graph by learning their underlying translation mapping. Graph translation could be highly desirable in many applications such as disaster management and rare event forecasting, where the rare and abnormal graph patterns (e.g., traffic congestions and terrorism events) will be inferred prior to their occurrence even without historical data on the abnormal patterns for this specific graph (e.g., a road network or human contact network). To this end, we propose a novel Graph-Translation-Generative Adversarial Networks (GT-GAN) which translates one mode of the input graphs to its target mode. GT-GAN consists of a graph translator where we propose new graph convolution and deconvolution layers to learn the global and local translation mapping. A new conditional graph discriminator has also been proposed to classify target graphs by conditioning on input graphs. Extensive experiments on multiple synthetic and real-world datasets demonstrate the effectiveness and scalability of the proposed GT-GAN.""","""Although one reviewer recommended accepting this paper, they were not willing to champion it during the discussion phase and did not seem to truly believe it is currently ready for publication. Thus I am recommending rejecting this submission."""
940,"""A Convergence Analysis of Gradient Descent for Deep Linear Neural Networks""","['Deep Learning', 'Learning Theory', 'Non-Convex Optimization']","""We analyze speed of convergence to global optimum for gradient descent training a deep linear neural network by minimizing the L2 loss over whitened data. Convergence at a linear rate is guaranteed when the following hold: (i) dimensions of hidden layers are at least the minimum of the input and output dimensions; (ii) weight matrices at initialization are approximately balanced; and (iii) the initial loss is smaller than the loss of any rank-deficient solution. The assumptions on initialization (conditions (ii) and (iii)) are necessary, in the sense that violating any one of them may lead to convergence failure. Moreover, in the important case of output dimension 1, i.e. scalar regression, they are met, and thus convergence to global optimum holds, with constant probability under a random initialization scheme. Our results significantly extend previous analyses, e.g., of deep linear residual networks (Bartlett et al., 2018).""","""This is a well written paper that contributes a clear advance to the understanding of how gradient descent behaves when training deep linear models. Reviewers were unanimously supportive."""
941,"""Identifying Generalization Properties in Neural Networks""","['generalization', 'PAC-Bayes', 'Hessian', 'perturbation']","""While it has not yet been proven, empirical evidence suggests that model generalization is related to local properties of the optima which can be described via the Hessian. We connect model generalization with the local property of a solution under the PAC-Bayes paradigm. In particular, we prove that model generalization ability is related to the Hessian, the higher-order ""smoothness"" terms characterized by the Lipschitz constant of the Hessian, and the scales of the parameters. Guided by the proof, we propose a metric to score the generalization capability of the model, as well as an algorithm that optimizes the perturbed model accordingly. ""","""This paper proposes a generalization metric depending on the Lipschitz of the Hessian. Pros: Paper has some nice experiments correlating their Hessian based generalization metric with the generalization gap, Cons: The paper does not compare its results with existing generalization bounds, as there is substantial work in the area now. It is not clear whether existing generalization bounds do not capture this phenomenon with different batch sizes/learning rates, and the necessity of having and explicit dependence on the Lipschitz of the Hessian. The bound by itself is also weak because of its dependence on number of parameters 'm'. The paper is poorly written and all reviewers complain about its readability. I suggest authors to address concerns of the reviewers before submitting again. """
942,"""Generative Ensembles for Robust Anomaly Detection""","['Anomaly Detection', 'Uncertainty', 'Out-of-Distribution', 'Generative Models']","""Deep generative models are capable of learning probability distributions over large, high-dimensional datasets such as images, video and natural language. Generative models trained on samples from p(x) ought to assign low likelihoods to out-of-distribution (OoD) samples from q(x), making them suitable for anomaly detection applications. We show that in practice, likelihood models are themselves susceptible to OoD errors, and even assign large likelihoods to images from other natural datasets. To mitigate these issues, we propose Generative Ensembles, a model-independent technique for OoD detection that combines density-based anomaly detection with uncertainty estimation. Our method outperforms ODIN and VIB baselines on image datasets, and achieves comparable performance to a classification model on the Kaggle Credit Fraud dataset.""","""This paper suggests the use of generative ensembles for detecting out-of-distribution samples. The reviewers found the paper easy to read, especially after the changes made during the rebuttal. However, further elaboration in the technical descriptions (and assumptions made) could make the work seem more mature, as R2 and R1 point out. The general feeling by reading the reviews and discussions is that this is promising work that, nevertheless, needs some more novel elements. A possible avenue for increasing the contribution of the paper is to follow R1s advice to extract more convincing insights from the results. """
943,"""Pay Less Attention with Lightweight and Dynamic Convolutions""","['Deep learning', 'sequence to sequence learning', 'convolutional neural networks', 'generative models']","""Self-attention is a useful mechanism to build generative models for language and images. It determines the importance of context elements by comparing each element to the current time step. In this paper, we show that a very lightweight convolution can perform competitively to the best reported self-attention results. Next, we introduce dynamic convolutions which are simpler and more efficient than self-attention. We predict separate convolution kernels based solely on the current time-step in order to determine the importance of context elements. The number of operations required by this approach scales linearly in the input length, whereas self-attention is quadratic. Experiments on large-scale machine translation, language modeling and abstractive summarization show that dynamic convolutions improve over strong self-attention models. On the WMT'14 English-German test set dynamic convolutions achieve a new state of the art of 29.7 BLEU.""","""Very solid work, recognized by all reviewers as worthy of acceptance. Additional readers also commented and there is interest in the open source implementation that the authors promise to provide."""
944,"""On Self Modulation for Generative Adversarial Networks""","['unsupervised learning', 'generative adversarial networks', 'deep generative modelling']","""Training Generative Adversarial Networks (GANs) is notoriously challenging. We propose and study an architectural modification, self-modulation, which improves GAN performance across different data sets, architectures, losses, regularizers, and hyperparameter settings. Intuitively, self-modulation allows the intermediate feature maps of a generator to change as a function of the input noise vector. While reminiscent of other conditioning techniques, it requires no labeled data. In a large-scale empirical study we observe a relative decrease of 5%-35% in FID. Furthermore, all else being equal, adding this modification to the generator leads to improved performance in 124/144 (86%) of the studied settings. Self-modulation is a simple architectural change that requires no additional parameter tuning, which suggests that it can be applied readily to any GAN.""","""This manuscript proposes an architectural improvement for generative adversarial network that allows the intermediate layers of a generator to be modulated by the input noise vector using conditional batch normalization. The reviewers find the paper simple and well-supported by extensive experimental results. There were some concerns about the impact of such an empirical study. However, the strength and simplicity of the technique means that the method could be of practical interest to the ICLR community."""
945,"""Task-GAN for Improved GAN based Image Restoration""",['Task-GAN: Improving Generative Adversarial Network for Image Restoration'],"""Deep Learning (DL) algorithms based on Generative Adversarial Network (GAN) have demonstrated great potentials in computer vision tasks such as image restoration. Despite the rapid development of image restoration algorithms using DL and GANs, image restoration for specific scenarios, such as medical image enhancement and super-resolved identity recognition, are still facing challenges. How to ensure visually realistic restoration while avoiding hallucination or mode- collapse? How to make sure the visually plausible results do not contain hallucinated features jeopardizing downstream tasks such as pathology identification and subject identification? Here we propose to resolve these challenges by coupling the GAN based image restoration framework with another task-specific network. With medical imaging restoration as an example, the proposed model conducts additional pathology recognition/classification task to ensure the preservation of detailed structures that are important to this task. Validated on multiple medical datasets, we demonstrate the proposed method leads to improved deep learning based image restoration while preserving the detailed structure and diagnostic features. Additionally, the trained task network show potentials to achieve super-human level performance in identifying pathology and diagnosis. Further validation on super-resolved identity recognition tasks also show that the proposed method can be generalized for diverse image restoration tasks.""","""This work presents a reconstruction GAN with an additional classification task in the objective loss function. Evaluations are carried out on medical and non-medical datasets. Reviewers raise multiple concerns around the following: - Novelty (all reviewers) - Inadequate comparison baselines (all reviewers) - Inadequate citations. (R2 & R3) Authors have not offered a rebuttal. Recommendation is reject. Work may be more suitable as an application paper for a medical conference or journal. """
946,"""Negotiating Team Formation Using Deep Reinforcement Learning""","['Reinforcement Learning', 'Negotiation', 'Team Formation', 'Cooperative Game Theory', 'Shapley Value']","""When autonomous agents interact in the same environment, they must often cooperate to achieve their goals. One way for agents to cooperate effectively is to form a team, make a binding agreement on a joint plan, and execute it. However, when agents are self-interested, the gains from team formation must be allocated appropriately to incentivize agreement. Various approaches for multi-agent negotiation have been proposed, but typically only work for particular negotiation protocols. More general methods usually require human input or domain-specific data, and so do not scale. To address this, we propose a framework for training agents to negotiate and form teams using deep reinforcement learning. Importantly, our method makes no assumptions about the specific negotiation protocol, and is instead completely experience driven. We evaluate our approach on both non-spatial and spatially extended team-formation negotiation environments, demonstrating that our agents beat hand-crafted bots and reach negotiation outcomes consistent with fair solutions predicted by cooperative game theory. Additionally, we investigate how the physical location of agents influences negotiation outcomes.""","""This paper was reviewed by three experts. Initially, the reviews were mixed with several concerns raised. After the author response, there continue to be concerns about need for significantly more experiments. If this were a journal, it is clear that recommendation would be ""major revision"". Since that option is not available and the paper clearly needs another round of reviews, we must unfortunately reject. We encourage the authors to incorporate reviewer feedback and submit a stronger manuscript at a future venue. """
947,"""AutoLoss: Learning Discrete Schedule for Alternate Optimization""","['Meta Learning', 'AutoML', 'Optimization Schedule']","""Many machine learning problems involve iteratively and alternately optimizing different task objectives with respect to different sets of parameters. Appropriately scheduling the optimization of a task objective or a set of parameters is usually crucial to the quality of convergence. In this paper, we present AutoLoss, a meta-learning framework that automatically learns and determines the optimization schedule. AutoLoss provides a generic way to represent and learn the discrete optimization schedule from metadata, allows for a dynamic and data-driven schedule in ML problems that involve alternating updates of different parameters or from different loss objectives. We apply AutoLoss on four ML tasks: d-ary quadratic regression, classification using a multi-layer perceptron (MLP), image generation using GANs, and multi-task neural machine translation (NMT). We show that the AutoLoss controller is able to capture the distribution of better optimization schedules that result in higher quality of convergence on all four tasks. The trained AutoLoss controller is generalizable -- it can guide and improve the learning of a new task model with different specifications, or on different datasets.""","""The paper suggests using meta-learning to tune the optimization schedule of alternative optimization problems. All of the reviewers agree that the paper is worthy of publication at ICLR. The authors have engaged with the reviewers and improved the paper since the submission. I asked the authors to address the rest of the comments in the camera ready version."""
948,"""Learning Cross-Lingual Sentence Representations via a Multi-task Dual-Encoder Model""","['sentence', 'embeddings', 'zero-shot', 'multilingual', 'multi-task', 'cross-lingual']","""A significant roadblock in multilingual neural language modeling is the lack of labeled non-English data. One potential method for overcoming this issue is learning cross-lingual text representations that can be used to transfer the performance from training on English tasks to non-English tasks, despite little to no task-specific non-English data. In this paper, we explore a natural setup for learning crosslingual sentence representations: the dual-encoder. We provide a comprehensive evaluation of our cross-lingual representations on a number of monolingual, crosslingual, and zero-shot/few-shot learning tasks, and also give an analysis of different learned cross-lingual embedding spaces.""","""Pros: - A new framework for learning sentence representations - Solid experiments and analyses - En-Zh / XNLI dataset was added, addressing the comment that no distant languages were considered; also ablation tests Cons: - The considered components are not novel, and their combination is straightforward - The set of downstream tasks is not very diverse (See R2) - Only high resource languages are considered (interesting to see it applied to real low resource languages) All reviewers agree that there is no modeling contribution. Overall, it is a solid paper but I do not believe that the contribution is sufficient. """
949,"""GamePad: A Learning Environment for Theorem Proving""","['Theorem proving', 'ITP', 'systems', 'neural embeddings']","""In this paper, we introduce a system called GamePad that can be used to explore the application of machine learning methods to theorem proving in the Coq proof assistant. Interactive theorem provers such as Coq enable users to construct machine-checkable proofs in a step-by-step manner. Hence, they provide an opportunity to explore theorem proving with human supervision. We use GamePad to synthesize proofs for a simple algebraic rewrite problem and train baseline models for a formalization of the Feit-Thompson theorem. We address position evaluation (i.e., predict the number of proof steps left) and tactic prediction (i.e., predict the next proof step) tasks, which arise naturally in tactic-based theorem proving.""","""This paper provides an RL environment defined over Coq, allowing for RL agents and other such systems to to be trained to propose tactics during the running of an ITP. I really like this general line of work, and the reviewers broadly speaking did as well. The one holdout is reviewer 3, who raises important concerns about the need for further evaluation. I understand and appreciate their points, and I think the authors should be careful to incorporate their feedback not only in final revisions to the paper, but in deciding what follow-on work to focus on. Nonetheless, and with all due respect to reviewer 3, who provided a review of acceptable quality, I am unsure the substance of their review merits a score as low as they have given. Considering the support the other reviews offer for the paper, I recommend acceptance for what the majority of reviewers believes is a good first step towards one day proving substantial new theorems using ITP-ML hybrids."""
950,"""Dense Morphological Network: An Universal Function Approximator""","['Mathematical Morphology', 'Neural Network', 'Activation Function', 'Universal Aproximatimation.']","""Artificial neural networks are built on the basic operation of linear combination and non-linear activation function. Theoretically this structure can approximate any continuous function with three layer architecture. But in practice learning the parameters of such network can be hard. Also the choice of activation function can greatly impact the performance of the network. In this paper we are proposing to replace the basic linear combination operation with non-linear operations that do away with the need of additional non-linear activation function. To this end we are proposing the use of elementary morphological operations (dilation and erosion) as the basic operation in neurons. We show that these networks (Denoted as Morph-Net) with morphological operations can approximate any smooth function requiring less number of parameters than what is necessary for normal neural networks. The results show that our network perform favorably when compared with similar structured network. We have carried out our experiments on MNIST, Fashion-MNIST, CIFAR10 and CIFAR100.""","""This work presents an interesting take on how to combine basic functions to lead to better activation functions. While the experiments in the paper show that the approach works well compared to the baselines that are used as reference, reviewers note that a more adequate assessment of the contribution would require comparing to stronger baselines or switching to tasks where the chosen baselines are indeed performing well. Authors are encouraged to follow the many suggestions of reviewers to strengthen their work."""
951,"""An Analysis of Composite Neural Network Performance from Function Composition Perspective""",[],"""This work investigates the performance of a composite neural network, which is composed of pre-trained neural network models and non-instantiated neural network models, connected to form a rooted directed graph. A pre-trained neural network model is generally a well trained neural network model targeted for a specific function. The advantages of adopting such a pre-trained model in a composite neural network are two folds. One is to benefit from other's intelligence and diligence and the other is saving the efforts in data preparation and resources and time in training. However, the overall performance of composite neural network is still not clear. In this work, we prove that a composite neural network, with high probability, performs better than any of its pre-trained components under certain assumptions. In addition, if an extra pre-trained component is added to a composite network, with high probability the overall performance will be improved. In the empirical evaluations, distinctively different applications support the above findings. ""","""Dear authors, All reviewers pointed out the fact that your result is about the expressivity of the big network rather than its accuracy, a result which is already known for the literature. I encourage you to carefully read all reviews should you wish to resubmit this work to a future conference."""
952,"""Graph HyperNetworks for Neural Architecture Search""","['neural', 'architecture', 'search', 'graph', 'network', 'hypernetwork', 'meta', 'learning', 'anytime', 'prediction']","""Neural architecture search (NAS) automatically finds the best task-specific neural network topology, outperforming many manual architecture designs. However, it can be prohibitively expensive as the search requires training thousands of different networks, while each training run can last for hours. In this work, we propose the Graph HyperNetwork (GHN) to amortize the search cost: given an architecture, it directly generates the weights by running inference on a graph neural network. GHNs model the topology of an architecture and therefore can predict network performance more accurately than regular hypernetworks and premature early stopping. To perform NAS, we randomly sample architectures and use the validation accuracy of networks with GHN generated weights as the surrogate search signal. GHNs are fast - they can search nearly 10 faster than other random search methods on CIFAR-10 and ImageNet. GHNs can be further extended to the anytime prediction setting, where they have found networks with better speed-accuracy tradeoff than the state-of-the-art manual designs.""","""The paper proposes an architecture search method based on graph hypernetworks (GHN). The core idea is that given a candidate architecture, GHN predicts its weights (similar to SMASH), which allows for fast evaluation w/o training the architecture from scratch. Unlike SMASH, GHN can operate on an arbitrary directed acyclic graph. Architecture search using GHN is fast and achieves competitive performance. Overall, this is a relevant contribution backed up by solid experiments, and should be accepted."""
953,"""RETHINKING SELF-DRIVING : MULTI -TASK KNOWLEDGE FOR BETTER GENERALIZATION AND ACCIDENT EXPLANATION ABILITY""","['Autonomous car', 'convolution network', 'image segmentation', 'depth estimation', 'generalization ability', 'explanation ability', 'multi-task learning']","""Current end-to-end deep learning driving models have two problems: (1) Poor generalization ability of unobserved driving environment when diversity of train- ing driving dataset is limited (2) Lack of accident explanation ability when driving models dont work as expected. To tackle these two problems, rooted on the be- lieve that knowledge of associated easy task is benificial for addressing difficult task, we proposed a new driving model which is composed of perception module for see and think and driving module for behave, and trained it with multi-task perception-related basic knowledge and driving knowledge stepwisely. Specifi- cally segmentation map and depth map (pixel level understanding of images) were considered as what & where and how far knowledge for tackling easier driving- related perception problems before generating final control commands for difficult driving task. The results of experiments demonstrated the effectiveness of multi- task perception knowledge for better generalization and accident explanation abil- ity. With our method the average sucess rate of finishing most difficult navigation tasks in untrained city of CoRL test surpassed current benchmark method for 15 percent in trained weather and 20 percent in untrained weathers.""","""The paper presents a unified system for perception and control that is trained in a step-wise fashion, with visual decoders to inspect scene parsing and understanding. Results demonstrate improved performance under certain conditions. But reviewers raise several concerns that must be addressed before the work is accepted. Reviewer Pros: + simple elegant design, easy to understand + provides some insight behind system function during failure conditions (error in perception vs control) + improves performance under a subset of tested conditions Reviewer Cons: - Concern about lack of novelty - Evaluation is limited in scope - References incomplete - Missing implementation details, hard to reproduce - Paper still contains many writing errors"""
954,"""S-System, Geometry, Learning, and Optimization: A Theory of Neural Networks""","['neural network theory', 'probability measure theory', 'probability coupling theory', 'S-System', 'optimization', 'random matrix', 'renormalization group', 'information geometry', 'coarse graining', 'hierarchy', 'activation function', 'symmetry']","""We present a formal measure-theoretical theory of neural networks (NN) built on {\it probability coupling theory}. Particularly, we present an algorithm framework, Hierarchical Measure Group and Approximate System (HMGAS), nicknamed S-System, of which NNs are special cases. In addition to many other results, the framework enables us to prove that 1) NNs implement {\it renormalization group (RG)} using information geometry, which points out that the large scale property to renormalize is dual Bregman divergence and completes the analog between NNs and RG; 2) and under a set of {\it realistic} boundedness and diversity conditions, for {\it large size nonlinear deep} NNs with a class of losses, including the hinge loss, all local minima are global minima with zero loss errors, using random matrix theory.""","""The paper is extremely difficult to read, even given that both reviewers have very strong math / theoretical background. Although it may potentially include interesting ideas, nothing in the work could not be understood by the ICLR audience. """
955,"""Learning to Refer to 3D Objects with Natural Language""","['Referential Language', '3D Objects', 'Part-Awareness', 'Neural Speakers', 'Neural Listeners']","""Human world knowledge is both structured and flexible. When people see an object, they represent it not as a pixel array but as a meaningful arrangement of semantic parts. Moreover, when people refer to an object, they provide descriptions that are not merely true but also relevant in the current context. Here, we combine these two observations in order to learn fine-grained correspondences between language and contextually relevant geometric properties of 3D objects. To do this, we employed an interactive communication task with human participants to construct a large dataset containing natural utterances referring to 3D objects from ShapeNet in a wide variety of contexts. Using this dataset, we developed neural listener and speaker models with strong capacity for generalization. By performing targeted lesions of visual and linguistic input, we discovered that the neural listener depends heavily on part-related words and associates these words correctly with the corresponding geometric properties of objects, suggesting that it has learned task-relevant structure linking the two input modalities. We further show that a neural speaker that is `listener-aware' --- that plans its utterances according to how an imagined listener would interpret its words in context --- produces more discriminative referring expressions than an `listener-unaware' speaker, as measured by human performance in identifying the correct object.""","""Paper develops a dataset and model for learning to refer to 3D objects. Reviewers raised concerns about lack of novelty. Fundamentally, it seems unclear what (if any) the take-away for an ML-audience would be after reading this paper. We encourage the authors to incorporate reviewer feedback and submit a stronger manuscript at a future (perhaps a more applied) venue. """
956,"""FUNCTIONAL VARIATIONAL BAYESIAN NEURAL NETWORKS""","['functional variational inference', 'Bayesian neural networks', 'stochastic processes']","""Variational Bayesian neural networks (BNN) perform variational inference over weights, but it is difficult to specify meaningful priors and approximating posteriors in a high-dimensional weight space. We introduce functional variational Bayesian neural networks (fBNNs), which maximize an Evidence Lower BOund (ELBO) defined directly on stochastic processes, i.e. distributions over functions. We prove that the KL divergence between stochastic processes is equal to the supremum of marginal KL divergences over all finite sets of inputs. Based on this, we introduce a practical training objective which approximates the functional ELBO using finite measurement sets and the spectral Stein gradient estimator. With fBNNs, we can specify priors which entail rich structure, including Gaussian processes and implicit stochastic processes. Empirically, we find that fBNNs extrapolate well using various structured priors, provide reliable uncertainty estimates, and can scale to large datasets.""","""This paper shows a promising new variational objective for Bayesian neural networks. The new objective is obtained by effectively considering a functional prior on the parameters. The paper is well-motivated and the mathematics are supported by theoretical justifications. There has been some discussion regarding the experimental section. On one hand, it contains several real and synthetic data which show the good performance of the proposed method. On the other hand, the reviewers requested deeper comparisons with state-of-the art (deep) GP models and more general problem settings. The AC decided that the paper can be accepted with the experiments contained in the new revision, although the authors would be strongly encouraged to address the reviewers comments in a non-cosmetic manner (as R2 put it). """
957,"""Optimal Attacks against Multiple Classifiers""","['online learning', 'nonconvex optimization', 'robust optimization']","""We study the problem of designing provably optimal adversarial noise algorithms that induce misclassification in settings where a learner aggregates decisions from multiple classifiers. Given the demonstrated vulnerability of state-of-the-art models to adversarial examples, recent efforts within the field of robust machine learning have focused on the use of ensemble classifiers as a way of boosting the robustness of individual models. In this paper, we design provably optimal attacks against a set of classifiers. We demonstrate how this problem can be framed as finding strategies at equilibrium in a two player, zero sum game between a learner and an adversary and consequently illustrate the need for randomization in adversarial attacks. The main technical challenge we consider is the design of best response oracles that can be implemented in a Multiplicative Weight Updates framework to find equilibrium strategies in the zero-sum game. We develop a series of scalable noise generation algorithms for deep neural networks, and show that it outperforms state-of-the-art attacks on various image classification tasks. Although there are generally no guarantees for deep learning, we show this is a well-principled approach in that it is provably optimal for linear classifiers. The main insight is a geometric characterization of the decision space that reduces the problem of designing best response oracles to minimizing a quadratic function over a set of convex polytopes.""","""Four reviewers have evaluated this paper. The reviewers have raised concerns about the specific formulation used for adversarial example generation which requires further clarity in motivation and interpretation. The reviewers have also made the point that the experimental evaluation is against previous work which tried to solve a different problem (black box based attack) and hence the conclusions are unconvincing."""
958,"""Graph Neural Networks with Generated Parameters for Relation Extraction""","['Graph Neural Networks', 'Relational Reasoning']","""Recently, progress has been made towards improving relational reasoning in machine learning field. Among existing models, graph neural networks (GNNs) is one of the most effective approaches for multi-hop relational reasoning. In fact, multi-hop relational reasoning is indispensable in many natural language processing tasks such as relation extraction. In this paper, we propose to generate the parameters of graph neural networks (GP-GNNs) according to natural language sentences, which enables GNNs to process relational reasoning on unstructured text inputs. We verify GP-GNNs in relation extraction from text. Experimental results on a human-annotated dataset and two distantly supervised datasets show that our model achieves significant improvements compared to the baselines. We also perform a qualitative analysis to demonstrate that our model could discover more accurate relations by multi-hop relational reasoning.""","""+ experiments on an interesting task: inferring relations which are not necessarily explicitly mentioned in a sentence but need to be induced relying on other relations + the idea to frame the relation prediction task as an inference task on a graph is interesting - the paper is not very well written, and it is hard to understand what exactly the contribution is. E.g., the authors contrast with previous work saying that previous work was relying on pre-defined graphs rather than inducing them. However, here they actually rely on predefined full graphs as well (i.e. full graphs connecting all entities). (See questions from R1) - the idea of predicting edge embeddings from the sentence is an interesting one. However, I do not see results studying alternative architectures (e.g., fixed transition matrices + gates / attention), or careful ablation studies. It is hard to say if this modification is indeed necessary / beneficial. (See also R3, agreeing that experiments look preliminary) - Extra baselines? E.g., what about layers of multi-head self-attention across entities? (as in Transformer). What about the number of parameters for the proposed model? Is there chance that it works better simply because it is a larger model? (See also R3) - evaluation only one dataset (not clear if any other datasets of this kind exist though) Overall, though I find the direction and certain aspects of the model quite interesting, the paper is not ready for publication."""
959,"""The meaning of ""most"" for visual question answering models""","['quantifier', 'evaluation methodology', 'psycholinguistics', 'visual question answering']","""The correct interpretation of quantifier statements in the context of a visual scene requires non-trivial inference mechanisms. For the example of ""most"", we discuss two strategies which rely on fundamentally different cognitive concepts. Our aim is to identify what strategy deep learning models for visual question answering learn when trained on such questions. To this end, we carefully design data to replicate experiments from psycholinguistics where the same question was investigated for humans. Focusing on the FiLM visual question answering model, our experiments indicate that a form of approximate number system emerges whose performance declines with more difficult scenes as predicted by Weber's law. Moreover, we identify confounding factors, like spatial arrangement of the scene, which impede the effectiveness of this system.""","""The paper studies an narrowly focused but interesting problem -- if the Visual Question answering model FILM from Perez et al (2018) is able to decide if most of the objects have a certain attribute or color. While the work itself is appreciate by the reviewers, concerns remain about the conclusion being limited in scope due to the synthetic nature of the data, and the analysis fairly narrow (a single model with a single very specific task). We encourage the authors to use reviewer feedback to make the manuscript stronger for a future deadline. """
960,"""Invariant-equivariant representation learning for multi-class data""","['representation learning', 'semantic representations', 'local vs global information', 'latent variable modelling', 'generative modelling', 'semi-supervised learning', 'variational autoencoders.']","""Representations learnt through deep neural networks tend to be highly informative, but opaque in terms of what information they learn to encode. We introduce an approach to probabilistic modelling that learns to represent data with two separate deep representations: an invariant representation that encodes the information of the class from which the data belongs, and an equivariant representation that encodes the symmetry transformation defining the particular data point within the class manifold (equivariant in the sense that the representation varies naturally with symmetry transformations). This approach to representation learning is conceptually transparent, easy to implement, and in-principle generally applicable to any data comprised of discrete classes of continuous distributions (e.g. objects in images, topics in language, individuals in behavioural data). We demonstrate qualitatively compelling representation learning and competitive quantitative performance, in both supervised and semi-supervised settings, versus comparable modelling approaches in the literature with little fine tuning.""","""The paper presents a new approach to learn separate class-invariant and class-equivariant latent representations, by training on labeled (and optional additional unlabelled) multi class data. Empirical results on MNIST and SVHN show that the method works well. Reviewers initially highlighted the following weaknesses of the paper: insufficient references and contrasting with related work (given that this problem space has been much explored before), limited novelty of the approach, limited experiments (MNIST only). One reviewer also mentioned a sometimes vague, overly hyperbolic, and meandering writeup. Authors did a commendable effort to improve the paper based on the reviews, adding new references, removing and rewriting parts of the paper to make it more focused, and providing experimental results on an additional dataset (SVHN). The paper did improve as a result. But while attenuated, the initial criticisms remain valid: the literature review and discussion remains short and too superficial. The peculiarities of the approach which grant it (modest) originality are insufficiently (theoretically and empirically) justified and not clearly enough put in context of the whole body of prior work. Consequently the proposed approach feels very ad-hoc. Finally the additional experiments are a step in the right direction, but experiments on only MNIST and SVHN are hardly enough in 2018 to convince the reader that a method has a universal potential and is more generally useful. Given the limited novelty, and in the absence of theoretical justification, experiments should be much more extensive, both in diversity of data/problems, and in the range of alternative approaches compared to, to build a convincing case. """
961,"""Learning Abstract Models for Long-Horizon Exploration""","['Reinforcement Learning', 'Hierarchical Reinforcement Learning', 'Model-based Reinforcement Learning', 'Exploration']","""In high-dimensional reinforcement learning settings with sparse rewards, performing effective exploration to even obtain any reward signal is an open challenge. While model-based approaches hold promise of better exploration via planning, it is extremely difficult to learn a reliable enough Markov Decision Process (MDP) in high dimensions (e.g., over 10^100 states). In this paper, we propose learning an abstract MDP over a much smaller number of states (e.g., 10^5), which we can plan over for effective exploration. We assume we have an abstraction function that maps concrete states (e.g., raw pixels) to abstract states (e.g., agent position, ignoring other objects). In our approach, a manager maintains an abstract MDP over a subset of the abstract states, which grows monotonically through targeted exploration (possible due to the abstract MDP). Concurrently, we learn a worker policy to travel between abstract states; the worker deals with the messiness of concrete states and presents a clean abstraction to the manager. On three of the hardest games from the Arcade Learning Environment (Montezuma's, Pitfall!, and Private Eye), our approach outperforms the previous state-of-the-art by over a factor of 2 in each game. In Pitfall!, our approach is the first to achieve superhuman performance without demonstrations.""","""The paper presents a novel approach to exploration in long-horizon / sparse reward RL settings. The approach is based on the notion of abstract states, a space that is lower-dimensional than the original state space, and in which transition dynamics can be learned and exploration is planned. A distributed algorithm is proposed for managing exploration in the abstract space (done by the manager), and learning to navigate between abstract states (workers). Empirical results show strong performance on hard exploration Atari games. The paper addresses a key challenge in reinforcement learning - learning and planning in long horizon MDPs. It presents an original approach to this problem, and demonstrates that it can be leveraged to achieve strong empirical results. At the same time, the reviewers and AC note several potential weaknesses, the focus here is on the subset that substantially affected the final acceptance decision. First, the paper deviates from the majority of current state of the art deep RL approaches by leveraging prior knowledge in the form of the RAM state. The cause for concern is not so much the use of the RAM information, but the comparison to other prior approaches using ""comparable amounts of prior knowledge"" - an argument that was considered misleading by the reviewers and AC. The reviewers make detailed suggestions on how to address these concerns in a future revision. Despite initially diverging assessments, the final consensus between the reviewers and AC was that the stated concerns would require a thorough revision of the paper and that it should not be accepted in its current stage. On a separate note, a lot of the discussion between R1 and the authors centered on whether more comparisons / a larger number of seeds should be run. The authors argued that the requested comparisons would be too costly. A suggestion for a future revision of the paper would be to only run a large number (e.g., 10) of seeds for the first 150M steps of each experiment, and presenting these results separately from the long-running experiments. This should be a cost efficient way to shed light on a particularly important range, and would help validate claims about sample efficiency."""
962,"""A Generative Model For Electron Paths""","['Molecules', 'Reaction Prediction', 'Graph Neural Networks', 'Deep Generative Models']","""Chemical reactions can be described as the stepwise redistribution of electrons in molecules. As such, reactions are often depicted using ""arrow-pushing"" diagrams which show this movement as a sequence of arrows. We propose an electron path prediction model (ELECTRO) to learn these sequences directly from raw reaction data. Instead of predicting product molecules directly from reactant molecules in one shot, learning a model of electron movement has the benefits of (a) being easy for chemists to interpret, (b) incorporating constraints of chemistry, such as balanced atom counts before and after the reaction, and (c) naturally encoding the sparsity of chemical reactions, which usually involve changes in only a small number of atoms in the reactants. We design a method to extract approximate reaction paths from any dataset of atom-mapped reaction SMILES strings. Our model achieves excellent performance on an important subset of the USPTO reaction dataset, comparing favorably to the strongest baselines. Furthermore, we show that our model recovers a basic knowledge of chemistry without being explicitly trained to do so.""","""The paper presents a graph neural network that represents the movements of electrons during chemical reactions, trained from a dataset to predict reactions outcomes. The paper is clearly written, the comparisons are sensical. There are some concerns by reviewer 3 about the experimental results: in particular the lack of a simpler baseline, and the experimental variance. I think the some of the important concerns from reviewer 3 were addressed in the rebuttal, and I hope the authors will update the manuscript accordingly. Overall, this is fitting for publication at ICLR 2019."""
963,"""Random mesh projectors for inverse problems""","['imaging', 'inverse problems', 'subspace projections', 'random Delaunay triangulations', 'CNN', 'geophysics', 'regularization']","""We propose a new learning-based approach to solve ill-posed inverse problems in imaging. We address the case where ground truth training samples are rare and the problem is severely ill-posed---both because of the underlying physics and because we can only get few measurements. This setting is common in geophysical imaging and remote sensing. We show that in this case the common approach to directly learn the mapping from the measured data to the reconstruction becomes unstable. Instead, we propose to first learn an ensemble of simpler mappings from the data to projections of the unknown image into random piecewise-constant subspaces. We then combine the projections to form a final reconstruction by solving a deconvolution-like problem. We show experimentally that the proposed method is more robust to measurement noise and corruptions not seen during training than a directly learned inverse.""","""This paper proposes a novel method of solving inverse problems that avoids direct inversion by first reconstructing various piecewise-constant projections of the unknown image (using a different CNN to learn each) and then combining them via optimization to solve the final inversion. Two of the reviewers requested more intuitions into why this two stage process would fight the inherent ambiguity. At the end of the discussion, two of the three reviewers are convinced by the derivations and empirical justification of the paper. The authors also have significantly improved the clarity of the manuscript throughout the discussion period. It would be interesting to see if there are any connections between such inversion via optimization with deep component analysis methods, e.g. Deep Component Analysis via Alternating Direction Neural Networks of Murdock et al. , that train neural architectures to effectively carry out the second step of optimization, as opposed to learning a feedforward mapping. """
964,"""Unsupervised Expectation Learning for Multisensory Binding""","['multisensory binding', 'expectation learning', 'unsupervised learning', 'Deep autoencoder', 'Growing-When-Required Network', 'animal recognition']","""Expectation learning is a continuous learning process which uses known multisensory bindings to modulate unisensory perception. When perceiving an event, we have an expectation on what we should see or hear which affects our unisensory perception. Expectation learning is known to enhance the unisensory perception of previously known multisensory events. In this work, we present a novel hybrid deep recurrent model based on audio/visual autoencoders, for unimodal stimulus representation and reconstruction, and a recurrent self-organizing network for multisensory binding of the representations. The model adapts concepts of expectation learning to enhance the unisensory representation based on the learned bindings. We demonstrate that the proposed model is capable of reconstructing signals from one modality by processing input of another modality for 43,500 Youtube videos in the animal subset of the AudioSet corpus. Our experiments also show that when using expectation learning, the proposed model presents state-of-the-art performance in representing and classifying unisensory stimuli.""","""The submission introduces a model that does learning of multisensory representations (by predicting one from the other), with an autoencoder structure. Generally, the reviewers liked the overall idea of the work, but found the clarity lacking, the evaluation insufficient (and not particularly state of the art), the requirement for paired training data quite limiting and the choices (VAE sometimes, autoencoder other times) somewhat ad hoc."""
965,"""Large-scale classification of structured objects using a CRF with deep class embedding""","['large-scale structure prediction', 'likelihood approximation', 'deep class embedding']","""This paper presents a novel deep learning architecture for classifying structured objects in ultrafine-grained datasets, where classes may not be clearly distinguishable by their appearance but rather by their context. We model sequences of images as linear-chain CRFs, and jointly learn the parameters from both local-visual features and neighboring class information. The visual features are learned by convolutional layers, whereas class-structure information is reparametrized by factorizing the CRF pairwise potential matrix. This forms a context-based semantic similarity space, learned alongside the visual similarities, and dramatically increases the learning capacity of contextual information. This new parametrization, however, forms a highly nonlinear objective function which is challenging to optimize. To overcome this, we develop a novel surrogate likelihood which allows for a local likelihood approximation of the original CRF with integrated batch-normalization. This model overcomes the difficulties of existing CRF methods to learn the contextual relationships thoroughly when there is a large number of classes and the data is sparse. The performance of the proposed method is illustrated on a huge dataset that contains images of retail-store product displays, and shows significantly improved results compared to linear CRF parametrization, unnormalized likelihood optimization, and RNN modeling.""","""The paper addresses the problem of large scale fine-grained classification by estimating pairwise potentials in a CRF model. The reviewers believe that the paper has some weaknesses including (1) the motivation for approximate learning is not clear (2) the approximate objective is not well studied and (3) the experiments are not convincing. The authors did not submit a rebuttal. I encourage the authors to take the feedback into account to improve the paper. """
966,"""GENERATING HIGH FIDELITY IMAGES WITH SUBSCALE PIXEL NETWORKS AND MULTIDIMENSIONAL UPSCALING""",[],"""The unconditional generation of high fidelity images is a longstanding benchmark for testing the performance of image decoders. Autoregressive image models have been able to generate small images unconditionally, but the extension of these methods to large images where fidelity can be more readily assessed has remained an open problem. Among the major challenges are the capacity to encode the vast previous context and the sheer difficulty of learning a distribution that preserves both global semantic coherence and exactness of detail. To address the former challenge, we propose the Subscale Pixel Network (SPN), a conditional decoder architecture that generates an image as a sequence of image slices of equal size. The SPN compactly captures image-wide spatial dependencies and requires a fraction of the memory and the computation. To address the latter challenge, we propose to use multidimensional upscaling to grow an image in both size and depth via intermediate stages corresponding to distinct SPNs. We evaluate SPNs on the unconditional generation of CelebAHQ of size 256 and of ImageNet from size 32 to 128. We achieve state-of-the-art likelihood results in multiple settings, set up new benchmark results in previously unexplored settings and are able to generate very high fidelity large scale samples on the basis of both datasets.""","""All reviewers recommend acceptance, with two reviewers in agreement that the results represent a significant advance for autoregressive generative models. The AC concurs. """
967,"""LayoutGAN: Generating Graphic Layouts with Wireframe Discriminators""",[],"""Layout is important for graphic design and scene generation. We propose a novel Generative Adversarial Network, called LayoutGAN, that synthesizes layouts by modeling geometric relations of different types of 2D elements. The generator of LayoutGAN takes as input a set of randomly-placed 2D graphic elements and uses self-attention modules to refine their labels and geometric parameters jointly to produce a realistic layout. Accurate alignment is critical for good layouts. We thus propose a novel differentiable wireframe rendering layer that maps the generated layout to a wireframe image, upon which a CNN-based discriminator is used to optimize the layouts in image space. We validate the effectiveness of LayoutGAN in various experiments including MNIST digit generation, document layout generation, clipart abstract scene generation and tangram graphic design.""","""Reviewers agree the paper should be accepted. See reviews below."""
968,"""DATNet: Dual Adversarial Transfer for Low-resource Named Entity Recognition""","['Low-resource', 'Named Entity Recognition']","""We propose a new architecture termed Dual Adversarial Transfer Network (DATNet) for addressing low-resource Named Entity Recognition (NER). Specifically, two variants of DATNet, i.e., DATNet-F and DATNet-P, are proposed to explore effective feature fusion between high and low resource. To address the noisy and imbalanced training data, we propose a novel Generalized Resource-Adversarial Discriminator (GRAD). Additionally, adversarial training is adopted to boost model generalization. We examine the effects of different components in DATNet across domains and languages and show that significant improvement can be obtained especially for low-resource data. Without augmenting any additional hand-crafted features, we achieve new state-of-the-art performances on CoNLL and Twitter NER---88.16% F1 for Spanish, 53.43% F1 for WNUT-2016, and 42.83% F1 for WNUT-2017.""","""A focused contribution that is clearly presented. That being said, the task of low-resource named entity recognition is fairly narrow and it is hard to tell how significant the empirical results are. The paper could be much stronger if it evaluated on a second task (and third task). Right now it is unclear whether the technique would generalize to other tasks."""
969,"""Sinkhorn AutoEncoders""","['generative models', 'autoencoders', 'optimal transport', 'sinkhorn algorithm']","""Optimal Transport offers an alternative to maximum likelihood for learning generative autoencoding models. We show how this principle dictates the minimization of the Wasserstein distance between the encoder aggregated posterior and the prior, plus a reconstruction error. We prove that in the non-parametric limit the autoencoder generates the data distribution if and only if the two distributions match exactly, and that the optimum can be obtained by deterministic autoencoders. We then introduce the Sinkhorn AutoEncoder (SAE), which casts the problem into Optimal Transport on the latent space. The resulting Wasserstein distance is minimized by backpropagating through the Sinkhorn algorithm. SAE models the aggregated posterior as an implicit distribution and therefore does not need a reparameterization trick for gradients estimation. Moreover, it requires virtually no adaptation to different prior distributions. We demonstrate its flexibility by considering models with hyperspherical and Dirichlet priors, as well as a simple case of probabilistic programming. SAE matches or outperforms other autoencoding models in visual quality and FID scores. ""","""The reviewers appreciated the contribution of combining Wasserstein Autoencoders with the Sinkhorn algorithm. Yet R4 as well as the author of the WAE paper (Ilya Tolstikhin) both expressed concerns about the empirical evaluation. While R1-R3 were all somewhat positive in their recommendation after the rebuttal, they all have somewhat lower confidence reviews, as is also clear by their comments. The AC decided to follow the recommendation of R4 as they were the most expert reviewer. The AC thus recommends to ""revise and resubmit"" the paper."""
970,"""The Laplacian in RL: Learning Representations with Efficient Approximations""","['Laplacian', 'reinforcement learning', 'representation']","""The smallest eigenvectors of the graph Laplacian are well-known to provide a succinct representation of the geometry of a weighted graph. In reinforcement learning (RL), where the weighted graph may be interpreted as the state transition process induced by a behavior policy acting on the environment, approximating the eigenvectors of the Laplacian provides a promising approach to state representation learning. However, existing methods for performing this approximation are ill-suited in general RL settings for two main reasons: First, they are computationally expensive, often requiring operations on large matrices. Second, these methods lack adequate justification beyond simple, tabular, finite-state settings. In this paper, we present a fully general and scalable method for approximating the eigenvectors of the Laplacian in a model-free RL context. We systematically evaluate our approach and empirically show that it generalizes beyond the tabular, finite-state setting. Even in tabular, finite-state settings, its ability to approximate the eigenvectors outperforms previous proposals. Finally, we show the potential benefits of using a Laplacian representation learned using our method in goal-achieving RL tasks, providing evidence that our technique can be used to significantly improve the performance of an RL agent.""","""This paper provides a novel and non-trivial method for approximating the eigenvectors of the Laplacian, in large or continuous state environments. Eigenvectors of the Laplacian have been used for proto-value functions and eigenoptions, but it has remained an open problem to extend their use to the non-tabular case. This paper makes an important advance towards this goal, and will be of interest to many that would like to learn state representations based on the geometric information given by the Laplacian. The paper could be made stronger by including a short discussion on why the limitations of this approach. Its an important new direction, but there must still be open questions (e.g., issues with the approach used to approximate the orthogonality constraint). It will be beneficial to readers to understand these issues."""
971,"""Neural Distribution Learning for generalized time-to-event prediction""","['Deep Learning', 'Survival Analysis', 'Event prediction', 'Time Series', 'Probabilistic Programming', 'Density Networks']","""Predicting the time to the next event is an important task in various domains. However, due to censoring and irregularly sampled sequences, time-to-event prediction has resulted in limited success only for particular tasks, architectures and data. Using recent advances in probabilistic programming and density networks, we make the case for a generalized parametric survival approach, sequentially predicting a distribution over the time to the next event. Unlike previous work, the proposed method can use asynchronously sampled features for censored, discrete, and multivariate data. Furthermore, it achieves good performance and near perfect calibration for probabilistic predictions without using rigid network-architectures, multitask approaches, complex learning schemes or non-trivial adaptations of cox-models. We firmly establish that this can be achieved in the standard neural network framework by simply switching out the output layer and loss function.""","""All reviewers agree to reject. While there were many positive points to this work, reviewers believed that it was not yet ready for acceptance."""
972,"""Domain Generalization via Invariant Representation under Domain-Class Dependency""","['domain generalization', 'adversarial learning', 'invariant feature learning']","""Learning domain-invariant representation is a dominant approach for domain generalization, where we need to build a classifier that is robust toward domain shifts induced by change of users, acoustic or lighting conditions, etc. However, prior domain-invariance-based methods overlooked the underlying dependency of classes (target variable) on source domains during optimization, which causes the trade-off between classification accuracy and domain-invariance, and often interferes with the domain generalization performance. This study first provides the notion of domain generalization under domain-class dependency and elaborates on the importance of considering the dependency by expanding the analysis of Xie et al. (2017). We then propose a method, invariant feature learning under optimal classifier constrains (IFLOC), which explicitly considers the dependency and maintains accuracy while improving domain-invariance. Specifically, the proposed method regularizes the representation so that it has as much domain information as the class labels, unlike prior methods that remove all domain information. Empirical validations show the superior performance of IFLOC to baseline methods, supporting the importance of the domain-class dependency in domain generalization and the efficacy of the proposed method for overcoming the issue.""","""This paper proposes a new solution to the problem of domain generalization where the label distribution may differ across domains. The authors argue that prior work which ignores this observation suffers from an accuracy-vs-invariance trade-off while their work does not. The main contribution of the work is to 1) consider the case of different label distributions across domains and 2) to propose a regularizer extension to Xie 2017 to handle this. There was disagreement between the reviewers on whether or not this contribution is significant enough to warrant publication. Two reviewers expressed concern of whether 1) naturally occurring data sources suffer substantially from this label distribution mismatch and 2) whether label distribution mismatch in practice results in significant performance loss for existing domain generalization techniques. Based on the experiments and discussions available now the answer to the above two points remains unclear. These key questions should be clarified and further justified before publication."""
973,"""CoT: Cooperative Training for Generative Modeling of Discrete Data""","['Generative Models', 'Sequence Modeling', 'Text Generation']","""We propose Cooperative Training (CoT) for training generative models that measure a tractable density for discrete data. CoT coordinately trains a generator G and an auxiliary predictive mediator M. The training target of M is to estimate a mixture density of the learned distribution G and the target distribution P, and that of G is to minimize the Jensen-Shannon divergence estimated through M. CoT achieves independent success without the necessity of pre-training via Maximum Likelihood Estimation or involving high-variance algorithms like REINFORCE. This low-variance algorithm is theoretically proved to be superior for both sample generation and likelihood prediction. We also theoretically and empirically show the superiority of CoT over most previous algorithms in terms of generative quality and diversity, predictive generalization ability and computational cost.""","""The paper proposes an original and interesting alternative to GANs for optimizing a (proxy to) Jensen-Shannon divergence for discrete sequence data. Experimental results seem promising. Official reviewers were largely positive based on originality and results. However, as it currently stands, the paper still makes false claims that are not well explained or supported, in particular its repeated central claim to provide a ""low-variance, bias-free algorithm"" to optimize JS. Given that these central issues were clearly pointed out in a review from a prior submission of this work to another venue (review reposted on the current OpenReview thread on Nov. 6), the AC feels that the authors had had plenty of time to look into them and address them in the paper, as well as occasions to reference and discuss relevant related work pointed in that review. The current version of the paper does neither. The algorithm is not unbiased for at least two reasons pointed out in discussions: a) in practice a parameterized mediator will be unable to match the true P+G, at best yielding a useful biased estimate (not unlike how GAN's parameterized discriminator induces bias). b) One would need to use REINFORCE (or similar) to get an unbiased estimate of the gradient in Eq. 13, a key detail omitted from the paper. From the discussion thread it is possible that authors were initially confused about the fact that this fundamental issue did not disappear with Eq. 13 (they commented ""most important idea we want to present in this paper is HOW TO avoid incorporating REINFORCE. Please refer to Eq.13, which is the key to the success of this.""). But rather, as guessed by a commentator, that a heuristic implementation, not explained in the paper, dropped the REINFORCE term thus effectively trading variance for bias. On December 4th authors posted a justification confirming heuristically dropping the REINFORCE terms when taking the gradient of Eq. 13, and said they could attach detailed analysis and experiment results in the camera-ready version. However if one of the ""most important idea"" of the paper is how to avoid REINFORCE (as still implied and highlighted in the abstract), the AC finds it worrisome that the paper had no explanation of when and how this was done, and no analysis of the bias induced by (unreportedly) dropping the term. The approach remains original, interesting, and potentially promising, but as it currently stands, AC and SAC agreed that inexact theoretical over-claiming and insufficient justification and in-depth analysis of key heuristic shortcuts/tradeoffs (however useful) are too important for their fixing to be entrusted to a final camera-ready revision step. A major revision that clearly adresses these issues in depth (both in how the approach is presented and in supporting experiments) will constitute a much more convincing, sound, and impactful research contribution. """
974,"""Deep Learning 3D Shapes Using Alt-az Anisotropic 2-Sphere Convolution""","['Spherical Convolution', 'Geometric deep learning', '3D shape analysis']","""The ground-breaking performance obtained by deep convolutional neural networks (CNNs) for image processing tasks is inspiring research efforts attempting to extend it for 3D geometric tasks. One of the main challenge in applying CNNs to 3D shape analysis is how to define a natural convolution operator on non-euclidean surfaces. In this paper, we present a method for applying deep learning to 3D surfaces using their spherical descriptors and alt-az anisotropic convolution on 2-sphere. A cascade set of geodesic disk filters rotate on the 2-sphere and collect spherical patterns and so to extract geometric features for various 3D shape analysis tasks. We demonstrate theoretically and experimentally that our proposed method has the possibility to bridge the gap between 2D images and 3D shapes with the desired rotation equivariance/invariance, and its effectiveness is evaluated in applications of non-rigid/ rigid shape classification and shape retrieval.""","""Strengths: Well written paper on a new kind of spherical convolution for use in spherical CNNs. Evaluated on rigid and non-rigid 3D shape recognition and retrieval problems. Paper provides solid strategy for efficient GPU implementation. Weaknesses: There was some misunderstanding about the properties of the alt-az convolution detected by one of the reviewers along with some points needing clarifications. However, discussion of these issues appears to have led to a resolution of the issues. Contention: The weaknesses above were discussed in some detail, but the procedure was not particularly contentious and the discussion unfolded well. All reviewers rate the paper as accept, the paper clearly provides value to the community and therefore should be accepted. """
975,"""Visualizing and Understanding the Semantics of Embedding Spaces via Algebraic Formulae""","['visualization', 'embeddings', 'representations', 't-sne', 'natural', 'language', 'processing', 'machine', 'learning', 'algebra']","""Embeddings are a fundamental component of many modern machine learning and natural language processing models. Understanding them and visualizing them is essential for gathering insights about the information they capture and the behavior of the models. State of the art in analyzing embeddings consists in projecting them in two-dimensional planes without any interpretable semantics associated to the axes of the projection, which makes detailed analyses and comparison among multiple sets of embeddings challenging. In this work, we propose to use explicit axes defined as algebraic formulae over embeddings to project them into a lower dimensional, but semantically meaningful subspace, as a simple yet effective analysis and visualization methodology. This methodology assigns an interpretable semantics to the measures of variability and the axes of visualizations, allowing for both comparisons among different sets of embeddings and fine-grained inspection of the embedding spaces. We demonstrate the power of the proposed methodology through a series of case studies that make use of visualizations constructed around the underlying methodology and through a user study. The results show how the methodology is effective at providing more profound insights than classical projection methods and how it is widely applicable to many other use cases.""","""Several visualizations are shown in this paper but it is unclear if they are novel."""
976,"""Unsupervised Control Through Non-Parametric Discriminative Rewards""","['deep reinforcement learning', 'goals', 'UVFA', 'mutual information']","""Learning to control an environment without hand-crafted rewards or expert data remains challenging and is at the frontier of reinforcement learning research. We present an unsupervised learning algorithm to train agents to achieve perceptually-specified goals using only a stream of observations and actions. Our agent simultaneously learns a goal-conditioned policy and a goal achievement reward function that measures how similar a state is to the goal state. This dual optimization leads to a co-operative game, giving rise to a learned reward function that reflects similarity in controllable aspects of the environment instead of distance in the space of observations. We demonstrate the efficacy of our agent to learn, in an unsupervised manner, to reach a diverse set of goals on three domains -- Atari, the DeepMind Control Suite and DeepMind Lab.""","""This paper introduces an unsupervised algorithm to learn a goal-conditioned policy and the reward function by formulating a mutual information maximization problem. The idea is interesting, but the experimental studies seem not rigorous enough. In the final version, I would like to see some more detailed analysis of the results obtained by the baselines (pixel approaches), as well as careful discussion on the relationship with other related work, such as Variational Intrinsic Control."""
977,"""Neural network gradient-based learning of black-box function interfaces""","['neural networks', 'black box functions', 'gradient descent']","""Deep neural networks work well at approximating complicated functions when provided with data and trained by gradient descent methods. At the same time, there is a vast amount of existing functions that programmatically solve different tasks in a precise manner eliminating the need for training. In many cases, it is possible to decompose a task to a series of functions, of which for some we may prefer to use a neural network to learn the functionality, while for others the preferred method would be to use existing black-box functions. We propose a method for end-to-end training of a base neural network that integrates calls to existing black-box functions. We do so by approximating the black-box functionality with a differentiable neural network in a way that drives the base network to comply with the black-box function interface during the end-to-end optimization process. At inference time, we replace the differentiable estimator with its external black-box non-differentiable counterpart such that the base network output matches the input arguments of the black-box function. Using this ``Estimate and Replace'' paradigm, we train a neural network, end to end, to compute the input to black-box functionality while eliminating the need for intermediate labels. We show that by leveraging the existing precise black-box function during inference, the integrated model generalizes better than a fully differentiable model, and learns more efficiently compared to RL-based methods.""","""The paper focuses on hybrid pipelines that contain black-boxes and neural networks, making it difficult to train the neural components due to non-differentiability. As a solution, this paper proposes to replace black-box functions with neural modules that approximate them during training, so that end-to-end training can be used, but at test time use the original black box modules. The authors propose a number of variations: offline, online, and hybrid of the two, to train the intermediate auxiliary networks. The proposed model is shown to be effective on a number of synthetic datasets. The reviewers and AC note the following potential weaknesses: (1) the reviewers found some of the experiment details to be scattered, (2) It was unclear what happens if there is a mismatch between the auxiliary network and the black box function it is approximating, especially if the function is one, like sorting, that is difficult for neural models to approximate, and (3) the text lacked description of real-world tasks for which such a hybrid pipeline would be useful. The authors provide comments and a revision to address these concerns. They added a section that described the experiment setup to aid reproducibility, and incorporated more details in the results and related work, as suggested by the reviewers. Although these changes go a long way, some of the concerns, especially regarding the mismatch between neural and black box function, still remain. Overall, the reviewers agreed that the issues had been addressed to a sufficient degree, and the paper should be accepted."""
978,"""Verification of Non-Linear Specifications for Neural Networks""","['Verification', 'Convex Optimization', 'Adversarial Robustness']","""Prior work on neural network verification has focused on specifications that are linear functions of the output of the network, e.g., invariance of the classifier output under adversarial perturbations of the input. In this paper, we extend verification algorithms to be able to certify richer properties of neural networks. To do this we introduce the class of convex-relaxable specifications, which constitute nonlinear specifications that can be verified using a convex relaxation. We show that a number of important properties of interest can be modeled within this class, including conservation of energy in a learned dynamics model of a physical system; semantic consistency of a classifier's output labels under adversarial perturbations and bounding errors in a system that predicts the summation of handwritten digits. Our experimental evaluation shows that our method is able to effectively verify these specifications. Moreover, our evaluation exposes the failure modes in models which cannot be verified to satisfy these specifications. Thus, emphasizing the importance of training models not just to fit training data but also to be consistent with specifications.""","""This paper proposes verification algorithms for a class of convex-relaxable specifications to evaluate the robustness of neural networks under adversarial examples. The reviewers were unanimous in their vote to accept the paper. Note: the remaining score of 5 belongs to a reviewer who agreed to acceptance in the discussion."""
979,"""Overlapping Community Detection with Graph Neural Networks""","['community detection', 'deep learning for graphs']","""Community detection in graphs is of central importance in graph mining, machine learning and network science. Detecting overlapping communities is especially challenging, and remains an open problem. Motivated by the success of graph-based deep learning in other graph-related tasks, we study the applicability of this framework for overlapping community detection. We propose a probabilistic model for overlapping community detection based on the graph neural network architecture. Despite its simplicity, our model outperforms the existing approaches in the community recovery task by a large margin. Moreover, due to the inductive formulation, the proposed model is able to perform out-of-sample community detection for nodes that were not present at training time""","""The paper provides an interesting combination of existing techniques (such as GCN and and the Bernoulli-Poisson link) to address the problem of overlapping community detection. However, there were concerns about lack of novelty, evaluation metrics, and missing comparisons with previous work. The authors did not provide a response to address these concerns."""
980,"""ARM: Augment-REINFORCE-Merge Gradient for Stochastic Binary Networks""","['Antithetic sampling', 'variable augmentation', 'deep discrete latent variable models', 'variance reduction', 'variational auto-encoder']","""To backpropagate the gradients through stochastic binary layers, we propose the augment-REINFORCE-merge (ARM) estimator that is unbiased, exhibits low variance, and has low computational complexity. Exploiting variable augmentation, REINFORCE, and reparameterization, the ARM estimator achieves adaptive variance reduction for Monte Carlo integration by merging two expectations via common random numbers. The variance-reduction mechanism of the ARM estimator can also be attributed to either antithetic sampling in an augmented space, or the use of an optimal anti-symmetric ""self-control"" baseline function together with the REINFORCE estimator in that augmented space. Experimental results show the ARM estimator provides state-of-the-art performance in auto-encoding variational inference and maximum likelihood estimation, for discrete latent variable models with one or multiple stochastic binary layers. Python code for reproducible research is publicly available.""","""This paper introduces a new way to estimate gradients of expectations of discrete random variables by introducing antithetic noise samples for use in a control variate. Quality: The experiments are mostly appropriate, although I disagree with the choice to present validation and test-set results instead of training-time results. If the goal of the method is to reduce variance, then checking whether optimization is improved (training loss) is the most direct measure. However reasonable people can disagree about this. I also think the toy experiment (copied from the REBAR and RELAX paper) is a bit too easy for this method, since it relies on taking two antithetic samples. I would have liked to see a categorical extension of the same experiment. Clarity: I think that this method will not have the impact it otherwise could because of the authors' fearless use of long equations and heavy notation throughout. This is unavoidable to some degree, but 1) The title of the paper isn't very descriptive 2) Why not follow previous work and use \theta instead of \phi for the parameters being optimized? The presentation has come a long way, but I fear that few besides our intrepid reviewers will have the stomach. I recommend providing more intuition throughout. Originality: The use of antithetic samples to reduce variance is old, but this seems like a well-thought-through and non-trivial application of the idea to this setting. Significance: Ultimately I think this is a new direction in gradient estimators for discrete RVs. I don't think this is the last word in this direction but it's both an empirical improvement, and will inspire further work."""
981,"""Globally Soft Filter Pruning For Efficient Convolutional Neural Networks""","['Filter Pruning', 'Model Compression', 'Efficient Convolutional Neural Networks']","""This paper propose a cumulative saliency based Globally Soft Filter Pruning (GSFP) scheme to prune redundant filters of Convolutional Neural Networks (CNNs).Specifically, the GSFP adopts a robust pruning method, which measures the global redundancy of the filter in the whole model by using the soft pruning strategy. In addition, in the model recovery process after pruning, we use the cumulative saliency strategy to improve the accuracy of pruning. GSFP has two advantages over previous works:(1) More accurate pruning guidance. For a pre-trained CNN model, the saliency of the filter varies with different input data. Therefore, accumulating the saliency of the filter over the entire data set can provide more accurate guidance for pruning. On the other hand, pruning from a global perspective is more accurate than local pruning. (2) More robust pruning strategy. We propose a reasonable normalization formula to prevent certain layers of filters in the network from being completely clipped due to excessive pruning rate.""","""This paper proposes new heuristics to prune and compress neural networks. The paper is well organized. However, reviewers are concerned that the novelty is relatively limited. The advantage of the proposed method is marginal on ImageNet. What is effective is not very clear. Therefore, recommend for rejection. """
982,"""Automata Guided Skill Composition""","['Skill composition', 'temporal logic', 'finite state automata']","""Skills learned through (deep) reinforcement learning often generalizes poorly across tasks and re-training is necessary when presented with a new task. We present a framework that combines techniques in formal methods with reinforcement learning (RL) that allows for the convenient specification of complex temporal dependent tasks with logical expressions and construction of new skills from existing ones with no additional exploration. We provide theoretical results for our composition technique and evaluate on a simple grid world simulation as well as a robotic manipulation task.""","""The authors present an interesting approach for combining finite state automata to compose new policies using temporal logic. The reviewers found this contribution interesting but had several questions that suggests that the current paper presentation could be significantly clarified and situated with respect to other literature. Given the strong pool of papers, this paper was borderline and the authors are encouraged to revise their paper to address the reviewers feedback. """
983,"""Sample-efficient policy learning in multi-agent Reinforcement Learning via meta-learning""","['Multi-agent', 'Reinforcement Learning', 'Meta-learning']","""To gain high rewards in muti-agent scenes, it is sometimes necessary to understand other agents and make corresponding optimal decisions. We can solve these tasks by first building models for other agents and then finding the optimal policy with these models. To get an accurate model, many observations are needed and this can be sample-inefficient. What's more, the learned model and policy can overfit to current agents and cannot generalize if the other agents are replaced by new agents. In many practical situations, each agent we face can be considered as a sample from a population with a fixed but unknown distribution. Thus we can treat the task against some specific agents as a task sampled from a task distribution. We apply meta-learning method to build models and learn policies. Therefore when new agents come, we can adapt to them efficiently. Experiments on grid games show that our method can quickly get high rewards.""","""The paper extends MAML so that a learned behavior can be quickly (sample-efficiently) adapted to a new agend (allied or opponent). The approach is tested on two simple tasks in 2D gridworld environments: chasing and path blocking. The experiments are very limited, they do not suffice to support the claims about the method. The authors did not enter a rebuttal and all the reviewers agree that the paper is not good enough for ICLR."""
984,"""I Know the Feeling: Learning to Converse with Empathy""","['dialogue generation', 'nlp applications', 'grounded text  generation', 'contextual representation learning']","""Beyond understanding what is being discussed, human communication requires an awareness of what someone is feeling. One challenge for dialogue agents is recognizing feelings in the conversation partner and replying accordingly, a key communicative skill that is trivial for humans. Research in this area is made difficult by the paucity of suitable publicly available datasets both for emotion and dialogues. This work proposes a new task for empathetic dialogue generation and EmpatheticDialogues, a dataset of 25k conversations grounded in emotional situations to facilitate training and evaluating dialogue systems. Our experiments indicate that dialogue models that use our dataset are perceived to be more empathetic by human evaluators, while improving on other metrics as well (e.g. perceived relevance of responses, BLEU scores), compared to models merely trained on large-scale Internet conversation data. We also present empirical comparisons of several ways to improve the performance of a given model by leveraging existing models or datasets without requiring lengthy re-training of the full model.""","""The reviewers raised a number of concerns including the usefulness of the presented dataset given that the collected data is acted rather than naturalistic (and the large body of research in affective computing explains that models trained on acted data cannot generalise to naturalistic data), no methodological novelty in the presented work, and relatively uninteresting application with very limited real-world application (it remains unclear whether having better empathetic dialogues would be truly crucial for any real-life application and, in addition, all work is based on acted rather than real-world data). The authors rebuttal addressed some of the reviewers concerns but not fully (especially when it comes to usefulness of the data). Overall, I believe that the effort to collect the presented database is noble and may be useful to the community to a small extent. However, given the unrealism of the data and, in turn, very limited (if any) generalisability of the presented to real-world scenarios, and lack of methodological contribution, I cannot recommend this paper for presentation at ICLR."""
985,"""Accelerated Sparse Recovery Under Structured Measurements""",['sparse recovery'],"""Extensive work on compressed sensing has yielded a rich collection of sparse recovery algorithms, each making different tradeoffs between recovery condition and computational efficiency. In this paper, we propose a unified framework for accelerating various existing sparse recovery algorithms without sacrificing recovery guarantees by exploiting structure in the measurement matrix. Unlike fast algorithms that are specific to particular choices of measurement matrices where the columns are Fourier or wavelet filters for example, the proposed approach works on a broad range of measurement matrices that satisfy a particular property. We precisely characterize this property, which quantifies how easy it is to accelerate sparse recovery for the measurement matrix in question. We also derive the time complexity of the accelerated algorithm, which is sublinear in the signal length in each iteration. Moreover, we present experimental results on real world data that demonstrate the effectiveness of the proposed approach in practice. ""","""The main idea of this paper is to use nearest neighbor search to to accelerate iterative thresholding based sparse recovery algorithms. All reviewers were underwhelmed by somewhat straightforward combination of existing results in sparse recovery and nearest-neighbor search. While the proposed method seems effective in practice, the paper has the feel of not being a fully publishable unit yet. Several technical questions were asked but no author feedback was provided to potentially lift this paper up."""
986,"""Jumpout: Improved Dropout for Deep Neural Networks with Rectified Linear Units""","['Dropout', 'deep neural networks with ReLU', 'local linear model']","""Dropout is a simple yet effective technique to improve generalization performance and prevent overfitting in deep neural networks (DNNs). In this paper, we discuss three novel observations about dropout to better understand the generalization of DNNs with rectified linear unit (ReLU) activations: 1) dropout is a smoothing technique that encourages each local linear model of a DNN to be trained on data points from nearby regions; 2) a constant dropout rate can result in effective neural-deactivation rates that are significantly different for layers with different fractions of activated neurons; and 3) the rescaling factor of dropout causes an inconsistency to occur between the normalization during training and testing conditions when batch normalization is also used. The above leads to three simple but nontrivial improvements to dropout resulting in our proposed method ""Jumpout."" Jumpout samples the dropout rate using a monotone decreasing distribution (such as the right part of a truncated Gaussian), so the local linear model at each data point is trained, with high probability, to work better for data points from nearby than from more distant regions. Instead of tuning a dropout rate for each layer and applying it to all samples, jumpout moreover adaptively normalizes the dropout rate at each layer and every training sample/batch, so the effective dropout rate applied to the activated neurons are kept the same. Moreover, we rescale the outputs of jumpout for a better trade-off that keeps both the variance and mean of neurons more consistent between training and test phases, which mitigates the incompatibility between dropout and batch normalization. Compared to the original dropout, jumpout shows significantly improved performance on CIFAR10, CIFAR100, Fashion- MNIST, STL10, SVHN, ImageNet-1k, etc., while introducing negligible additional memory and computation costs.""","""The paper introduces a new variant of the Dropout method. The reviewers agree that the procedure is clear. However, motivations behind the method are heuristic, and have to lean much on empirical evidence. A strong motivation behind the procedure is lacking, and the motivation behind the method is unclear. Furthermore, the empirical evidence is lacking in detail and could use better comparisons with existing literature."""
987,"""Machine Translation With Weakly Paired Bilingual Documents""","['Natural Language Processing', 'Machine Translation', 'Unsupervised Learning']","""Neural machine translation, which achieves near human-level performance in some languages, strongly relies on the availability of large amounts of parallel sentences, which hinders its applicability to low-resource language pairs. Recent works explore the possibility of unsupervised machine translation with monolingual data only, leading to much lower accuracy compared with the supervised one. Observing that weakly paired bilingual documents are much easier to collect than bilingual sentences, e.g., from Wikipedia, news websites or books, in this paper, we investigate the training of translation models with weakly paired bilingual documents. Our approach contains two components/steps. First, we provide a simple approach to mine implicitly bilingual sentence pairs from document pairs which can then be used as supervised signals for training. Second, we leverage the topic consistency of two weakly paired documents and learn the sentence-to-sentence translation by constraining the word distribution-level alignments. We evaluate our proposed method on weakly paired documents from Wikipedia on four tasks, the widely used WMT16 German pseudo-formula English and WMT13 Spanish pseudo-formula English tasks, and obtain pseudo-formula / pseudo-formula and pseudo-formula / pseudo-formula BLEU points separately, outperforming state-of-the-art unsupervised results by more than 5 BLEU points and reducing the gap between unsupervised translation and supervised translation up to 50\%. ""","""This paper proposes a new method to mine sentence from Wikipedia and use them to train an MT system, and also a topic-based loss function. In particular, the first contribution, which is the main aspect of the proposal is effective, outperforming methods for fully unsupervised learning. The main concern with the proposed method, or at least it's description in the paper, is that it isn't framed appropriately with respect to previous work on mining parallel sentences from comparable corpora such as Wikipedia. Based on interaction in the reviews, I feel that things are now framed a bit better, and there are additional baselines, but still the explanation in the paper isn't framed with respect to this previous work, and also the baselines are not competitive, despite previous work reporting very nice results for these previous methods. I feel like this could be a very nice paper at some point if it's re-written with the appropriate references to previous work, and experimental results where the baselines are done appropriately. Thus at this time I'm not recommending that the paper be accepted, but encourage the authors to re-submit a revised version in the future."""
988,"""Zero-training Sentence Embedding via Orthogonal Basis""","['Natural Language Processing', 'Sentence Embeddings']","""We propose a simple and robust training-free approach for building sentence representations. Inspired by the Gram-Schmidt Process in geometric theory, we build an orthogonal basis of the subspace spanned by a word and its surrounding context in a sentence. We model the semantic meaning of a word in a sentence based on two aspects. One is its relatedness to the word vector subspace already spanned by its contextual words. The other is its novel semantic meaning which shall be introduced as a new basis vector perpendicular to this existing subspace. Following this motivation, we develop an innovative method based on orthogonal basis to combine pre-trained word embeddings into sentence representation. This approach requires zero training and zero parameters, along with efficient inference performance. We evaluate our approach on 11 downstream NLP tasks. Experimental results show that our model outperforms all existing zero-training alternatives in all the tasks and it is competitive to other approaches relying on either large amounts of labelled data or prolonged training time.""","""The paper proposes a simple approach for computing a sentence embedding as a weighted combination of pre-trained word embeddings, which obtains nice results on a number of tasks. The approach is described as training-free but does require computing principal components of word embedding subspaces on the test set (similarly to some earlier work). The reviewers are generally in agreement that the approach is interesting, and the results are encouraging. However, there is some concern about the clarity of the paper and in particular the placement of the work in relation to other methods. There is also a bit of concern about whether there is sufficient novelty compared to Arora et al. 2017, which also compose sentence embeddings as weighted combinations of word embeddings, and also use a principal subspace of embeddings in the test set. This AC feels that the method here is sufficiently different from Arora et al., but agrees with the reviewers that the paper clarity needs to be improved, so that the community can appreciate what is gained from the new aspects of the approach and what conclusions should be drawn from each experimental comparison."""
989,"""Deconfounding Reinforcement Learning in Observational Settings""","['confounder', 'causal inference', 'reinforcement learning']","""In this paper, we propose a general formulation to cope with a family of reinforcement learning tasks in observational settings, that is, learning good policies solely from the historical data produced by real environments with confounders (i.e., the factors affecting both actions and rewards). Based on the proposed approach, we extend one representative of reinforcement learning algorithms: the Actor-Critic method, to its deconfounding variant, which is also straightforward to be applied to other algorithms. In addition, due to lack of datasets in this direction, a benchmark is developed for deconfounding reinforcement learning algorithms by revising OpenAI Gym and MNIST. We demonstrate that the proposed algorithms are superior to traditional reinforcement learning algorithms in confounded environments. To the best of our knowledge, this is the first time that confounders are taken into consideration for addressing full reinforcement learning problems.""","""The paper studies RL based on data with confounders, where the confounders can affect both rewards and actions. The setting is relevant in many problems and can have much potential. This work is an interesting and useful attempt. However, reviewers raised many questions regarding the problem setup and its comparison to related areas like causal inference. While the author response provided further helpful details, the questions remained among the reviewers. Therefore, the paper is not recommended for acceptance in its current stage; more work is needed to better motivate the setting and clarify its relation to other areas. Furthermore, the paper should probably discuss its relation to (1) partially observable MDP; and (2) off-policy RL."""
990,"""Multi-class classification without multi-class labels""","['classification', 'unsupervised learning', 'semi-supervised learning', 'problem reduction', 'weak supervision', 'cross-task', 'learning', 'deep learning', 'neural network']","""This work presents a new strategy for multi-class classification that requires no class-specific labels, but instead leverages pairwise similarity between examples, which is a weaker form of annotation. The proposed method, meta classification learning, optimizes a binary classifier for pairwise similarity prediction and through this process learns a multi-class classifier as a submodule. We formulate this approach, present a probabilistic graphical model for it, and derive a surprisingly simple loss function that can be used to learn neural network-based models. We then demonstrate that this same framework generalizes to the supervised, unsupervised cross-task, and semi-supervised settings. Our method is evaluated against state of the art in all three learning paradigms and shows a superior or comparable accuracy, providing evidence that learning multi-class classification without multi-class labels is a viable learning option.""","""This paper provides a technique to learn multi-class classifiers without multi-class labels, by modeling the multi-class labels as hidden variables and optimizing the likelihood of the input variables and the binary similarity labels. The majority of reviewers voted to accept."""
991,"""Adapting Auxiliary Losses Using Gradient Similarity""","['auxiliary losses', 'transfer learning', 'task similarity', 'deep learning', 'deep reinforcement learning']","""One approach to deal with the statistical inefficiency of neural networks is to rely on auxiliary losses that help to build useful representations. However, it is not always trivial to know if an auxiliary task will be helpful for the main task and when it could start hurting. We propose to use the cosine similarity between gradients of tasks as an adaptive weight to detect when an auxiliary loss is helpful to the main loss. We show that our approach is guaranteed to converge to critical points of the main task and demonstrate the practical usefulness of the proposed algorithm in a few domains: multi-task supervised learning on subsets of ImageNet, reinforcement learning on gridworld, and reinforcement learning on Atari games.""","""This paper tackles the problem of using auxiliary losses to help regularize and aid the learning of a ""goal"" task. The approach proposes avoiding the learning of irrelevant or contradictory details from the auxiliary task at the expense of the ""goal"" tasks by observing cosine similarity between the auxiliary and main tasks and ignore those gradients which are too dissimilar. To justify such a setup one must first show that such negative interference occurs in practice, warranting explicit attention. Then one must show that their algorithm effectively mitigates this interference and at the same time provides some useful signal in combination with the main learning objective. During the review process there was a significant discussion as to whether the proposed approach sufficiently justified its need and usefulness as defined above. One major point of contention is whether to compare against the multi-task literature. The authors claim that prior multi-task learning literature is out of scope of this work since their goal is not to measure performance on all tasks used during learning. However, this claim does not invalidate the reviewer's request for comparison against multi-task learning work. In fact, the authors *should* verify that their method outperforms state-of-the-art multi-task learning methods. Not because they too are studying performance across all tasks, but because their method which knows to prioritize one task during training should certainly outperform the learning paradigms which have no special preference to one of the tasks. A main issue with the current draft centers around the usefulness of the proposed algorithm. First, whether the gradient co-sine similarity is a necessary condition to avoid negative interference and 2) to show at least empirically that auxiliary losses do offer improved performance over optimizing the goal task alone. Based on the experiments now available the answers to these questions remains unclear and thus the paper is not yet recommended for publication."""
992,"""Unsupervised Speech Recognition via Segmental Empirical Output Distribution Matching""","['Unsupervised speech recognition', 'unsupervised learning', 'phoneme classification']","""We consider the problem of training speech recognition systems without using any labeled data, under the assumption that the learner can only access to the input utterances and a phoneme language model estimated from a non-overlapping corpus. We propose a fully unsupervised learning algorithm that alternates between solving two sub-problems: (i) learn a phoneme classifier for a given set of phoneme segmentation boundaries, and (ii) refining the phoneme boundaries based on a given classifier. To solve the first sub-problem, we introduce a novel unsupervised cost function named Segmental Empirical Output Distribution Matching, which generalizes the work in (Liu et al., 2017) to segmental structures. For the second sub-problem, we develop an approximate MAP approach to refining the boundaries obtained from Wang et al. (2017). Experimental results on TIMIT dataset demonstrate the success of this fully unsupervised phoneme recognition system, which achieves a phone error rate (PER) of 41.6%. Although it is still far away from the state-of-the-art supervised systems, we show that with oracle boundaries and matching language model, the PER could be improved to 32.5%. This performance approaches the supervised system of the same model architecture, demonstrating the great potential of the proposed method.""","""This paper is about unsupervised learning for ASR, by matching the acoustic distribution, learned unsupervisedly, with a prior phone-lm distribution. Overall, the results look good on TIMIT. Reviewers agree that this is a well written paper and that it has interesting results. Strengths - Novel formulation for unsupervised ASR, and a non-trivial extension to previously proposed unsupervised classification to segmental level. - Well written, with strong results. Improved results and analysis based on review feedback. Weaknesses - Results are on TIMIT -- a small phone recognition task. - Unclear how it extends to large vocabulary ASR tasks, and tasks that have large scale training data, and RNNs that may learn implicit LMs. The authors propose to deal with this in future work. Overall, the reviewers agree that this is an excellent contribution with strong results. Therefore, it is recommended that the paper be accepted."""
993,"""Simple Black-box Adversarial Attacks""",[],"""The construction of adversarial images is a search problem in high dimensions within a small region around a target image. The goal is to find an imperceptibly modified image that is misclassified by a target model. In the black-box setting, only sporadic feedback is provided through occasional model evaluations. In this paper we provide a new algorithm whose search strategy is based on an intriguingly simple iterative principle: We randomly pick a low frequency component of the discrete cosine transform (DCT) and either add or subtract it to the target image. Model evaluations are only required to identify whether an operation decreases the adversarial loss. Despite its simplicity, the proposed method can be used for targeted and untargeted attacks --- resulting in previously unprecedented query efficiency in both settings. We require a median of 600 black-box model queries (ResNet-50) to produce an adversarial ImageNet image, and we successfully attack Google Cloud Vision with 2500 median queries, averaging to a cost of only $3 per image. We argue that our proposed algorithm should serve as a strong baseline for future adversarial black-box attacks, in particular because it is extremely fast and can be implemented in less than 20 lines of PyTorch code. ""","""The paper considers a procedure for the generation of adversarial examples under a black box setting. The authors claim simplicity as one of the main selling points, with which reviewers agreed, while also noting that the results were impressive or ""promising"". There were concerns over novelty and some confusion over the contribution compared to Guo et al, which I believe has been clarified. The highest confidence reviewer (AnonReviewer2), a researcher with significant expertise in adversarial examples, raised issues of inconsistent threat models (and therefore unfair comparisons regarding query efficiency), missing baselines. A misunderstanding about comparison against a concurrent submission to ICLR 2019 was resolved on the basis that the relevant results are mentioned but not originally presented in the concurrent submission. While I disagree with AnonReviewer2 that results on attacking a particular image from previous work (when run against the Google Cloud Vision API) would be informative, the reviewer has remaining unaddressed concerns about the fairness of comparison (comparing against results reported in previous work rather than re-run in the same setting), and rightly points out that as many variables should be controlled for as possible when making comparisons. Running all methods under the same experimental setting with the same *collection* of query images is therefore appropriate. The authors have not responded to AnonReviewer2's updated post-rebuttal review, and with the remaining sticking point of fairness of comparison with respect to query efficiency I must recommend rejection at this point in time, while noting that all reviewers considered the method promising; I thus would expect to see the method successfully published in the near future once issues of the experimental protocol have been solidified."""
994,"""Decoupled Weight Decay Regularization""","['optimization', 'regularization', 'weight decay', 'Adam']","""L pseudo-formula regularization and weight decay regularization are equivalent for standard stochastic gradient descent (when rescaled by the learning rate), but as we demonstrate this is \emph{not} the case for adaptive gradient algorithms, such as Adam. While common implementations of these algorithms employ L pseudo-formula regularization (often calling it ``weight decay'' in what may be misleading due to the inequivalence we expose), we propose a simple modification to recover the original formulation of weight decay regularization by \emph{decoupling} the weight decay from the optimization steps taken w.r.t. the loss function. We provide empirical evidence that our proposed modification (i) decouples the optimal choice of weight decay factor from the setting of the learning rate for both standard SGD and Adam and (ii) substantially improves Adam's generalization performance, allowing it to compete with SGD with momentum on image classification datasets (on which it was previously typically outperformed by the latter). Our proposed decoupled weight decay has already been adopted by many researchers, and the community has implemented it in TensorFlow and PyTorch; the complete source code for our experiments is available at \url{pseudo-url}""","""Evaluating this paper is somewhat awkward because it has already been through multiple reviewing cycles, and in the meantime, the trick has already become widely adopted and inspired interesting follow-up work. Much of the paper is devoted to reviewing this follow-up work. I think it's clearly time for this to be made part of the published literature, so I recommend acceptance. (And all reviewers are in agreement that the paper ought to be accepted.) The paper proposes, in the context of Adam, to apply literal weight decay in place of L2 regularization. An impressively thorough set of experiments are given to demonstrate the improved generalization performance, as well as a decoupling of the hyperparameters. Previous versions of the paper suffered from a lack of theoretical justification for the proposed method. Ordinarily, in such cases, one would worry that the improved results could be due to some sort of experimental confound. But AdamW has been validated by so many other groups on a range of domains that the improvement is well established. And other researchers have offered possible explanations for the improvement. """
995,"""Recall Traces: Backtracking Models for Efficient Reinforcement Learning""","['Model free RL', 'Variational Inference']","""In many environments only a tiny subset of all states yield high reward. In these cases, few of the interactions with the environment provide a relevant learning signal. Hence, we may want to preferentially train on those high-reward states and the probable trajectories leading to them. To this end, we advocate for the use of a \textit{backtracking model} that predicts the preceding states that terminate at a given high-reward state. We can train a model which, starting from a high value state (or one that is estimated to have high value), predicts and samples which (state, action)-tuples may have led to that high value state. These traces of (state, action) pairs, which we refer to as Recall Traces, sampled from this backtracking model starting from a high value state, are informative as they terminate in good states, and hence we can use these traces to improve a policy. We provide a variational interpretation for this idea and a practical algorithm in which the backtracking model samples from an approximate posterior distribution over trajectories which lead to large rewards. Our method improves the sample efficiency of both on- and off-policy RL algorithms across several environments and tasks. ""","""The paper presents ""recall traces"", a model based approach designed to improve reinforcement learning in sparse reward settings. The approach learns a generative model of trajectories leading to high-reward states, and is subsequently used to augment the real experience collected by the agent. This novel take on combining model-based and model-free learning is conceptually well motivated and is empirically shown to improve sample efficiency on several benchmark tasks. The reviewers noted the following potential weaknesses in their initial reviews: the paper could provide a clearer motivation of why the proposed approach is expected to lead to performance improvements, and how it relates to learning (and uses of) a forward model. Details of the method, e.g., model parameterization is unclear, and the effect of hyperparameter choices is not fully evaluated. The authors provided detailed replies to all reviewer suggestions, and ran extensive new experiments, including experiments to address questions about hyperparameter settings, and an entirely new use of the proposed model in a learning from demonstration setting. The authors also clarified the paper as requested by the reviewers. The reviewers have not responded to the rebuttal, but in the AC's assessment their concerns have been adequately addressed. The reviewers have updated their scores in response to the rebuttal, and the consensus is to accept the paper. The AC notes that the authors seem unaware of related work by Oh et al. ""Self Imitation Learning"" which was published at ICML 2018. The paper is based on a similar conceptual motivation but imitates high-value traces directly, instead of using a generative model. The authors should include a discussion of how their paper relates to this earlier work in their camera ready version."""
996,"""Learned optimizers that outperform on wall-clock and validation loss""","['Learned Optimizers', 'Meta-Learning']","""Deep learning has shown that learned functions can dramatically outperform hand-designed functions on perceptual tasks. Analogously, this suggests that learned update functions may similarly outperform current hand-designed optimizers, especially for specific tasks. However, learned optimizers are notoriously difficult to train and have yet to demonstrate wall-clock speedups over hand-designed optimizers, and thus are rarely used in practice. Typically, learned optimizers are trained by truncated backpropagation through an unrolled optimization process. The resulting gradients are either strongly biased (for short truncations) or have exploding norm (for long truncations). In this work we propose a training scheme which overcomes both of these difficulties, by dynamically weighting two unbiased gradient estimators for a variational loss on optimizer performance. This allows us to train neural networks to perform optimization faster than well tuned first-order methods. Moreover, by training the optimizer against validation loss, as opposed to training loss, we are able to use it to train models which generalize better than those trained by first order methods. We demonstrate these results on problems where our learned optimizer trains convolutional networks in a fifth of the wall-clock time compared to tuned first-order methods, and with an improvement""","""The paper conveys interesting idea but need more work in terms of fair empirical study and also improvement of the writing. The AC based her summary only on the technical argumentation presented by reviewers and authors. """
997,"""A Frank-Wolfe Framework for Efficient and Effective Adversarial Attacks""",[],"""Depending on how much information an adversary can access to, adversarial attacks can be classified as white-box attack and black-box attack. In both cases, optimization-based attack algorithms can achieve relatively low distortions and high attack success rates. However, they usually suffer from poor time and query complexities, thereby limiting their practical usefulness. In this work, we focus on the problem of developing efficient and effective optimization-based adversarial attack algorithms. In particular, we propose a novel adversarial attack framework for both white-box and black-box settings based on the non-convex Frank-Wolfe algorithm. We show in theory that the proposed attack algorithms are efficient with an pseudo-formula convergence rate. The empirical results of attacking Inception V3 model and ResNet V2 model on the ImageNet dataset also verify the efficiency and effectiveness of the proposed algorithms. More specific, our proposed algorithms attain the highest attack success rate in both white-box and black-box attacks among all baselines, and are more time and query efficient than the state-of-the-art.""","""While there was some support for the ideas presented, the majority of the reviewers did not think the submission is ready for publication at ICLR. Significant concerns were raised about clarity of the exposition."""
998,"""Hallucinations in Neural Machine Translation""","['nmt', 'translate', 'dynamics', 'rnn']","""Neural machine translation (NMT) systems have reached state of the art performance in translating text and are in wide deployment. Yet little is understood about how these systems function or break. Here we show that NMT systems are susceptible to producing highly pathological translations that are completely untethered from the source material, which we term hallucinations. Such pathological translations are problematic because they are are deeply disturbing of user trust and easy to find with a simple search. We describe a method to generate hallucinations and show that many common variations of the NMT architecture are susceptible to them. We study a variety of approaches to reduce the frequency of hallucinations, including data augmentation, dynamical systems and regularization techniques, showing that data augmentation significantly reduces hallucination frequency. Finally, we analyze networks that produce hallucinations and show that there are signatures in the attention matrix as well as in the hidden states of the decoder.""","""Strengths - Hallucinations are a problem for seq2seq models, esp trained on small datasets Weankesses - Hallucinations are known to exists, the analyses / observations are not very novel - The considered space of hallucinations source (i.e. added noise) is fairly limited, it is not clear that these are the most natural sources of hallucination and not clear if the methods defined to combat these types would generalize to other types. E.g., I'd rather see hallucinations appearing when running NMT on some natural (albeit noisy) corpus, rather than defining the noise model manually. - The proposed approach is not particularly interesting, and may not be general. Alternative techniques (e.g., modeling coverage) have been proposed in the past. - A wider variety of language pairs, amounts of data, etc needed to validate the methods. This is an empirical paper, I would expect higher quality of evaluation. Two reviewers argued that the baseline system is somewhat weak and the method is not very exciting. """
999,"""Dimensionality Reduction for Representing the Knowledge of Probabilistic Models""","['metric learning', 'distance learning', 'dimensionality reduction', 'bound guarantees']","""Most deep learning models rely on expressive high-dimensional representations to achieve good performance on tasks such as classification. However, the high dimensionality of these representations makes them difficult to interpret and prone to over-fitting. We propose a simple, intuitive and scalable dimension reduction framework that takes into account the soft probabilistic interpretation of standard deep models for classification. When applying our framework to visualization, our representations more accurately reflect inter-class distances than standard visualization techniques such as t-SNE. We show experimentally that our framework improves generalization performance to unseen categories in zero-shot learning. We also provide a finite sample error upper bound guarantee for the method.""","""This paper introduces an approach for reducing the dimensionality of training data examples in a way that preserves information about soft target probabilistic representations provided by a teacher model, with applications such as zero-shot learning and distillation. The authors provide an extensive theoretical and empirical analysis, showing performance improvements in zero shot learning and finite sample error upper bounds. The reviewers generally agree this is a good paper that should be published."""
1000,"""How Important is a Neuron""","['attribution', 'saliency', 'influence']","""The problem of attributing a deep networks prediction to its input/base features is well-studied (cf. Simonyan et al. (2013)). We introduce the notion of conductance to extend the notion of attribution to understanding the importance of hidden units. Informally, the conductance of a hidden unit of a deep network is the flow of attribution via this hidden unit. We can use conductance to understand the importance of a hidden unit to the prediction for a specific input, or over a set of inputs. We justify conductance in multiple ways via a qualitative comparison with other methods, via some axiomatic results, and via an empirical evaluation based on a feature selection task. The empirical evaluations are done using the Inception network over ImageNet data, and a convolutinal network over text data. In both cases, we demonstrate the effectiveness of conductance in identifying interesting insights about the internal workings of these networks.""","""This paper proposes a new measure to quantify the contribution of an individual neuron within a deep neural network. Interpretability and better understanding of the inner workings of neural networks are important questions, and all reviewers agree that this work is contributing an interesting approach and results."""
1001,"""Combinatorial Attacks on Binarized Neural Networks""","['binarized neural networks', 'combinatorial optimization', 'integer programming']","""Binarized Neural Networks (BNNs) have recently attracted significant interest due to their computational efficiency. Concurrently, it has been shown that neural networks may be overly sensitive to ``attacks"" -- tiny adversarial changes in the input -- which may be detrimental to their use in safety-critical domains. Designing attack algorithms that effectively fool trained models is a key step towards learning robust neural networks. The discrete, non-differentiable nature of BNNs, which distinguishes them from their full-precision counterparts, poses a challenge to gradient-based attacks. In this work, we study the problem of attacking a BNN through the lens of combinatorial and integer optimization. We propose a Mixed Integer Linear Programming (MILP) formulation of the problem. While exact and flexible, the MILP quickly becomes intractable as the network and perturbation space grow. To address this issue, we propose IProp, a decomposition-based algorithm that solves a sequence of much smaller MILP problems. Experimentally, we evaluate both proposed methods against the standard gradient-based attack (PGD) on MNIST and Fashion-MNIST, and show that IProp performs favorably compared to PGD, while scaling beyond the limits of the MILP.""","""The paper provides a novel attack method and contributes to evaluating the robustness of neural networks with recently proposed defenses. The evaluation is convincing overall and the authors have answered most questions from the reviewers. We recommend acceptance. """
1002,"""Aggregated Momentum: Stability Through Passive Damping""","['momentum', 'optimization', 'deep learning', 'neural networks']","""Momentum is a simple and widely used trick which allows gradient-based optimizers to pick up speed along low curvature directions. Its performance depends crucially on a damping coefficient. Largecamping coefficients can potentially deliver much larger speedups, but are prone to oscillations and instability; hence one typically resorts to small values such as 0.5 or 0.9. We propose Aggregated Momentum (AggMo), a variant of momentum which combines multiple velocity vectors with different damping coefficients. AggMo is trivial to implement, but significantly dampens oscillations, enabling it to remain stable even for aggressive damping coefficients such as 0.999. We reinterpret Nesterov's accelerated gradient descent as a special case of AggMo and analyze rates of convergence for quadratic objectives. Empirically, we find that AggMo is a suitable drop-in replacement for other momentum methods, and frequently delivers faster convergence with little to no tuning.""","""Dear authors, Reviewers liked the idea of your new optimizer and found the experiments convincing. However, they also would have liked to get better insights on the place of AggMo in the existing optimization literature. Given that the related work section is quite small, I encourage you to expand it based on the works mentioned in the reviews."""
1003,"""Nested Dithered Quantization for Communication Reduction in Distributed Training""","['machine learning', 'distributed training', 'dithered quantization', 'nested quantization', 'distributed compression']","""In distributed training, the communication cost due to the transmission of gradients or the parameters of the deep model is a major bottleneck in scaling up the number of processing nodes. To address this issue, we propose dithered quantization for the transmission of the stochastic gradients and show that training with Dithered Quantized Stochastic Gradients (DQSG) is similar to the training with unquantized SGs perturbed by an independent bounded uniform noise, in contrast to the other quantization methods where the perturbation depends on the gradients and hence, complicating the convergence analysis. We study the convergence of training algorithms using DQSG and the trade off between the number of quantization levels and the training time. Next, we observe that there is a correlation among the SGs computed by workers that can be utilized to further reduce the communication overhead without any performance loss. Hence, we develop a simple yet effective quantization scheme, nested dithered quantized SG (NDQSG), that can reduce the communication significantly without requiring the workers communicating extra information to each other. We prove that although NDQSG requires significantly less bits, it can achieve the same quantization variance bound as DQSG. Our simulation results confirm the effectiveness of training using DQSG and NDQSG in reducing the communication bits or the convergence time compared to the existing methods without sacrificing the accuracy of the trained model.""","""The reviewers found that the paper needs more compelling empirical study."""
1004,"""A Walk with SGD: How SGD Explores Regions of Deep Network Loss?""",[],"""The non-convex nature of the loss landscape of deep neural networks (DNN) lends them the intuition that over the course of training, stochastic optimization algorithms explore different regions of the loss surface by entering and escaping many local minima due to the noise induced by mini-batches. But is this really the case? This question couples the geometry of the DNN loss landscape with how stochastic optimization algorithms like SGD interact with it during training. Answering this question may help us qualitatively understand the dynamics of deep neural network optimization. We show evidence through qualitative and quantitative experiments that mini-batch SGD rarely crosses barriers during DNN optimization. As we show, the mini-batch induced noise helps SGD explore different regions of the loss surface using a seemingly different mechanism. To complement this finding, we also investigate the qualitative reason behind the slowing down of this exploration when using larger batch-sizes. We show this happens because gradients from larger batch-sizes align more with the top eigenvectors of the Hessian, which makes SGD oscillate in the proximity of the parameter initialization, thus preventing exploration.""","""The reviewers agree that the paper needs significantly more work to improve presentation and is not fully empirically and conceptually convincing."""
1005,"""RotDCF: Decomposition of Convolutional Filters for Rotation-Equivariant Deep Networks""",[],"""Explicit encoding of group actions in deep features makes it possible for convolutional neural networks (CNNs) to handle global deformations of images, which is critical to success in many vision tasks. This paper proposes to decompose the convolutional filters over joint steerable bases across the space and the group geometry simultaneously, namely a rotation-equivariant CNN with decomposed convolutional filters (RotDCF). This decomposition facilitates computing the joint convolution, which is proved to be necessary for the group equivariance. It significantly reduces the model size and computational complexity while preserving performance, and truncation of the bases expansion serves implicitly to regularize the filters. On datasets involving in-plane and out-of-plane object rotations, RotDCF deep features demonstrate greater robustness and interpretability than regular CNNs. The stability of the equivariant representation to input variations is also proved theoretically. The RotDCF framework can be extended to groups other than rotations, providing a general approach which achieves both group equivariance and representation stability at a reduced model size.""","""This paper builds on the recent DCFNet (Decomposed Convolutional Filters) architecture to incorporate rotation equivariance while preserving stability. The core idea is to decompose the trainable filters into a steerable representation and learn over a subset of the coefficients of that representation. Reviewers all agreed that this is a solid contribution that advances research into group equivariant CNNs, bringing efficiency gains and stability guarantees, albeit these appear to be incremental with respect to the techniques developed in the DCFNet work. In summary, the AC believes this to be a valuable contribution and therefore recommends acceptance. """
1006,"""Adaptive Pruning of Neural Language Models for Mobile Devices""","['Inference-time pruning', 'Neural Language Models']","""Neural language models (NLMs) exist in an accuracy-efficiency tradeoff space where better perplexity typically comes at the cost of greater computation complexity. In a software keyboard application on mobile devices, this translates into higher power consumption and shorter battery life. This paper represents the first attempt, to our knowledge, in exploring accuracy-efficiency tradeoffs for NLMs. Building on quasi-recurrent neural networks (QRNNs), we apply pruning techniques to provide a ""knob"" to select different operating points. In addition, we propose a simple technique to recover some perplexity using a negligible amount of memory. Our empirical evaluations consider both perplexity as well as energy consumption on a Raspberry Pi, where we demonstrate which methods provide the best perplexity-power consumption operating point. At one operating point, one of the techniques is able to provide energy savings of 40% over the state of the art with only a 17% relative increase in perplexity.""","""The area chair agrees with the authors and the reviewers that the topic of this work is relevant and important. The area chair however shares the concerns of the reviewers about the setup and the empirical evaluation: - Having one model that can be pruned to varying sizes at run-time is convenient, but in practice it is likely to be OK to do the pruning at training time. In light of this, the empirical results are not so impressive. - Without quantization, distillation and fused ops, the value of the empirical results seems questionable as these are important and well-known techniques that are often used in practice. A more thorough evaluation that includes these techniques would make the paper much stronger."""
1007,"""Understanding the Effectiveness of Lipschitz-Continuity in Generative Adversarial Nets""","['GANs', 'Lipschitz-continuity', 'convergence']","""In this paper, we investigate the underlying factor that leads to the failure and success in training of GANs. Specifically, we study the property of the optimal discriminative function pseudo-formula and show that pseudo-formula in most GANs can only reflect the local densities at pseudo-formula , which means the value of pseudo-formula for points in the fake distribution ( pseudo-formula ) does not contain any information useful about the location of other points in the real distribution ( pseudo-formula ). Given that the supports of the real and fake distributions are usually disjoint, we argue that such a pseudo-formula and its gradient tell nothing about ""how to pull pseudo-formula to pseudo-formula "", which turns out to be the fundamental cause of failure in training of GANs. We further demonstrate that a well-defined distance metric (including the dual form of Wasserstein distance with a compacted constraint) does not necessarily ensure the convergence of GANs. Finally, we propose Lipschitz-continuity condition as a general solution and show that in a large family of GAN objectives, Lipschitz condition is capable of connecting pseudo-formula and pseudo-formula through pseudo-formula such that the gradient pseudo-formula at each sample \sim P_g points towards some real sample \sim P_r$.""","""The paper investigates problems that can arise for a certain version of the dual form of the Wasserstein distance, which is proved in Appendix I. While the theoretical analysis seems correct, the significance of the distribution is limited by the fact, that the specific dual form analysed is not commonly used in other works. Furthermore, the assumption that the optimal function is differentiable is often not fulfilled neither. The paper would herefore be significantly strengthen by making more clear to which methods used in practice the insights carry over. """
1008,"""Overcoming the Disentanglement vs Reconstruction Trade-off via Jacobian Supervision""","['disentangling', 'autoencoders', 'jacobian', 'face manipulation']","""A major challenge in learning image representations is the disentangling of the factors of variation underlying the image formation. This is typically achieved with an autoencoder architecture where a subset of the latent variables is constrained to correspond to specific factors, and the rest of them are considered nuisance variables. This approach has an important drawback: as the dimension of the nuisance variables is increased, image reconstruction is improved, but the decoder has the flexibility to ignore the specified factors, thus losing the ability to condition the output on them. In this work, we propose to overcome this trade-off by progressively growing the dimension of the latent code, while constraining the Jacobian of the output image with respect to the disentangled variables to remain the same. As a result, the obtained models are effective at both disentangling and reconstruction. We demonstrate the applicability of this method in both unsupervised and supervised scenarios for learning disentangled representations. In a facial attribute manipulation task, we obtain high quality image generation while smoothly controlling dozens of attributes with a single model. This is an order of magnitude more disentangled factors than state-of-the-art methods, while obtaining visually similar or superior results, and avoiding adversarial training.""","""The paper proposes a new way to tackle the trade-off between disentanglement and reconstruction, by training a teacher autoencoder that learns to disentangle, then distilling into a student model. The distillation is encouraged with a loss term that constrains the Jacobian in an interesting way. The qualitative results with image manipulation are interesting and the general idea seems to be well-liked by the reviewers (and myself). The main weaknesses of the paper seem to be in the evaluation. Disentanglement is not exactly easy to measure as such. But overall the various ablation studies do show that the Jacobian regularization term improves meaningfully over Fader nets. Given the quality of the results and the fact that this work moves the needle in an important (albeit hard to define) area of learning disentangled representations, I think would be a good piece of work to present at ICLR so I recommend acceptance."""
1009,"""Unsupervised Learning of Sentence Representations Using Sequence Consistency""","['sentence representation', 'unsupervised learning', 'LSTM']","""Computing universal distributed representations of sentences is a fundamental task in natural language processing. We propose ConsSent, a simple yet surprisingly powerful unsupervised method to learn such representations by enforcing consistency constraints on sequences of tokens. We consider two classes of such constraints sequences that form a sentence and between two sequences that form a sentence when merged. We learn sentence encoders by training them to distinguish between consistent and inconsistent examples, the latter being generated by randomly perturbing consistent examples in six different ways. Extensive evaluation on several transfer learning and linguistic probing tasks shows improved performance over strong unsupervised and supervised baselines, substantially surpassing them in several cases. Our best results are achieved by training sentence encoders in a multitask setting and by an ensemble of encoders trained on the individual tasks.""","""The overall view of the reviewers is that the paper is not quite good enough as it stands. The reviewers also appreciates the contributions so taking the comments into account and resubmit elsewhere is encouraged. """
1010,"""CAMOU: Learning Physical Vehicle Camouflages to Adversarially Attack Detectors in the Wild""","['Adversarial Attack', 'Object Detection', 'Synthetic Simulation']","""In this paper, we conduct an intriguing experimental study about the physical adversarial attack on object detectors in the wild. In particular, we learn a camouflage pattern to hide vehicles from being detected by state-of-the-art convolutional neural network based detectors. Our approach alternates between two threads. In the first, we train a neural approximation function to imitate how a simulator applies a camouflage to vehicles and how a vehicle detector performs given images of the camouflaged vehicles. In the second, we minimize the approximated detection score by searching for the optimal camouflage. Experiments show that the learned camouflage can not only hide a vehicle from the image-based detectors under many test cases but also generalizes to different environments, vehicles, and object detectors.""","""This work develops a method for learning camouflage patterns that could be painted onto a 3d object in order to reliably fool an image-based object detector. Experiments are conducted in a simulated environment. All reviewers agree that the problem and approach are interesting. Reviewers 1 and 3 are highly positive, while Reviewer 2 believes that real-world experiments are necessary to substantiate the claims of the paper. While such experiments would certainly enhance the impact of the work, I agree with Reviewers 1 and 3 that the current approach is sufficiently interesting and well-developed on its own."""
1011,"""Cross-Entropy Loss Leads To Poor Margins""","['Cross-entropy loss', 'Binary classification', 'Low-rank features', 'Adversarial examples', 'Differential training']","""Neural networks could misclassify inputs that are slightly different from their training data, which indicates a small margin between their decision boundaries and the training dataset. In this work, we study the binary classification of linearly separable datasets and show that linear classifiers could also have decision boundaries that lie close to their training dataset if cross-entropy loss is used for training. In particular, we show that if the features of the training dataset lie in a low-dimensional affine subspace and the cross-entropy loss is minimized by using a gradient method, the margin between the training points and the decision boundary could be much smaller than the optimal value. This result is contrary to the conclusions of recent related works such as (Soudry et al., 2018), and we identify the reason for this contradiction. In order to improve the margin, we introduce differential training, which is a training paradigm that uses a loss function defined on pairs of points from each class. We show that the decision boundary of a linear classifier trained with differential training indeed achieves the maximum margin. The results reveal the use of cross-entropy loss as one of the hidden culprits of adversarial examples and introduces a new direction to make neural networks robust against them.""","""The paper challenges claims about cross-entropy loss attaining max margin when applied to linear classifier and linearly separable data. This is important in moving forward with the development of better loss functions. The main criticism of the paper is that the results are incremental and can be easily obtained from previous work. The authors expressed certain concerns about the reviewing process. In the interest of dissipating any doubts, we collected two additional referee reports. Although one referee is positive about the paper, four other referees agree that the paper is not strong enough. """
1012,"""M^3RL: Mind-aware Multi-agent Management Reinforcement Learning""","['Multi-agent Reinforcement Learning', 'Deep Reinforcement Learning']","""Most of the prior work on multi-agent reinforcement learning (MARL) achieves optimal collaboration by directly learning a policy for each agent to maximize a common reward. In this paper, we aim to address this from a different angle. In particular, we consider scenarios where there are self-interested agents (i.e., worker agents) which have their own minds (preferences, intentions, skills, etc.) and can not be dictated to perform tasks they do not want to do. For achieving optimal coordination among these agents, we train a super agent (i.e., the manager) to manage them by first inferring their minds based on both current and past observations and then initiating contracts to assign suitable tasks to workers and promise to reward them with corresponding bonuses so that they will agree to work together. The objective of the manager is to maximize the overall productivity as well as minimize payments made to the workers for ad-hoc worker teaming. To train the manager, we propose Mind-aware Multi-agent Management Reinforcement Learning (M^3RL), which consists of agent modeling and policy learning. We have evaluated our approach in two environments, Resource Collection and Crafting, to simulate multi-agent management problems with various task settings and multiple designs for the worker agents. The experimental results have validated the effectiveness of our approach in modeling worker agents' minds online, and in achieving optimal ad-hoc teaming with good generalization and fast adaptation.""","""The paper addresses a variant of multi-agent reinforcement learning that aligns well with real-world applications - it considers the case where agents may have individual, diverging preferences. The proposed approach trains a ""manager"" agent which coordinates the self-interested worker agents by assigning them appropriate tasks and rewarding successful task completion (through contract generation). The approach is empirically validated on two grid-world domains: resource collection and crafting. The reviewers point out that this formulation is closely related to the principle-agent problem known in the economics literature, and see a key contribution of the paper in bringing this type of problem into the deep RL space. The reviewers noted several potential weaknesses: They asked to clarify the relation to prior work, especially on the principle-agents work done in other areas, as well as connections to real world applications. In this context, they also noted that the significance of the contribution was unclear. Several modeling choices were questioned, including the choice of using rule-based agents for the empirical results presented in the main paper, and the need for using deep learning for contract generation. They asked the authors to provide additional details regarding scalability and sample complexity of the approach. The authors carefully addressed the reviewer concerns, and the reviewers have indicated that they are satisfied with the response and updates to the paper. The consensus is to accept the paper. The AC is particularly pleased to see that the authors plan to open source their code so that experiments can be replicated, and encourages them to do so in a timely manner. The AC also notes that the figures in the paper are very small, and often not readable in print - please increase figure and font sizes in the camera ready version to ensure the paper is legible when printed."""
1013,"""A new dog learns old tricks: RL finds classic optimization algorithms""","['reinforcement learning', 'algorithms', 'adwords', 'knapsack', 'secretary']","""This paper introduces a novel framework for learning algorithms to solve online combinatorial optimization problems. Towards this goal, we introduce a number of key ideas from traditional algorithms and complexity theory. First, we draw a new connection between primal-dual methods and reinforcement learning. Next, we introduce the concept of adversarial distributions (universal and high-entropy training sets), which are distributions that encourage the learner to find algorithms that work well in the worst case. We test our new ideas on a number of optimization problem such as the AdWords problem, the online knapsack problem, and the secretary problem. Our results indicate that the models have learned behaviours that are consistent with the traditional optimal algorithms for these problems.""","""This paper is concerned with solving Online Combinatorial Optimization (OCO) problems using reinforcement learning (RL). There is a well-established traditional family of approaches to solving OCO problems, therefore the attempt itself to solve them with RL is very intriguing, as this provides insights about the capabilities of RL in a new but at the same time well understood class of problems. The reviewers agree that this approach is not entirely new. While past similar efforts take away some of the novelty of this paper, the reviewers and AC believe that still the setting considered here contains novel and interesting elements. All reviewers were unconvinced that this work can provide strong claims about using RL to learn any primal-dual algorithm. This takes away some of the papers impact, but thanks to discussion the authors managed to clarify some hand-wavy claims and toned-down the claims that were not convincing. Therefore, it was agreed that the new revision still provides some useful insight into the RL and primal-dual connection, even without a complete formal connection. """
1014,"""Exploration by Uncertainty in Reward Space""","['Policy Exploration', 'Uncertainty in Reward Space']","""Efficient exploration plays a key role in reinforcement learning tasks. Commonly used dithering strategies, such as-greedy, try to explore the action-state space randomly; this can lead to large demand for samples. In this paper, We propose an exploration method based on the uncertainty in reward space. There are two policies in this approach, the exploration policy is used for exploratory sampling in the environment, then the benchmark policy try to update by the data proven by the exploration policy. Benchmark policy is used to provide the uncertainty in reward space, e.g. td-error, which guides the exploration policy updating. We apply our method on two grid-world environments and four Atari games. Experiment results show that our method improves learning speed and have a better performance than baseline policies""","""The paper has some nice ideas for efficient exploration, but reviewers think more work is needed before it is ready for publication. In particular, the paper should have an improved discussion of state-of-the-art work on exploration, compare the difference and benefits of the proposed approach, and then conduct proper experiments to validate the claims."""
1015,"""Rethinking learning rate schedules for stochastic optimization""","['SGD', 'learning rate', 'step size schedules', 'stochastic approximation', 'stochastic optimization', 'deep learning', 'non-convex optimization', 'stochastic gradient descent']","""There is a stark disparity between the learning rate schedules used in the practice of large scale machine learning and what are considered admissible learning rate schedules prescribed in the theory of stochastic approximation. Recent results, such as in the 'super-convergence' methods which use oscillating learning rates, serve to emphasize this point even more. One plausible explanation is that non-convex neural network training procedures are better suited to the use of fundamentally different learning rate schedules, such as the ``cut the learning rate every constant number of epochs'' method (which more closely resembles an exponentially decaying learning rate schedule); note that this widely used schedule is in stark contrast to the polynomial decay schemes prescribed in the stochastic approximation literature, which are indeed shown to be (worst case) optimal for classes of convex optimization problems. The main contribution of this work shows that the picture is far more nuanced, where we do not even need to move to non-convex optimization to show other learning rate schemes can be far more effective. In fact, even for the simple case of stochastic linear regression with a fixed time horizon, the rate achieved by any polynomial decay scheme is sub-optimal compared to the statistical minimax rate (by a factor of condition number); in contrast the ```''cut the learning rate every constant number of epochs'' provides an exponential improvement (depending only logarithmically on the condition number) compared to any polynomial decay scheme. Finally, it is important to ask if our theoretical insights are somehow fundamentally tied to quadratic loss minimization (where we have circumvented minimax lower bounds for more general convex optimization problems)? Here, we conjecture that recent results which make the gradient norm small at a near optimal rate, for both convex and non-convex optimization, may also provide more insights into learning rate schedules used in practice. ""","""R4 recommends acceptance while R2 is lukewarm and R1 argues for rejection to revise the presentation of the paper. As we unfortunately need to reject borderline papers given the space constraints, the AC recommends ""revise and resubmit""."""
1016,"""Learning Hash Codes via Hamming Distance Targets""","['information retrieval', 'learning to hash', 'cbir']","""We present a powerful new loss function and training scheme for learning binary hash codes with any differentiable model and similarity function. Our loss function improves over prior methods by using log likelihood loss on top of an accurate approximation for the probability that two inputs fall within a Hamming distance target. Our novel training scheme obtains a good estimate of the true gradient by better sampling inputs and evaluating loss terms between all pairs of inputs in each minibatch. To fully leverage the resulting hashes, we use multi-indexing. We demonstrate that these techniques provide large improvements to a similarity search tasks. We report the best results to date on competitive information retrieval tasks for Imagenet and SIFT 1M, improving recall from 73% to 85% and reducing query cost by a factor of 2-8, respectively.""","""The paper proposes learning a hash function that maps high dimensional data to binary codes, and uses multi-index hashing for efficient retrieval. The paper discusses similar results to ""Similarity estimation techniques from rounding algorithms, M Charikar, 2002"" without citing this paper. The proposed learning idea is also similar to ""Binary Reconstructive Embedding, B. Kulis, T. Darrell, NIPS'09"" without citation. Please study the learning to hash literature and discuss the similarities and differences with your approach. Due to missing citations and lack of novelty, I believe the paper does not pass the bar for acceptance at ICLR. PS: PQ and its better variants (optimized PQ and cartesian k-means) are from a different family of quantization techniques as pointed out by R3 and multi-index hashing is not directly applicable to such techniques. Regardless, I am also surprised that your technique just using hamming distance is able to outperform PQ using lookup table distance. """
1017,"""Context Mover's Distance & Barycenters: Optimal transport of contexts for building representations""","['representation learning', 'wasserstein distance', 'wasserstein barycenter', 'entailment']","""We propose a unified framework for building unsupervised representations of entities and their compositions, by viewing each entity as a histogram (or distribution) over its contexts. This enables us to take advantage of optimal transport and construct representations that effectively harness the geometry of the underlying space containing the contexts. Our method captures uncertainty via modelling the entities as distributions and simultaneously provides interpretability with the optimal transport map, hence giving a novel perspective for building rich and powerful feature representations. As a guiding example, we formulate unsupervised representations for text, and demonstrate it on tasks such as sentence similarity and word entailment detection. Empirical results show strong advantages gained through the proposed framework. This approach can potentially be used for any unsupervised or supervised problem (on text or other modalities) with a co-occurrence structure, such as any sequence data. The key tools at the core of this framework are Wasserstein distances and Wasserstein barycenters.""","""The paper proposes to build word representation based on a histogram over context word vectors, allowing them to measure distances between words in terms of optimal transport between these histograms. An empirical analysis shows that the proposed approach is competitive with others on semantic textual similarity and hypernym detection tasks. While the idea is definitely interesting, the paper would be streghten by a more extensive empirical analysis. """
1018,"""Interactive Agent Modeling by Learning to Probe""","['Agent Modeling', 'Theory of Mind', 'Deep Reinforcement Learning', 'Multi-agent Reinforcement Learning']","""The ability of modeling the other agents, such as understanding their intentions and skills, is essential to an agent's interactions with other agents. Conventional agent modeling relies on passive observation from demonstrations. In this work, we propose an interactive agent modeling scheme enabled by encouraging an agent to learn to probe. In particular, the probing agent (i.e. a learner) learns to interact with the environment and with a target agent (i.e., a demonstrator) to maximize the change in the observed behaviors of that agent. Through probing, rich behaviors can be observed and are used for enhancing the agent modeling to learn a more accurate mind model of the target agent. Our framework consists of two learning processes: i) imitation learning for an approximated agent model and ii) pure curiosity-driven reinforcement learning for an efficient probing policy to discover new behaviors that otherwise can not be observed. We have validated our approach in four different tasks. The experimental results suggest that the agent model learned by our approach i) generalizes better in novel scenarios than the ones learned by passive observation, random probing, and other curiosity-driven approaches do, and ii) can be used for enhancing performance in multiple applications including distilling optimal planning to a policy net, collaboration, and competition. A video demo is available at pseudo-url""","""The submission proposes a setting of two agents, one of them probing the other (the latter being the ""demonstrator""). The probing is done in a way that learns to imitate the expert's behavior, with some curiosity-driven reward that maximizes the chance that the probing agents has the expert do trajectories that the probing agent hasn't seen before. All the reviewers found the idea and experiments interesting. The major concern is whether the setup and the environments are too contrived. At least 2 reviewers commented on the fact that the environments/dataset seemed engineered for success of the given method, which is a concern about how this method would generalize to something other than the proposed setup. I also share the concern with R3 regarding the practicality of the proposed method: it is not obvious to me what problems this would actually be *useful* for, given that the method requires online interaction with an expert agent in order to succeed. The space of such scenarios where we can continuously probe an expert agent many many times for free/cheap is very small and frankly I'm not entirely sure why you would need to do imitation learning in that case at all (if the method was shown to work using only a state, rather than requiring a state/action pair from the expert, then maybe it'd be more useful). It's a tough call, but despite the nice results and interesting ideas, I think the method lacks generality and practical utility/significance and thus at this point I cannot recommend acceptance in its current form."""
1019,"""Improved resistance of neural networks to adversarial images through generative pre-training""","['adversarial images', 'Boltzmann machine', 'mean field approximation']","""We train a feed forward neural network with increased robustness against adversarial attacks compared to conventional training approaches. This is achieved using a novel pre-trained building block based on a mean field description of a Boltzmann machine. On the MNIST dataset the method achieves strong adversarial resistance without data augmentation or adversarial training. We show that the increased adversarial resistance is correlated with the generative performance of the underlying Boltzmann machine.""","""No reviewer has made a strong case for accepting this paper or championed it so I am recommending rejecting it. The unfavorable reviewers, although they mention real issues, have not highlighted some of the most important barriers to accepting this work. One major, but not necessarily dispositive, concern is that the paper only presents results on MNIST. However, even if we put aside this concern, there are several issues with the motivation and approach of this paper. If this technique is actually good at improving the model outside the clean image distribution, then the paper should show that and not just L2 worst case perturbations. To quote the intro of the paper: ""How can deep learning systems successfully generalise and at the same time be extremely vulnerable to minute changes in the input?"" The answer is: they don't generalize and this work does not show us improved generalization. Even a small amount of test error in the data distribution suggests that the closest test error to a given point will often be quite close to the starting point, although this is easier to see with linear models. The best way to fix this work would be to study (average case) error on noisy distributions (as in the concurrent submission pseudo-url )."""
1020,"""Convolutional CRFs for Semantic Segmentation""","['conditional random fields', 'semantic segmentation', 'computer vision', 'structured learning']","""For the challenging semantic image segmentation task the best performing models have traditionally combined the structured modelling capabilities of Conditional Random Fields (CRFs) with the feature extraction power of CNNs. In more recent works however, CRF post-processing has fallen out of favour. We argue that this is mainly due to the slow training and inference speeds of CRFs, as well as the difficulty of learning the internal CRF parameters. To overcome both issues we propose to add the assumption of conditional independence to the framework of fully-connected CRFs. This allows us to reformulate the inference in terms of convolutions, which can be implemented highly efficiently on GPUs.Doing so speeds up inference and training by two orders of magnitude. All parameters of the convolutional CRFs can easily be optimized using backpropagation. Towards the goal of facilitating further CRF research we have made our implementations publicly available.""","""The authors replace the large filtering step in the permutohedral lattice with a spatially varying convolutional kernel. They show that inference is more efficient and training is easier. In practice, the synthetic experiments seem to show a greater improvement than appears in real data. There are concerns about the clarity, lack of theoretical proofs, and at times overstated claims that do not have sufficient support. The ratings before the rebuttal and discussion were 7-4-6. After, R1 adjusted their score from 6 to 4. R2 initially gave a 7 but later said ""I think the authors missed an opportunity here. I rated it as an accept, because I saw what it could have been after a good revision. The core idea is good, but fully agree with R1 and R3 that the paper needs work (which the authors were not willing to do). I checked the latest revision (as of Monday morning). None of R3's writing/claims issues are fixed, neither were my additional experimental requests, not even R1's typos."" There is therefore a consensus among reviewers for reject. """
1021,"""Multi-Scale Stacked Hourglass Network for Human Pose Estimation""","['Human pose estimation', 'Hourglass network', 'Multi-scale analysis']","""Stacked hourglass network has become an important model for Human pose estimation. The estimation of human body posture depends on the global information of the keypoints type and the local information of the keypoints location. The consistent processing of inputs and constraints makes it difficult to form differentiated and determined collaboration mechanisms for each stacked hourglass network. In this paper, we propose a Multi-Scale Stacked Hourglass (MSSH) network to high-light the differentiation capabilities of each Hourglass network for human pose estimation. The pre-processing network forms feature maps of different scales,and dispatch them to various locations of the stack hourglass network, where the small-scale features reach the front of stacked hourglass network, and large-scale features reach the rear of stacked hourglass network. And a new loss function is proposed for multi-scale stacked hourglass network. Different keypoints have different weight coefficients of loss function at different scales, and the keypoints weight coefficients are dynamically adjusted from the top-level hourglass network to the bottom-level hourglass network. Experimental results show that the pro-posed method is competitive with respect to the comparison algorithm on MPII and LSP datasets.""","""The paper presents a multi-scale extension of the hourglass network. As the reviewers point out, the paper is below ICLR publication standard due to low novelty (i.e., multi-scale extension is not a new idea) and significance (i.e., not a significant performance gain against the state-of-the-art method or other baselines)."""
1022,"""GANSynth: Adversarial Neural Audio Synthesis""","['GAN', 'Audio', 'WaveNet', 'NSynth', 'Music']","""Efficient audio synthesis is an inherently difficult machine learning task, as human perception is sensitive to both global structure and fine-scale waveform coherence. Autoregressive models, such as WaveNet, model local structure at the expense of global latent structure and slow iterative sampling, while Generative Adversarial Networks (GANs), have global latent conditioning and efficient parallel sampling, but struggle to generate locally-coherent audio waveforms. Herein, we demonstrate that GANs can in fact generate high-fidelity and locally-coherent audio by modeling log magnitudes and instantaneous frequencies with sufficient frequency resolution in the spectral domain. Through extensive empirical investigations on the NSynth dataset, we demonstrate that GANs are able to outperform strong WaveNet baselines on automated and human evaluation metrics, and efficiently generate audio several orders of magnitude faster than their autoregressive counterparts. ""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - novel approach to audio synthesis - strong qualitative and quantitative results - extensive evaluation 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - small grammatical issues (mostly resolved in the revision). 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. No major points of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be accepted. """
1023,"""Analysing Mathematical Reasoning Abilities of Neural Models""","['mathematics', 'dataset', 'algebraic', 'reasoning']","""Mathematical reasoning---a core ability within human intelligence---presents some unique challenges as a domain: we do not come to understand and solve mathematical problems primarily on the back of experience and evidence, but on the basis of inferring, learning, and exploiting laws, axioms, and symbol manipulation rules. In this paper, we present a new challenge for the evaluation (and eventually the design) of neural architectures and similar system, developing a task suite of mathematics problems involving sequential questions and answers in a free-form textual input/output format. The structured nature of the mathematics domain, covering arithmetic, algebra, probability and calculus, enables the construction of training and test spits designed to clearly illuminate the capabilities and failure-modes of different architectures, as well as evaluate their ability to compose and relate knowledge and learned processes. Having described the data generation process and its potential future expansions, we conduct a comprehensive analysis of models from two broad classes of the most powerful sequence-to-sequence architectures and find notable differences in their ability to resolve mathematical problems and generalize their knowledge. """,""" Pros: - A useful and well-structured dataset which will be of use to the community - Well-written and clear (though see Reviewer 2's comment concerning the clarity of the model description section) - Good methodology Cons: - There is a question about why a new dataset is needed rather than a combination of previous datasets and also why these datasets couldn't be harvested from school texts directly. Presumably it would've been a lot more work but please address the issue in your rebuttal. - Evaluation: Reviewer 3 is concerned that the evaluation should perhaps have included more mathematics-specific models (a couple of which are mentioned in the text). On the other hand, Reviewer 2 is concerned that the specific choices (e.g. ""thinking steps"") made for the general models are non-standard in seq-2-seq models. I haven't heard about the thinking step approach but perhaps it's out there somewhere. It would be helpful generally to have more discussion about the reasoning involved in these decisions. I think this is a useful contribution to the community, well written and thoughtfully constructed. I am tentatively accepting this paper with the understanding that you will engage directly with the reviewers to address their concerns about the evaluation section. Please in particular use the rebuttal period to focus on the clarity of the model description and the motivation for the particular models chosen. Also consider adding additional experiments to allay the concerns of the reviewers."""
1024,"""Coarse-grain Fine-grain Coattention Network for Multi-evidence Question Answering""","['question answering', 'reading comprehension', 'nlp', 'natural language processing', 'attention', 'representation learning']","""End-to-end neural models have made significant progress in question answering, however recent studies show that these models implicitly assume that the answer and evidence appear close together in a single document. In this work, we propose the Coarse-grain Fine-grain Coattention Network (CFC), a new question answering model that combines information from evidence across multiple documents. The CFC consists of a coarse-grain module that interprets documents with respect to the query then finds a relevant answer, and a fine-grain module which scores each candidate answer by comparing its occurrences across all of the documents with the query. We design these modules using hierarchies of coattention and self-attention, which learn to emphasize different parts of the input. On the Qangaroo WikiHop multi-evidence question answering task, the CFC obtains a new state-of-the-art result of 70.6% on the blind test set, outperforming the previous best by 3% accuracy despite not using pretrained contextual encoders.""","""The paper presents a method for coarse and fine inference for question answering. It originally measured performance only on WikiHop and then later added experiments on TriviaQA. The results are good. One of the concerns regarding the paper was the novelty of the work, and lack of enough experiments. However, the addition of TriviaQA results allays some of that concern. I'd suggest citing the paper by Swayamdipta et al from last year that attempted coarse to fine inference for TriviaQA: Multi-Mention Learning for Reading Comprehension with Neural Cascades. Swabha Swayamdipta, Ankur P. Parikh and Tom Kwiatkowski. Proceedings of ICLR 2018. Overall, there is relative consensus that the paper is good with a new method and some strong results."""
1025,"""Trajectory VAE for multi-modal imitation""","['imitation learning', 'latent variable model', 'variational autoencoder', 'diverse behaviour']","""We address the problem of imitating multi-modal expert demonstrations in sequential decision making problems. In many practical applications, for example video games, behavioural demonstrations are readily available that contain multi-modal structure not captured by typical existing imitation learning approaches. For example, differences in the observed players' behaviours may be representative of different underlying playstyles. In this paper, we use a generative model to capture different emergent playstyles in an unsupervised manner, enabling the imitation of a diverse range of distinct behaviours. We utilise a variational autoencoder to learn an embedding of the different types of expert demonstrations on the trajectory level, and jointly learn a latent representation with a policy. In experiments on a range of 2D continuous control problems representative of Minecraft environments, we empirically demonstrate that our model can capture a multi-modal structured latent space from the demonstrated behavioural trajectories. ""","""The paper considers the problem of imitating multi-modal expert demonstrations using a variational auto-encoder to embed demonstrated trajectories into a structured latent space. The problem is important, and the paper is well written. The model is shown to work well on toy examples. However, as pointed out by the reviewers, given that multi-modal has been studied before, the approach should have been compared both in theory and in practice to existing methods and baselines (e.g., InfoGAIL). Furthermore, the technical contribution is somewhat limited as it using an existing model on a new application domain."""
1026,"""Generative Feature Matching Networks""","['Generative Deep Neural Networks', 'Feature Matching', 'Maximum Mean Discrepancy', 'Generative Adversarial Networks']","""We propose a non-adversarial feature matching-based approach to train generative models. Our approach, Generative Feature Matching Networks (GFMN), leverages pretrained neural networks such as autoencoders and ConvNet classifiers to perform feature extraction. We perform an extensive number of experiments with different challenging datasets, including ImageNet. Our experimental results demonstrate that, due to the expressiveness of the features from pretrained ImageNet classifiers, even by just matching first order statistics, our approach can achieve state-of-the-art results for challenging benchmarks such as CIFAR10 and STL10.""","""The paper proposes a method of training implicit generative models based on moment matching in the feature spaces of pre-trained feature extractors, derived from autoencoders or classifiers. The authors also propose a trick for tracking the moving averages by appealing to the Adam optimizer and deriving updates based on the implied loss function of a moving average update. It was generally agreed that the paper was well written and easy to follow, that empirical results were good, but that the novelty is relatively low. Generative models have been built out of pre-trained classifiers before (e.g. generative plug & play networks), feature matching losses for generator networks have been proposed before (e.g. Salimans et al, 2016). The contribution here is mainly the extensive empirical analysis plus the AMA trick. After receiving exclusively confidence score 3 reviews, I sought the opinion of a 4th reviewer, an expert on GANs and GAN-like generative models. Their remaining sticking points, after a rapid rebuttal, are with possible degeneracies in the loss function and class-level information leakage from pre-trained classifiers, making these results are not properly ""unconditional"". The authors rebutted this by suggesting that unlike Salimans et al (2016), there is no signal backpropagated from the label layer, but I find this particularly unconvincing: the objective in that work maximizes a ""none-of-the-above"" class (and thus minimizes *all* classes). The gradient backpropagated to the generator is uninformative about which particular class a sample should imitate, but the features learned by the discriminator needing to discriminate between classes shape those gradients in a particular way all the same, and the result is samples that look like distinct CIFAR classes. In the same way, the gradients used to train GFMN are ""shaped"" by particular class-discriminative features when trained against a classifier feature extractor. From my own perspective, while there is no theory presented to support why this method is a good idea (why matching arbitrary features unconnected with the generative objective should lead to good results), the idea of optimizing a moment matching objective in classifier feature space is rather obvious, and it is unsurprising that with enough ""elbow grease"" it can be made to work. The Adam moving average trick is interesting but a deeper analysis and ablation of why this works would have helped convince the reader that it is principled. This paper was very much on the borderline. Aside from quibbles over the fairness of comparisons above, I was forced to ask myself whether I could imagine that this would be a widely read, influential, and frequently cited piece of work. I believe that the carefully done empirical investigation has its merits, but that the core ideas are rather obvious and the added novelty of a poorly understood stabilized moving average is not enough to warrant acceptance."""
1027,"""SNIP: SINGLE-SHOT NETWORK PRUNING BASED ON CONNECTION SENSITIVITY""","['neural network pruning', 'connection sensitivity']","""Pruning large neural networks while maintaining their performance is often desirable due to the reduced space and time complexity. In existing methods, pruning is done within an iterative optimization procedure with either heuristically designed pruning schedules or additional hyperparameters, undermining their utility. In this work, we present a new approach that prunes a given network once at initialization prior to training. To achieve this, we introduce a saliency criterion based on connection sensitivity that identifies structurally important connections in the network for the given task. This eliminates the need for both pretraining and the complex pruning schedule while making it robust to architecture variations. After pruning, the sparse network is trained in the standard way. Our method obtains extremely sparse networks with virtually the same accuracy as the reference network on the MNIST, CIFAR-10, and Tiny-ImageNet classification tasks and is broadly applicable to various architectures including convolutional, residual and recurrent networks. Unlike existing methods, our approach enables us to demonstrate that the retained connections are indeed relevant to the given task.""","""This method proposes a criterion (SNIP) to prune neural networks before training. The pro is that SNIP can find the architecturally important parameters in the network without full training. The con is that SNIP only evaluated on small datasets (mnist, cifar, tiny-imagenet) and it's uncertain if the same heuristic works on large-scale dataset. Small datasets can always achieve high pruning ratio, so evaluation on ImageNet is quite important for pruning work. The reviewers have consensus on accept. The authors are recommended to compare with previous work [1][2] to make the paper more convincing. [1] Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. NIPS, 2015. [2] Yiwen Guo, Anbang Yao, and Yurong Chen. Dynamic network surgery for efficient dnns. NIPS, 2016."""
1028,"""Assumption Questioning: Latent Copying and Reward Exploitation in Question Generation""","['question generation', 'answer questioning', 'pointer networks', 'reinforcement learning']","""Question generation is an important task for improving our ability to process natural language data, with additional challenges over other sequence transformation tasks. Recent approaches use modifications to a Seq2Seq architecture inspired by advances in machine translation, but unlike translation the input and output vocabularies overlap significantly, and there are many different valid questions for each input. Approaches using copy mechanisms and reinforcement learning have shown promising results, but there are ambiguities in the exact implementation that have not yet been investigated. We show that by removing inductive bias from the model and allowing the choice of generation path to become latent, we achieve substantial improvements over implementations biased with both naive and smart heuristics. We perform a human evaluation to confirm these findings. We show that although policy gradient methods may be used to decouple training from the ground truth and optimise directly for quality metrics that have previously been assumed to be good choices, these objectives are poorly aligned with human judgement and the model simply learns to exploit the weaknesses of the reward source. Finally, we show that an adversarial objective learned directly from the ground truth data is not able to generate a useful training signal.""","""This paper investigates copying mechanisms and reward functions in sequence to sequence models for question generation. The key findings are threefold: (1) when the alignments between input and output are weak, it is better to use latent copying mechanism to soften the model bias toward copying, (2) while policy gradient methods might be able to improve automatic scores, their results poorly align with human evaludation, and (3) the use of adversarial objective also does not lead to useful training signals. Pros: The task is well motivated and the paper presents potentially useful negative results on policy gradient and adversarial training. Cons: All reviewers found the clarity and organization of the paper requires improvements. Also, the proposed methods are reletively incremental and the empirical results are not strong. While the rebuttal answered some of the clarification questions, it does not address major concerns about the novelty and contributions. Verdict: Reject due to relatively weak contributions and novelty."""
1029,"""iRDA Method for Sparse Convolutional Neural Networks""","['sparse convolutional neural networks', 'regularized dual averaging']","""We propose a new approach, known as the iterative regularized dual averaging (iRDA), to improve the efficiency of convolutional neural networks (CNN) by significantly reducing the redundancy of the model without reducing its accuracy. The method has been tested for various data sets, and proven to be significantly more efficient than most existing compressing techniques in the deep learning literature. For many popular data sets such as MNIST and CIFAR-10, more than 95% of the weights can be zeroed out without losing accuracy. In particular, we are able to make ResNet18 with 95% sparsity to have an accuracy that is comparable to that of a much larger model ResNet50 with the best 60% sparsity as reported in the literature.""","""This paper proposes an iterative regularized dual averaging method to sparsify CNN weights during learning. The main contribution seems to be in an iterative procedure where the weights are pruned out greedily by observing the sparsity of the averaged gradients. The reviewers agree that the idea seems straightforward and novelty is limited. For this reason, I recommend to reject this paper. """
1030,"""Neural Speed Reading with Structural-Jump-LSTM""","['natural language processing', 'speed reading', 'recurrent neural network', 'classification']","""Recurrent neural networks (RNNs) can model natural language by sequentially ''reading'' input tokens and outputting a distributed representation of each token. Due to the sequential nature of RNNs, inference time is linearly dependent on the input length, and all inputs are read regardless of their importance. Efforts to speed up this inference, known as ''neural speed reading'', either ignore or skim over part of the input. We present Structural-Jump-LSTM: the first neural speed reading model to both skip and jump text during inference. The model consists of a standard LSTM and two agents: one capable of skipping single words when reading, and one capable of exploiting punctuation structure (sub-sentence separators (,:), sentence end symbols (.!?), or end of text markers) to jump ahead after reading a word. A comprehensive experimental evaluation of our model against all five state-of-the-art neural reading models shows that Structural-Jump-LSTM achieves the best overall floating point operations (FLOP) reduction (hence is faster), while keeping the same accuracy or even improving it compared to a vanilla LSTM that reads the whole text.""","""The authors obtain nice speed improvements by learning to skip and jump over input words when processing text with an LSTM. At some points the reviewers considered the work incremental since similar ideas have already been explored, but at the end two of the reviewers ended up endorsing the paper with strong support."""
1031,"""Efficient Augmentation via Data Subsampling""","['data augmentation', 'invariance', 'subsampling', 'influence']","""Data augmentation is commonly used to encode invariances in learning methods. However, this process is often performed in an inefficient manner, as artificial examples are created by applying a number of transformations to all points in the training set. The resulting explosion of the dataset size can be an issue in terms of storage and training costs, as well as in selecting and tuning the optimal set of transformations to apply. In this work, we demonstrate that it is possible to significantly reduce the number of data points included in data augmentation while realizing the same accuracy and invariance benefits of augmenting the entire dataset. We propose a novel set of subsampling policies, based on model influence and loss, that can achieve a 90% reduction in augmentation set size while maintaining the accuracy gains of standard data augmentation.""","""The paper proposes several subsampling policies to achieve a clear reduction in the size of augmented data while maintaining the accuracy of using a standard data augmentation method. The paper in general is clearly written and easy to follow, and provides sufficiently convincing experimental results to support the claim. After reading the authors' response and revision, the reviewers have reached a general consensus that the paper is above the acceptance bar. """
1032,"""A Resizable Mini-batch Gradient Descent based on a Multi-Armed Bandit""","['Batch size', 'Optimization', 'Mini-batch gradient descent', 'Multi-armed bandit']","""Determining the appropriate batch size for mini-batch gradient descent is always time consuming as it often relies on grid search. This paper considers a resizable mini-batch gradient descent (RMGD) algorithm based on a multi-armed bandit that achieves performance equivalent to that of best fixed batch-size. At each epoch, the RMGD samples a batch size according to a certain probability distribution proportional to a batch being successful in reducing the loss function. Sampling from this probability provides a mechanism for exploring different batch size and exploiting batch sizes with history of success. After obtaining the validation loss at each epoch with the sampled batch size, the probability distribution is updated to incorporate the effectiveness of the sampled batch size. Experimental results show that the RMGD achieves performance better than the best performing single batch size. It is surprising that the RMGD achieves better performance than grid search. Furthermore, it attains this performance in a shorter amount of time than grid search.""","""It is a simple but good idea to consider the choice of mini-batch size as a multi-armed bandit problem. Experiments also show a slight improvement compared to the best fixed batch size. The main concerns from the reviewers are that (1) treating the choice of hyper-parameters as a bandit problem is known and has been exploited in different context, and this paper is limited to the choice of the mini-batch size, (2) the improvement in the test error is not significant. The authors' feedback did not solve the concerns raised by R2. This paper conveys a nice idea but as the current form it falls slightly below the standard of the ICLR publications. One direction for improvement, as suggested by the reviewer, would be extending the idea for a wider hyper-parameter selection problems."""
1033,"""Explaining Adversarial Examples with Knowledge Representation""","['adversarial example', 'knowledge representation', 'distribution imitation']","""Adversarial examples are modified samples that preserve original image structures but deviate classifiers. Researchers have put efforts into developing methods for generating adversarial examples and finding out origins. Past research put much attention on decision boundary changes caused by these methods. This paper, in contrast, discusses the origin of adversarial examples from a more underlying knowledge representation point of view. Human beings can learn and classify prototypes as well as transformations of objects. While neural networks store learned knowledge in a more hybrid way of combining all prototypes and transformations as a whole distribution. Hybrid storage may lead to lower distances between different classes so that small modifications can mislead the classifier. A one-step distribution imitation method is designed to imitate distribution of the nearest different class neighbor. Experiments show that simply by imitating distributions from a training set without any knowledge of the classifier can still lead to obvious impacts on classification results from deep networks. It also implies that adversarial examples can be in more forms than small perturbations. Potential ways of alleviating adversarial examples are discussed from the representation point of view. The first path is to change the encoding of data sent to the training step. Training data that are more prototypical can help seize more robust and accurate structural knowledge. The second path requires constructing learning frameworks with improved representations.""","""The reviewers have agreed this paper is not ready for publication at ICLR. """
1034,"""DEEP GEOMETRICAL GRAPH CLASSIFICATION""","['Graph classification', 'Deep Learning', 'Graph pooling', 'Embedding']","""Most of the existing Graph Neural Networks (GNNs) are the mere extension of the Convolutional Neural Networks (CNNs) to graphs. Generally, they consist of several steps of message passing between the nodes followed by a global indiscriminate feature pooling function. In many data-sets, however, the nodes are unlabeled or their labels provide no information about the similarity between the nodes and the locations of the nodes in the graph. Accordingly, message passing may not propagate helpful information throughout the graph. We show that this conventional approach can fail to learn to perform even simple graph classification tasks. We alleviate this serious shortcoming of the GNNs by making them a two step method. In the first of the proposed approach, a graph embedding algorithm is utilized to obtain a continuous feature vector for each node of the graph. The embedding algorithm represents the graph as a point-cloud in the embedding space. In the second step, the GNN is applied to the point-cloud representation of the graph provided by the embedding method. The GNN learns to perform the given task by inferring the topological structure of the graph encoded in the spatial distribution of the embedded vectors. In addition, we extend the proposed approach to the graph clustering problem and a new architecture for graph clustering is proposed. Moreover, the spatial representation of the graph is utilized to design a graph pooling algorithm. We turn the problem of graph down-sampling into a column sampling problem, i.e., the sampling algorithm selects a subset of the nodes whose feature vectors preserve the spatial distribution of all the feature vectors. We apply the proposed approach to several popular benchmark data-sets and it is shown that the proposed geometrical approach strongly improves the state-of-the-art result for several data-sets. For instance, for the PTC data-set, we improve the state-of-the-art result for more than 22 %.""","""The extension of convnets to non-Euclidean data is a major theme of research in computer vision and signal processing. This paper is concerned with Graph structured datasets. The main idea seems to be interesting: to improve graph neural nets by first embedding the graph in a Euclidean space reducing it to a point cloud, and then exploiting the induced topological structure implicit in the point cloud. However, all reviewers found this paper hard to read and improperly motivated due to poor writing quality. The experimental results are somewhat promising but not completely convincing, and the proposed framework lacks a solid theoretical footing. Hence, the AC cannot recommend acceptance at ICLR-2019. """
1035,"""Building Dynamic Knowledge Graphs from Text using Machine Reading Comprehension""","['recurrent graph networks', 'dynamic knowledge base construction', 'entity state tracking', 'machine reading comprehension']","""We propose a neural machine-reading model that constructs dynamic knowledge graphs from procedural text. It builds these graphs recurrently for each step of the described procedure, and uses them to track the evolving states of participant entities. We harness and extend a recently proposed machine reading comprehension(MRC) model to query for entity states, since these states are generally communicated in spans of text and MRC models perform well in extracting entity-centric spans. The explicit, structured, and evolving knowledge graph representations that our model constructs can be used in downstream question answering tasks to improve machine comprehension of text, as we demonstrate empirically. On two comprehension tasks from the recently proposed ProPara dataset, our model achieves state-of-the-art results. We further show that our model is competitive on the Recipes dataset, suggesting it may be generally applicable.""","""This paper investigates a new approach to machine reading for procedural text, where the task of reading comprehension is formulated as dynamic construction of a procedural knowledge graph. The proposed model constructs a recurrent knowledge graph (as a bipartite graph between entities and location nodes) and tracks the entity states for two domains: scientific processes and recipes. Pros: The idea of formulating reading comprehension as dynamic construction of a knowledge graph is novel and interesting. The proposed model is tested on two different domains: scientific processes (ProPara) and cooking recipes. Cons: The initial submission didn't have the experimental results on the full recipe dataset and also had several clarity issues, all of which have been resolved through the rebuttal. Verdict: Accept. An interesting task & models with solid empirical results. """
1036,"""Selfless Sequential Learning""","['Lifelong learning', 'Continual Learning', 'Sequential learning', 'Regularization']","""Sequential learning, also called lifelong learning, studies the problem of learning tasks in a sequence with access restricted to only the data of the current task. In this paper we look at a scenario with fixed model capacity, and postulate that the learning process should not be selfish, i.e. it should account for future tasks to be added and thus leave enough capacity for them. To achieve Selfless Sequential Learning we study different regularization strategies and activation functions. We find that imposing sparsity at the level of the representation (i.e. neuron activations) is more beneficial for sequential learning than encouraging parameter sparsity. In particular, we propose a novel regularizer, that encourages representation sparsity by means of neural inhibition. It results in few active neurons which in turn leaves more free neurons to be utilized by upcoming tasks. As neural inhibition over an entire layer can be too drastic, especially for complex tasks requiring strong representations, our regularizer only inhibits other neurons in a local neighbourhood, inspired by lateral inhibition processes in the brain. We combine our novel regularizer with state-of-the-art lifelong learning methods that penalize changes to important previously learned parts of the network. We show that our new regularizer leads to increased sparsity which translates in consistent performance improvement on diverse datasets.""","""Two of the reviewers raised their scores during the discussion phase noting that the revised version was clearer and addressed some of their concerns. As a result, all the reviewers ultimately recommended acceptance. They particularly enjoyed the insights that the authors shared from their experiments and appreciated that the experiments were quite thorough. All the reviewers mentioned that the work seemed somewhat incremental, but given the results, insights and empirical evaluation decided that it would still be a valuable contribution to the conference. One reviewer added feedback about how to improve the writing and clarity of the paper for the camera ready version."""
1037,"""Generative Models from the perspective of Continual Learning""","['Generative Models', 'Continual Learning']","""Which generative model is the most suitable for Continual Learning? This paper aims at evaluating and comparing generative models on disjoint sequential image generation tasks. We investigate how several models learn and forget, considering various strategies: rehearsal, regularization, generative replay and fine-tuning. We used two quantitative metrics to estimate the generation quality and memory ability. We experiment with sequential tasks on three commonly used benchmarks for Continual Learning (MNIST, Fashion MNIST and CIFAR10). We found that among all models, the original GAN performs best and among Continual Learning strategies, generative replay outperforms all other methods. Even if we found satisfactory combinations on MNIST and Fashion MNIST, training generative models sequentially on CIFAR10 is particularly instable, and remains a challenge.""","""This paper presents empirical evaluation and comparison of different generative models (such as GANs and VAE) in the continual learning setting. To avoid catastrophic forgetting, the following strategies are considered: rehearsal, regularization, generative replay and fine-tuning. The empirical evaluations are carried out using three datasets (MNIST, Fashion MNIST and CIFAR). While all reviewers and AC acknowledge the importance and potential usefulness of studying and comparing different generative models in continual learning, they raised several important concerns that place this paper bellow the acceptance bar: (1) in an empirical study paper, an in-depth analysis and more insightful evaluations are required to better understand the benefits and shortcomings of the available models (R1 and R2), e.g. analyzing why generative replay fails to improve VAE, why is rehearsal better for likelihood models, and in general why certain combinations are more effective than others see more suggestions in R1s and R2s comments. The authors discussed in their response to the reviews some of these questions, but a more detailed analysis is required to fully understand the benefits of this empirical study. (2) The evaluation is geared towards quality metrics for the generative models and lacks evaluation for catastrophic forgetting in continual learning (hence it favours GANs models) -- See R3s suggestion how to improve. To conclude, the reviewers and AC suggest that in its current state the manuscript is not ready for a publication. We hope the reviews are useful for improving and revising the paper. """
1038,"""Learning sparse relational transition models""","['Deictic reference', 'relational model', 'rule-based transition model']","""We present a representation for describing transition models in complex uncertain domains using relational rules. For any action, a rule selects a set of relevant objects and computes a distribution over properties of just those objects in the resulting state given their properties in the previous state. An iterative greedy algorithm is used to construct a set of deictic references that determine which objects are relevant in any given state. Feed-forward neural networks are used to learn the transition distribution on the relevant objects' properties. This strategy is demonstrated to be both more versatile and more sample efficient than learning a monolithic transition model in a simulated domain in which a robot pushes stacks of objects on a cluttered table.""","""pros: - the paper is well-written and precise - the proposed method is novel - valuable for real-world problems cons: - Reviewer 2 expresses some concern about the organization of the paper and over-generality in the exposition - There could be more discussion of scalability"""
1039,"""Feed-forward Propagation in Probabilistic Neural Networks with Categorical and Max Layers""","['probabilistic neural network', 'uncertainty', 'dropout', 'bayesian', 'softmax', 'argmax', 'logsumexp']","""Probabilistic Neural Networks deal with various sources of stochasticity: input noise, dropout, stochastic neurons, parameter uncertainties modeled as random variables, etc. In this paper we revisit a feed-forward propagation approach that allows one to estimate for each neuron its mean and variance w.r.t. all mentioned sources of stochasticity. In contrast, standard NNs propagate only point estimates, discarding the uncertainty. Methods propagating also the variance have been proposed by several authors in different context. The view presented here attempts to clarify the assumptions and derivation behind such methods, relate them to classical NNs and broaden their scope of applicability. The main technical contributions are new approximations for the distributions of argmax and max-related transforms, which allow for fully analytic uncertainty propagation in networks with softmax and max-pooling layers as well as leaky ReLU activations. We evaluate the accuracy of the approximation and suggest a simple calibration. Applying the method to networks with dropout allows for faster training and gives improved test likelihoods without the need of sampling.""","""Reviewers are in a consensus and recommended to accept after engaging with the authors. Please take reviewers' comments into consideration to improve your submission for the camera ready. """
1040,"""Visceral Machines: Risk-Aversion in Reinforcement Learning with Intrinsic Physiological Rewards""","['Reinforcement Learning', 'Simulation', 'Affective Computing']",""" As people learn to navigate the world, autonomic nervous system (e.g., ``fight or flight) responses provide intrinsic feedback about the potential consequence of action choices (e.g., becoming nervous when close to a cliff edge or driving fast around a bend.) Physiological changes are correlated with these biological preparations to protect one-self from danger. We present a novel approach to reinforcement learning that leverages a task-independent intrinsic reward function trained on peripheral pulse measurements that are correlated with human autonomic nervous system responses. Our hypothesis is that such reward functions can circumvent the challenges associated with sparse and skewed rewards in reinforcement learning settings and can help improve sample efficiency. We test this in a simulated driving environment and show that it can increase the speed of learning and reduce the number of collisions during the learning stage.""","""The paper considers the problem of incorporating human physiological feedback into an autonomous driving system, where minimization of a predicted arousal response is used as an additional source of reward signal, with the intuition that this could be used as a proxy for training a policy that is risk-averse. Reviewers were generally positive about the novelty and relevance of the approach but had methodological concerns. In particular, concerns about the weighting of the intrinsic vs. extrinsic reward (why under different settings the optimal tradeoff parameter was different, how this affects the optimal policy if the influence of the intrinsic reward is not decreased with time). Additional baseline experiments were requested and performed, and the paper was modified to significantly incorporate other feedback such as drawing connections to imitation learning. A title change was proposed and accepted to reflect the focus on the application of risk aversion (I'd ask that the authors update the paper OpenReview metadata to reflect this). At a high level, I believe this is an original and interesting contribution to the literature. I have not heard from two of three reviewers regarding whether their concerns were addressed, but given that their concerns appear to me to have been addressed (and their initial scores indicated that the work met the bar for acceptance, if only marginally), I am inclined to recommend acceptance."""
1041,"""Sequenced-Replacement Sampling for Deep Learning""","['deep neural networks', 'stochastic gradient descent', 'sequenced-replacement sampling']","""We propose sequenced-replacement sampling (SRS) for training deep neural networks. The basic idea is to assign a fixed sequence index to each sample in the dataset. Once a mini-batch is randomly drawn in each training iteration, we refill the original dataset by successively adding samples according to their sequence index. Thus we carry out replacement sampling but in a batched and sequenced way. In a sense, SRS could be viewed as a way of performing ""mini-batch augmentation"". It is particularly useful for a task where we have a relatively small images-per-class such as CIFAR-100. Together with a longer period of initial large learning rate, it significantly improves the classification accuracy in CIFAR-100 over the current state-of-the-art results. Our experiments indicate that training deeper networks with SRS is less prone to over-fitting. In the best case, we achieve an error rate as low as 10.10%.""","""This paper proposes a new batching strategy for training deep nets. The idea is to have the properties of sampling with replacement while reducing the chance of not touching an example in a given epoch. Experimental results show that this can improve performance on one of the tasks considered. However the reviewers consistently agree that the experimental validation of this work is much too limited. Furthermore the motivation for the approach should be more clearly established."""
1042,"""Traditional and Heavy Tailed Self Regularization in Neural Network Models""","['statistical mechanics', 'self-regularization', 'random matrix', 'glassy behavior', 'heavy-tailed']","""Random Matrix Theory (RMT) is applied to analyze the weight matrices of Deep Neural Networks (DNNs), including both production quality, pre-trained models such as AlexNet and Inception, and smaller models trained from scratch, such as LeNet5 and a miniature-AlexNet. Empirical and theoretical results clearly indicate that the empirical spectral density (ESD) of DNN layer matrices displays signatures of traditionally-regularized statistical models, even in the absence of exogenously specifying traditional forms of regularization, such as Dropout or Weight Norm constraints. Building on recent results in RMT, most notably its extension to Universality classes of Heavy-Tailed matrices, we develop a theory to identify 5+1 Phases of Training, corresponding to increasing amounts of Implicit Self-Regularization. For smaller and/or older DNNs, this Implicit Self-Regularization is like traditional Tikhonov regularization, in that there is a ""size scale"" separating signal from noise. For state-of-the-art DNNs, however, we identify a novel form of Heavy-Tailed Self-Regularization, similar to the self-organization seen in the statistical physics of disordered systems. This implicit Self-Regularization can depend strongly on the many knobs of the training process. By exploiting the generalization gap phenomena, we demonstrate that we can cause a small model to exhibit all 5+1 phases of training simply by changing the batch size.""","""While it appears that the authors have done significant amount of work to investigate this topic, there are concerns that the theorems are not rigorously/precisely presented, and it is unclear how they can guide the design and training of neural network models in practice. The response and revision of the authors do not provide sufficient materials to address these concerns. """
1043,"""Effective Path: Know the Unknowns of Neural Network""",[],"""Despite their enormous success, there is still no solid understanding of deep neural networks working mechanism. As such, researchers have demonstrated DNNs are vulnerable to small input perturbation, i.e., adversarial attacks. This work proposes the effective path as a new approach to exploring DNNs' internal organization. The effective path is an ensemble of synapses and neurons, which is reconstructed from a trained DNN using our activation-based backward algorithm. The per-image effective path can be aggregated to the class-level effective path, through which we observe that adversarial images activate effective path different from normal images. We propose an effective path similarity-based method to detect adversarial images and demonstrate its high accuracy and broad applicability. ""","""The paper presents an approach to estimate the ""effective path"" of examples in a network to reach a decision, and consider this to analyze if examples might be adversarial. Reviewers think the paper lacks some clarity and experiments. They point to a confusion between interpretability and adversarial attacks, they ask questions about computational complexity, and point to some unsubstanciated claims. Authors have not responded to reviewers. Overall, I concur with the reviewers to reject the paper."""
1044,"""State-Regularized Recurrent Networks""","['recurrent network', 'finite state machines', 'state-regularized', 'interpretability and explainability']","""Recurrent networks are a widely used class of neural architectures. They have, however, two shortcomings. First, it is difficult to understand what exactly they learn. Second, they tend to work poorly on sequences requiring long-term memorization, despite having this capacity in principle. We aim to address both shortcomings with a class of recurrent networks that use a stochastic state transition mechanism between cell applications. This mechanism, which we term state-regularization, makes RNNs transition between a finite set of learnable states. We show that state-regularization (a) simplifies the extraction of finite state automata modeling an RNN's state transition dynamics, and (b) forces RNNs to operate more like automata with external memory and less like finite state machines.""","""the authors propose to incorporate an additional layer between the consecutive steps in LSTM by introducing a radial basis function layer (with dot product kernel and softmax) followed by a linear layer to make LSTM similar to or better at (by being more explicit) capturing DFA-like transition. the motivation is relatively straightforward, but it does not really resolve the issue of whether existing formulations of RNN's cannot capture such transition. since this was not shown theoretically nor intuitively, it is important for empirical evaluations to be thorough and clearly show that the proposed approach does indeed outperform the vanilla LSTM (with peepholes) when the capacity (e.g., the number of parameters) matches. unfortunately it has been the consensus among the reviewers that more thorough comparison on more conventional benchmarks are needed to convince them of the merit of the proposed approach."""
1045,"""Why Do Neural Response Generation Models Prefer Universal Replies?""","['Neural Response Generation', 'Universal Replies', 'Optimization Goal Analysis', 'Max-Marginal Ranking Regularization']","""Recent advances in neural Sequence-to-Sequence (Seq2Seq) models reveal a purely data-driven approach to the response generation task. Despite its diverse variants and applications, the existing Seq2Seq models are prone to producing short and generic replies, which blocks such neural network architectures from being utilized in practical open-domain response generation tasks. In this research, we analyze this critical issue from the perspective of the optimization goal of models and the specific characteristics of human-to-human conversational corpora. Our analysis is conducted by decomposing the goal of Neural Response Generation (NRG) into the optimizations of word selection and ordering. It can be derived from the decomposing that Seq2Seq based NRG models naturally tend to select common words to compose responses, and ignore the semantic of queries in word ordering. On the basis of the analysis, we propose a max-marginal ranking regularization term to avoid Seq2Seq models from producing the generic and uninformative responses. The empirical experiments on benchmarks with several metrics have validated our analysis and proposed methodology.""","""This paper seeks to shed light on why seq2seq models favor generic replies. The problem is an important one, unfortunately the responses proposed in the paper are not satisfactory. Most reviewers note problems and general lack of rigorousness in the assumptions used to produce the theoretical part of the paper (e.g., strong assumption of independence of generated words). The experiments themselves are not convincing enough to warrant acceptance by themselves."""
1046,"""The role of over-parametrization in generalization of neural networks""","['Generalization', 'Over-Parametrization', 'Neural Networks', 'Deep Learning']","""Despite existing work on ensuring generalization of neural networks in terms of scale sensitive complexity measures, such as norms, margin and sharpness, these complexity measures do not offer an explanation of why neural networks generalize better with over-parametrization. In this work we suggest a novel complexity measure based on unit-wise capacities resulting in a tighter generalization bound for two layer ReLU networks. Our capacity bound correlates with the behavior of test error with increasing network sizes (within the range reported in the experiments), and could partly explain the improvement in generalization with over-parametrization. We further present a matching lower bound for the Rademacher complexity that improves over previous capacity lower bounds for neural networks. ""","""I agree with the reviewers that this is a strong contribution and provides new insights, even if it doesn't quite close the problem. p.s.: It seems that centering the weight matrices at initialization is a key idea. The authors note that Dziugaite and Roy used bounds that were based on the distance to initialization, but that their reported numerical generalization bounds also increase with the increasing network size. Looking back at that work, they look at networks where the size increases by a very large factor (going from e.g. 400,000 parameters roughly to over 1.2 million, so a factor of 2.5), at the same time the bound increases by a much smaller factor. The type of increase also seems much less severe than those pictured in Figures 3/5. Since Dzugate and Roy's bounds involved optimization, perhaps the increase there is merely apparent."""
1047,"""Diversity is All You Need: Learning Skills without a Reward Function""","['reinforcement learning', 'unsupervised learning', 'skill discovery']","""Intelligent creatures can explore their environments and learn useful skills without supervision. In this paper, we propose ``Diversity is All You Need''(DIAYN), a method for learning useful skills without a reward function. Our proposed method learns skills by maximizing an information theoretic objective using a maximum entropy policy. On a variety of simulated robotic tasks, we show that this simple objective results in the unsupervised emergence of diverse skills, such as walking and jumping. In a number of reinforcement learning benchmark environments, our method is able to learn a skill that solves the benchmark task despite never receiving the true task reward. We show how pretrained skills can provide a good parameter initialization for downstream tasks, and can be composed hierarchically to solve complex, sparse reward tasks. Our results suggest that unsupervised discovery of skills can serve as an effective pretraining mechanism for overcoming challenges of exploration and data efficiency in reinforcement learning.""","""There is consensus among the reviewer that this is a good paper. It is a bit incremental compared to Gregor et al 2016. This paper show quite better empirical results."""
1048,"""Learning when to Communicate at Scale in Multiagent Cooperative and Competitive Tasks""","['multiagent', 'communication', 'competitive', 'cooperative', 'continuous', 'emergent', 'reinforcement learning']","""Learning when to communicate and doing that effectively is essential in multi-agent tasks. Recent works show that continuous communication allows efficient training with back-propagation in multi-agent scenarios, but have been restricted to fully-cooperative tasks. In this paper, we present Individualized Controlled Continuous Communication Model (IC3Net) which has better training efficiency than simple continuous communication model, and can be applied to semi-cooperative and competitive settings along with the cooperative settings. IC3Net controls continuous communication with a gating mechanism and uses individualized rewards foreach agent to gain better performance and scalability while fixing credit assignment issues. Using variety of tasks including StarCraft BroodWars explore and combat scenarios, we show that our network yields improved performance and convergence rates than the baselines as the scale increases. Our results convey that IC3Net agents learn when to communicate based on the scenario and profitability.""","""All reviewers agree that the proposed is interesting and innovative. One reviewer argues that some additional baseline comparisons could be beneficial and the other two suggest inclusion of additional explanations and discussions of the results. The authors rebuttal alleviated most of the concerns. All reviewers are very appreciative of the quality of the work overall and recommend probable acceptance. I agree with this score and recommend this work for poster presentation at ICLR."""
1049,"""INTERPRETABLE CONVOLUTIONAL FILTER PRUNING""",[],"""The sophisticated structure of Convolutional Neural Network (CNN) allows for outstanding performance, but at the cost of intensive computation. As significant redundancies inevitably present in such a structure, many works have been proposed to prune the convolutional filters for computation cost reduction. Although extremely effective, most works are based only on quantitative characteristics of the convolutional filters, and highly overlook the qualitative interpretation of individual filters specific functionality. In this work, we interpreted the functionality and redundancy of the convolutional filters from different perspectives, and proposed a functionality-oriented filter pruning method. With extensive experiment results, we proved the convolutional filters qualitative significance regardless of magnitude, demonstrated significant neural network redundancy due to repetitive filter functions, and analyzed the filter functionality defection under inappropriate retraining process. Such an interpretable pruning approach not only offers outstanding computation cost optimization over previous filter pruning methods, but also interprets filter pruning process.""","""The current version of the paper receives a unanimous rejection from reviewers, as the final proposal. """
1050,"""On Meaning-Preserving Adversarial Perturbations for Sequence-to-Sequence Models""","['Sequence-to-sequence', 'adversarial attacks', 'evaluation', 'meaning preservation', 'machine translation']","""Adversarial examples have been shown to be an effective way of assessing the robustness of neural sequence-to-sequence (seq2seq) models, by applying perturbations to the input of a model leading to large degradation in performance. However, these perturbations are only indicative of a weakness in the model if they do not change the semantics of the input in a way that would change the expected output. Using the example of machine translation (MT), we propose a new evaluation framework for adversarial attacks on seq2seq models taking meaning preservation into account and demonstrate that existing methods may not preserve meaning in general. Based on these findings, we propose new constraints for attacks on word-based MT systems and show, via human and automatic evaluation, that they produce more semantically similar adversarial inputs. Furthermore, we show that performing adversarial training with meaning-preserving attacks is beneficial to the model in terms of adversarial robustness without hurting test performance.""","""This paper present a framework for creating meaning-preserving adversarial examples. It then proposes two attacks within this framework: one based on k-NN in the word embedding space, and another one based on character swapping. Overall, the goal of constructing such meaning-preserving attacks is very interesting. However, it is unclear how successful the proposed approach really is in the context of this goal. Additionally, it is not clear how much novelty there is compared to already existing methods that have a very similar aim."""
1051,"""Universal Stagewise Learning for Non-Convex Problems with Convergence on Averaged Solutions""","['optimization', 'sgd', 'adagrad']","""Although stochastic gradient descent (SGD) method and its variants (e.g., stochastic momentum methods, AdaGrad) are algorithms of choice for solving non-convex problems (especially deep learning), big gaps still remain between the theory and the practice with many questions unresolved. For example, there is still a lack of theories of convergence for SGD and its variants that use stagewise step size and return an averaged solution in practice. In addition, theoretical insights of why adaptive step size of AdaGrad could improve non-adaptive step size of SGD is still missing for non-convex optimization. This paper aims to address these questions and fill the gap between theory and practice. We propose a universal stagewise optimization framework for a broad family of non-smooth non-convex problems with the following key features: (i) at each stage any suitable stochastic convex optimization algorithms (e.g., SGD or AdaGrad) that return an averaged solution can be employed for minimizing a regularized convex problem; (ii) the step size is decreased in a stagewise manner; (iii) an averaged solution is returned as the final solution. % that is selected from all stagewise averaged solutions with sampling probabilities increasing as the stage number. Our theoretical results of stagewise {\ada} exhibit its adaptive convergence, therefore shed insights on its faster convergence than stagewise SGD for problems with slowly growing cumulative stochastic gradients. To the best of our knowledge, these new results are the first of their kind for addressing the unresolved issues of existing theories mentioned earlier. Besides theoretical contributions, our empirical studies show that our stagewise variants of SGD, AdaGrad improve the generalization performance of existing variants/implementations of SGD and AdaGrad. ""","""This paper develops a stagewise optimization framework for solving non smooth and non convex problems. The idea is to use standard convex solvers to iteratively optimize a regularized objective with penalty centered at previous iterates - which is standard in many proximal methods. The paper combines this with the analysis for non-smooth functions giving a more general convergence results. Reviewers agree on the usefulness and novelty of the contribution. Initially there were concerns about lack of comparison with current results, but updated version have addressed this issue. The main weakness is that the results only holds for \mu weekly convex functions and the algorithm depends on the knowledge of \mu. Despite this limitations, reviewers believe that the paper has enough new material and I suggest for publication. I suggest authors to address these issues in the final version. """
1052,"""Invariance and Inverse Stability under ReLU""","['deep neural networks', 'invertibility', 'invariance', 'robustness', 'ReLU networks']","""We flip the usual approach to study invariance and robustness of neural networks by considering the non-uniqueness and instability of the inverse mapping. We provide theoretical and numerical results on the inverse of ReLU-layers. First, we derive a necessary and sufficient condition on the existence of invariance that provides a geometric interpretation. Next, we move to robustness via analyzing local effects on the inverse. To conclude, we show how this reverse point of view not only provides insights into key effects, but also enables to view adversarial examples from different perspectives.""","""The main strength of the paper is to provide a clear mathematical characterization of invertible neural networks. The reviewers and the AC also note potential weakness including 1) the exposition of the paper can be much improved; 2) it's unclear how these analyses can help improve the training algorithm or architecture design since these characterizations are likely not computable; 3) the novelty compared to previous work Carlsson et al. 2017 may not be enough for ICLR acceptance. These weakness are considered critical issues by the AC in the decision. """
1053,"""Neural Logic Machines""","['Neuro-Symbolic Computation', 'Logic Induction']","""We propose the Neural Logic Machine (NLM), a neural-symbolic architecture for both inductive learning and logic reasoning. NLMs exploit the power of both neural networks---as function approximators, and logic programming---as a symbolic processor for objects with properties, relations, logic connectives, and quantifiers. After being trained on small-scale tasks (such as sorting short arrays), NLMs can recover lifted rules, and generalize to large-scale tasks (such as sorting longer arrays). In our experiments, NLMs achieve perfect generalization in a number of tasks, from relational reasoning tasks on the family tree and general graphs, to decision making tasks including sorting arrays, finding shortest paths, and playing the blocks world. Most of these tasks are hard to accomplish for neural networks or inductive logic programming alone.""",""" pros: - The paper presents an interesting forward chaining model which makes use of meta-level expansions and reductions on predicate arguments in a neat way to reduce complexity. As Reviewer 3 points out, there are a number of other papers from the neuro-symbolic community that learn relations (logic tensor networks is one good reference there). However using these meta-rules you can mix predicates of different arities in a principled way in the construction of the rules, which is something I haven't seen. - The paper is reasonably well written (see cons for specific issues) - There is quite a broad evaluation across a number of different tasks. I appreciated that you integrated this into an RL setting for tasks like blocks world. - The results are good on small datasets and generalize well cons: - (scalability) As both Reviewers 1 and 3 point out, there are scalability issues as a function of the predicate arity in computing the set of permutations for the output predicate computation. - (interpretability) As Reviewer 2 notes, unlike del-ILP, it is not obvious how symbolic rules can be extracted. This is an important point to address up front in the text. - (clarity) The paper is confusing or ambiguous in places: -Initially I read the 1,2,3 sequence at the top of 3 to be a deduction (and was confused) rather than three applications of the meta-rules. Maybe instead of calling that section ""primitive logic rules"" you can call them ""logical meta-rules"". -Another confusion, also mentioned by reviewer 3 is that you are assuming that free variables (e.g. the ""x"" in the expression ""Clear(x)"") are implicitly considered universally quantified in your examples but you don't say this anywhere. If I have the fact ""Clear(x)"" as an input fact, then presumably you will interpret this as ""for all x Clear(x)"" and provide an input tensor to the first layer which will have all 1.0's along the ""Clear"" relation dimension, right? -It seems that you are making the assumption that you will never need to apply a predicate to the same object in multiple arguments? If not, I don't see why you say that the shape of the tensor will be m x (m-1) instead of m^2. You need to be able to do this to get reflexivity for example: ""a <= a"". -I think you are implicitly making the closed world assumption (CWA) and should say so. -On pg. 4 you say ""The facts are tensors that encode relations among multiple objectives, as described in Sec. 2.2."". What do you mean by ""objectives""? I would say the facts are tensors that encode relations among multiple objects. -On pg. 5 you say ""We finish this subsection, continuing with the blocks world to illustrate the forward propagation in NLM"". I see no mention of blocks world in this paragraph. It just seems like a description of what happens at one block, generically. -In many places you say that this model can compute deduction on first-order predicate calculus (FOPC) but it seems to me that you are limited to horn logic (rule logic) in which there is at most one positive literal per clause (i.e. rules of the form: b1 AND b2 AND ... AND bn => h). From what I can tell you cannot handle deduction on clauses such as b1 AND b2 => h1 or (h2 and h3). -There is not enough description of the exact setup for each experiment. For example in blocks world, how do you choose predicates for each layer? How many exactly for each experiment? You make it seem on p3 that you can handle recursive predicates but this seems to not have been worked out completely in the appendix. You should make this clear. -In figure 1 you list Move as if its a predicate like On but it's a very different thing. On is predicate describing a relation in one state. Move is an action which updates a state by changing the values of predicates. They should not be presented in the same way. -You use ""min"" and ""max"" for ""and"" and ""or"" respectively. Other approaches have found that using the product t-norm t-norm(x,y) = x * y helps with gradient propagation. del-ILP discusses this in more detail on p 19. Did you try these variations? -I think it would be helpful to somewhere explicitly describe the actual MLP model you use for deduction including layer sizes and activation functions. -p. 5. typo: ""Such a parameter sharing mechanism is crucial to the generalization ability of NLM to problems ov varying sizes."" (""ov"" -> ""of"") -p. 6. sec 3.1 typo: ""For ILP, the set of pre-conditions of the symbols is used direclty as input of the system."" (""direclty"" -> ""directly"") I think this is a valuable contribution and novel in the particulars of the architecture (eg. expand/reduce) and am recommending acceptance. But I would like to see a real effort made to sharpen the writing and make the exposition crystal clear. Please in particular pay attention to Reviewer 3's comments. """
1054,"""An Adversarial Learning Framework for a Persona-based Multi-turn Dialogue Model""","['conversation model', 'dialogue system', 'adversarial net', 'persona']","""In this paper, we extend the persona-based sequence-to-sequence (Seq2Seq) neural network conversation model to a multi-turn dialogue scenario by modifying the state-of-the-art hredGAN architecture to simultaneously capture utterance attributes such as speaker identity, dialogue topic, speaker sentiments and so on. The proposed system, phredGAN has a persona-based HRED generator (PHRED) and a conditional discriminator. We also explore two approaches to accomplish the conditional discriminator: (1) pseudo-formula , a system that passes the attribute representation as an additional input into a traditional adversarial discriminator, and (2) pseudo-formula , a dual discriminator system which in addition to the adversarial discriminator, collaboratively predicts the attribute(s) that generated the input utterance. To demonstrate the superior performance of phredGAN over the persona SeqSeq model, we experiment with two conversational datasets, the Ubuntu Dialogue Corpus (UDC) and TV series transcripts from the Big Bang Theory and Friends. Performance comparison is made with respect to a variety of quantitative measures as well as crowd-sourced human evaluation. We also explore the trade-offs from using either variant of pseudo-formula on datasets with many but weak attribute modalities (such as with Big Bang Theory and Friends) and ones with few but strong attribute modalities (customer-agent interactions in Ubuntu dataset).""","""This work presents extensions of dialogue systems to simultaneously capture speakers' ""personas"" (in the framing of Li et al's work) and adapt to them. While the ideas are interesting, reviewers note that the incremental contribution compared to previous work is a bit too limited for ICLR's expectation, without being offset by strongly convincing experimental results. Authors are encouraged to incorporated their ideas into future submissions after having combined them with other insights to provide a stronger overall contribution."""
1055,"""Generalization and Regularization in DQN""","['generalization', 'reinforcement learning', 'dqn', 'regularization', 'transfer learning', 'multitask']","""Deep reinforcement learning (RL) algorithms have shown an impressive ability to learn complex control policies in high-dimensional environments. However, despite the ever-increasing performance on popular benchmarks like the Arcade Learning Environment (ALE), policies learned by deep RL algorithms can struggle to generalize when evaluated in remarkably similar environments. These results are unexpected given the fact that, in supervised learning, deep neural networks often learn robust features that generalize across tasks. In this paper, we study the generalization capabilities of DQN in order to aid in understanding this mismatch between generalization in deep RL and supervised learning methods. We provide evidence suggesting that DQN overspecializes to the domain it is trained on. We then comprehensively evaluate the impact of traditional methods of regularization from supervised learning, pseudo-formula and dropout, and of reusing learned representations to improve the generalization capabilities of DQN. We perform this study using different game modes of Atari 2600 games, a recently introduced modification for the ALE which supports slight variations of the Atari 2600 games used for benchmarking in the field. Despite regularization being largely underutilized in deep RL, we show that it can, in fact, help DQN learn more general features. These features can then be reused and fine-tuned on similar tasks, considerably improving the sample efficiency of DQN.""","""The authors have presented an empirical study of generalization and regularization in DQN. They evaluate generalization on different variants of Atari games and show that dropout and L2 regularization are beneficial. The paper does not contain any major revelations, nor does it propose new algorithms or approaches, but it is a well-written and clear demonstration, and it would be interesting to the deep RL community. However, the reviewers did not feel that the paper met the bar for publication at ICLR because the experiments were not more comprehensive, which would be expected for an empirical study. The AC will side with the reviewers but hopes that the authors will expand their study and resubmit to another venue in the future."""
1056,"""Multi-Agent Dual Learning""","['Dual Learning', 'Machine Learning', 'Neural Machine Translation']","""Dual learning has attracted much attention in machine learning, computer vision and natural language processing communities. The core idea of dual learning is to leverage the duality between the primal task (mapping from domain X to domain Y) and dual task (mapping from domain Y to X) to boost the performances of both tasks. Existing dual learning framework forms a system with two agents (one primal model and one dual model) to utilize such duality. In this paper, we extend this framework by introducing multiple primal and dual models, and propose the multi-agent dual learning framework. Experiments on neural machine translation and image translation tasks demonstrate the effectiveness of the new framework. In particular, we set a new record on IWSLT 2014 German-to-English translation with a 35.44 BLEU score, achieve a 31.03 BLEU score on WMT 2014 English-to-German translation with over 2.6 BLEU improvement over the strong Transformer baseline, and set a new record of 49.61 BLEU score on the recent WMT 2018 English-to-German translation.""","""A paper that studies two tasks: machine translation and image translation. The authors propose a new multi-agent dual learning technique that takes advantage of the symmetry of the problem. The empirical gains over a competitive baseline are quite solid. The reviewers consistently liked the paper but have in some cases fairly low confidence in their assessment."""
1057,"""Learning to control self-assembling morphologies: a study of generalization via modularity""","['modularity', 'compostionality', 'graphs', 'dynamics', 'network']","""Much of contemporary sensorimotor learning assumes that one is already given a complex agent (e.g., a robotic arm) and the goal is to learn to control it. In contrast, this paper investigates a modular co-evolution strategy: a collection of primitive agents learns to self-assemble into increasingly complex collectives in order to solve control tasks. Each primitive agent consists of a limb and a neural controller. Limbs may choose to link up to form collectives, with linking being treated as a dynamic action. When two limbs link, a joint is added between them, actuated by the 'parent' limb's controller. This forms a new 'single' agent, which may further link with other agents. In this way, complex morphologies can emerge, controlled by a policy whose architecture is in explicit correspondence with the morphology. In experiments, we demonstrate that agents with these modular and dynamic topologies generalize better to test-time environments compared to static and monolithic baselines. Project videos are available at pseudo-url""","""Strengths: A co-evolution of body connectivity and its topology mimicing control policy is presented. Weaknesses: Reviewers found the paper to be lacking in detail. The importance of message passing in achieving the given results is clear on one example but not some others. Some reviewers had questions regarding the baseline comparisons. The authors provided lengthy details in responses on the discussion board, but reviewers likely had limited time to fully reread the many changes that were listed. AC: The physics in the motions shown in the video require signficant further explanation. It looks like the ball joints can directly attach themselves to the ground, and make that link stand up. Thus it seems that the robots are not underactuated and can effectively grab arbitrary points in the environment. Also it is strange to see the robot parts dynamically fly together as if attracted by a magnet. The physics needs significant further explanation. Points of Contention: The R2 review is positive on the paper (7), with a moderate confidence (3). R1 contributed additional questions during the discussion, but R2 and R3 were silent. The AC further examined the submission (paper and video). The reviewers and the AC are in consensus regarding the many details that are behind the system that are still not understood. The AC is also skeptical of the non-physical nature of the motion, or the unspecified behavior of fully-actuated contacts with the ground. """
1058,"""Adversarial Attacks for Optical Flow-Based Action Recognition Classifiers""","['adversarial attacks', 'action recognition', 'video classification']","""The success of deep learning research has catapulted deep models into production systems that our society is becoming increasingly dependent on, especially in the image and video domains. However, recent work has shown that these largely uninterpretable models exhibit glaring security vulnerabilities in the presence of an adversary. In this work, we develop a powerful untargeted adversarial attack for action recognition systems in both white-box and black-box settings. Action recognition models differ from image-classification models in that their inputs contain a temporal dimension, which we explicitly target in the attack. Drawing inspiration from image classifier attacks, we create new attacks which achieve state-of-the-art success rates on a two-stream classifier trained on the UCF-101 dataset. We find that our attacks can significantly degrade a models performance with sparsely and imperceptibly perturbed examples. We also demonstrate the transferability of our attacks to black-box action recognition systems.""","""With scores of 5, 4 and 3 the paper is just too far away from the threshold for acceptance."""
1059,"""Learning to Search Efficient DenseNet with Layer-wise Pruning""","['reinforcement learning', 'DenseNet', 'neural network compression']","""Deep neural networks have achieved outstanding performance in many real-world applications with the expense of huge computational resources. The DenseNet, one of the recently proposed neural network architecture, has achieved the state-of-the-art performance in many visual tasks. However, it has great redundancy due to the dense connections of the internal structure, which leads to high computational costs in training such dense networks. To address this issue, we design a reinforcement learning framework to search for efficient DenseNet architectures with layer-wise pruning (LWP) for different tasks, while retaining the original advantages of DenseNet, such as feature reuse, short paths, etc. In this framework, an agent evaluates the importance of each connection between any two block layers, and prunes the redundant connections. In addition, a novel reward-shaping trick is introduced to make DenseNet reach a better trade-off between accuracy and float point operations (FLOPs). Our experiments show that DenseNet with LWP is more compact and efficient than existing alternatives. ""","""The paper proposes to apply Neural Architecture Search for pruning DenseNet. The reviewers and AC note the potential weaknesses of the paper in various aspects, and decided that the authors need more works to publish. """
1060,"""ClariNet: Parallel Wave Generation in End-to-End Text-to-Speech""","['text-to-speech', 'deep generative models', 'end-to-end training', 'text to waveform']","""In this work, we propose a new solution for parallel wave generation by WaveNet. In contrast to parallel WaveNet (van Oord et al., 2018), we distill a Gaussian inverse autoregressive flow from the autoregressive WaveNet by minimizing a regularized KL divergence between their highly-peaked output distributions. Our method computes the KL divergence in closed-form, which simplifies the training algorithm and provides very efficient distillation. In addition, we introduce the first text-to-wave neural architecture for speech synthesis, which is fully convolutional and enables fast end-to-end training from scratch. It significantly outperforms the previous pipeline that connects a text-to-spectrogram model to a separately trained WaveNet (Ping et al., 2018). We also successfully distill a parallel waveform synthesizer conditioned on the hidden representation in this end-to-end model.""","""The authors discuss an improved distillation scheme for parallel WaveNet using a Gaussian inverse autoregressive flow, which can be computed in closed-form, thus simplifying training. The work received favorable comments from the reviewers, along with a number of suggestions for improvement which have improved the draft considerably. The AC agrees with the reviewers that the work is a valuable contribution, particularly in the context of end-to-end neural text-to-speech systems. """
1061,"""Reinforced Imitation Learning from Observations""","['imitation learning', 'state-only observations', 'self-exploration']","""Imitation learning is an effective alternative approach to learn a policy when the reward function is sparse. In this paper, we consider a challenging setting where an agent has access to a sparse reward function and state-only expert observations. We propose a method which gradually balances between the imitation learning cost and the reinforcement learning objective. Built upon an existing imitation learning method, our approach works with state-only observations. We show, through navigation scenarios, that (i) an agent is able to efficiently leverage sparse rewards to outperform standard state-only imitation learning, (ii) it can learn a policy even when learner's actions are different from the expert, and (iii) the performance of the agent is not bounded by that of the expert due to the optimized usage of sparse rewards.""","""This paper proposes to combine rewards obtained through IRL from rewards coming from the environment, and evaluate the algorithm on grid world environments. The problem setting is important and of interest to the ICLR community. While the revised paper addresses the concerns about the lack of a stochastic environment problem, the reviewers still have major concerns regarding the novelty and significance of the algorithmic contribution, as well as the limited complexity of the experimental domains. As such, the paper does not meet the bar for publication at ICLR."""
1062,"""Strength in Numbers: Trading-off Robustness and Computation via Adversarially-Trained Ensembles""","['adversarial examples', 'adversarial robustness', 'visualisation', 'ensembles']","""While deep learning has led to remarkable results on a number of challenging problems, researchers have discovered a vulnerability of neural networks in adversarial settings, where small but carefully chosen perturbations to the input can make the models produce extremely inaccurate outputs. This makes these models particularly unsuitable for safety-critical application domains (e.g. self-driving cars) where robustness is extremely important. Recent work has shown that augmenting training with adversarially generated data provides some degree of robustness against test-time attacks. In this paper we investigate how this approach scales as we increase the computational budget given to the defender. We show that increasing the number of parameters in adversarially-trained models increases their robustness, and in particular that ensembling smaller models while adversarially training the entire ensemble as a single model is a more efficient way of spending said budget than simply using a larger single model. Crucially, we show that it is the adversarial training of the ensemble, rather than the ensembling of adversarially trained models, which provides robustness.""","""The work brings little novelty compared to existing literature. """
1063,"""Improving MMD-GAN Training with Repulsive Loss Function""","['generative adversarial nets', 'loss function', 'maximum mean discrepancy', 'image generation', 'unsupervised learning']","""Generative adversarial nets (GANs) are widely used to learn the data sampling process and their performance may heavily depend on the loss functions, given a limited computational budget. This study revisits MMD-GAN that uses the maximum mean discrepancy (MMD) as the loss function for GAN and makes two contributions. First, we argue that the existing MMD loss function may discourage the learning of fine details in data as it attempts to contract the discriminator outputs of real data. To address this issue, we propose a repulsive loss function to actively learn the difference among the real data by simply rearranging the terms in MMD. Second, inspired by the hinge loss, we propose a bounded Gaussian kernel to stabilize the training of MMD-GAN with the repulsive loss function. The proposed methods are applied to the unsupervised image generation tasks on CIFAR-10, STL-10, CelebA, and LSUN bedroom datasets. Results show that the repulsive loss function significantly improves over the MMD loss at no additional computational cost and outperforms other representative loss functions. The proposed methods achieve an FID score of 16.21 on the CIFAR-10 dataset using a single DCGAN network and spectral normalization.""","""The submission proposes two new things: a repulsive loss for MMD loss optimization and a bounded RBF kernel that stabilizes training of MMD-GAN. The submission has a number of unsupervised image modeling experiments on standard benchmarks and shows reasonable performance. All in all, this is an interesting piece of work that has a number of interesting ideas (e.g. the PICO method, which is useful to know). I agree with R2 that the RBF kernel seems somewhat hacky in its introduction, despite working well in practice. That being said, the repulsive loss seems like something the research community would benefit from finding out more about, and I think the experiments and discussion are sufficiently extensive to warrant publication."""
1064,"""A Study of Robustness of Neural Nets Using Approximate Feature Collisions""",[],"""In recent years, various studies have focused on the robustness of neural nets. While it is known that neural nets are not robust to examples with adversarially chosen perturbations as a result of linear operations on the input data, we show in this paper there could be a convex polytope within which all examples are misclassified by neural nets due to the properties of ReLU activation functions. We propose a way to find such polytopes empirically and demonstrate that such polytopes exist in practice. Furthermore, we show that such polytopes exist even after constraining the examples to be a composition of image patches, resulting in perceptibly different examples at different locations in the polytope that are all misclassified. ""","""The paper presents a novel view on adversarial examples, where models using ReLU are inherently sensitive to adversarial examples because ReLU activations yield a polytope of examples with exactly the same activation. Reviewers found the finding interesting and novel but argue it is limited in impact. I also found the idea interesting but the paper could probably be improved as all reviewers have remarked. Overall, I found it borderline but probably not enough for acceptance."""
1065,"""Convergence Properties of Deep Neural Networks on Separable Data""","['learning dynamics', 'gradient descent', 'classification', 'optimization', 'cross-entropy', 'hinge loss', 'implicit regularization', 'gradient starvation']","""While a lot of progress has been made in recent years, the dynamics of learning in deep nonlinear neural networks remain to this day largely misunderstood. In this work, we study the case of binary classification and prove various properties of learning in such networks under strong assumptions such as linear separability of the data. Extending existing results from the linear case, we confirm empirical observations by proving that the classification error also follows a sigmoidal shape in nonlinear architectures. We show that given proper initialization, learning expounds parallel independent modes and that certain regions of parameter space might lead to failed training. We also demonstrate that input norm and features' frequency in the dataset lead to distinct convergence speeds which might shed some light on the generalization capabilities of deep neural networks. We provide a comparison between the dynamics of learning with cross-entropy and hinge losses, which could prove useful to understand recent progress in the training of generative adversarial networks. Finally, we identify a phenomenon that we baptize gradient starvation where the most frequent features in a dataset prevent the learning of other less frequent but equally informative features.""","""The manuscript proposes to analyze the learning dynamics of deep networks with separable data. A variety of results are provided under various assumptions. The reviewers and AC note the assumptions required for the analysis are quite strong, and perhaps too strong to provide useful insight into real problems. Reviewers also cite issues with writing and the breadth of the title (this was much improved after rebuttal)."""
1066,"""Using Ontologies To Improve Performance In Massively Multi-label Prediction""","['multi-label', 'Bayesian network', 'ontology']","""Massively multi-label prediction/classification problems arise in environments like health-care or biology where it is useful to make very precise predictions. One challenge with massively multi-label problems is that there is often a long-tailed frequency distribution for the labels, resulting in few positive examples for the rare labels. We propose a solution to this problem by modifying the output layer of a neural network to create a Bayesian network of sigmoids which takes advantage of ontology relationships between the labels to help share information between the rare and the more common labels. We apply this method to the two massively multi-label tasks of disease prediction (ICD-9 codes) and protein function prediction (Gene Ontology terms) and obtain significant improvements in per-label AUROC and average precision.""","""The paper proposes a nice approach to massively multi-label problems with rare labels which may only have a limited number of positive examples; the approach uses Bayes nets to exploit the relationships among the labels in the output layer of a neural nets. The paper is clearly written and the approach seems promising, however, the reviewers would like to see even more convincing empirical results. """
1067,"""Unicorn: Continual learning with a universal, off-policy agent""","['reinforcement learning', 'continual learning', 'universal value functions', 'off-policy learning', 'multi-task']","""Some real-world domains are best characterized as a single task, but for others this perspective is limiting. Instead, some tasks continually grow in complexity, in tandem with the agent's competence. In continual learning there are no explicit task boundaries or curricula. As learning agents have become more powerful, continual learning remains one of the frontiers that has resisted quick progress. To test continual learning capabilities we consider a challenging 3D domain with an implicit sequence of tasks and sparse rewards. We propose a novel agent architecture called Unicorn, which demonstrates strong continual learning and outperforms several baseline agents on the proposed domain. The agent achieves this by jointly representing and efficiently learning multiple policies for multiple goals, using a parallel off-policy learning setup. ""","""The authors present an interesting approach but there were multiple significant concerns with the clarity of the presentation, and some concern with the significance of the experimental results."""
1068,"""Mixture of Pre-processing Experts Model for Noise Robust Deep Learning on Resource Constrained Platforms""","['noise robust', 'object detection']","""Deep learning on an edge device requires energy efficient operation due to ever diminishing power budget. Intentional low quality data during the data acquisition for longer battery life, and natural noise from the low cost sensor degrade the quality of target output which hinders adoption of deep learning on an edge device. To overcome these problems, we propose simple yet efficient mixture of pre-processing experts (MoPE) model to handle various image distortions including low resolution and noisy images. We also propose to use adversarially trained auto encoder as a pre-processing expert for the noisy images. We evaluate our proposed method for various machine learning tasks including object detection on MS-COCO 2014 dataset, multiple object tracking problem on MOT-Challenge dataset, and human activity recognition on UCF 101 dataset. Experimental results show that the proposed method achieves better detection, tracking and activity recognition accuracies under noise without sacrificing accuracies for the clean images. The overheads of our proposed MoPE are 0.67% and 0.17% in terms of memory and computation compared to the baseline object detection network.""","""As the reviewers point out, the paper seems to be below the ICLR publication bar due to low novelty and limited significance. """
1069,"""Robustness May Be at Odds with Accuracy""","['adversarial examples', 'robust machine learning', 'robust optimization', 'deep feature representations']","""We show that there exists an inherent tension between the goal of adversarial robustness and that of standard generalization. Specifically, training robust models may not only be more resource-consuming, but also lead to a reduction of standard accuracy. We demonstrate that this trade-off between the standard accuracy of a model and its robustness to adversarial perturbations provably exists even in a fairly simple and natural setting. These findings also corroborate a similar phenomenon observed in practice. Further, we argue that this phenomenon is a consequence of robust classifiers learning fundamentally different feature representations than standard classifiers. These differences, in particular, seem to result in unexpected benefits: the features learned by robust models tend to align better with salient data characteristics and human perception.""","""This paper provides interesting discussions on the trade-off between model accuracy and robustness to adversarial examples. All reviewers found that both empirical studies and theoretical results are solid. The paper is very well written. The visualization results are very intuitive. I recommend acceptance. """
1070,"""DEEP-TRIM: REVISITING L1 REGULARIZATION FOR CONNECTION PRUNING OF DEEP NETWORK""","['L1 regularization', 'deep neural network', 'deep compression']","""State-of-the-art deep neural networks (DNNs) typically have tens of millions of parameters, which might not fit into the upper levels of the memory hierarchy, thus increasing the inference time and energy consumption significantly, and prohibiting their use on edge devices such as mobile phones. The compression of DNN models has therefore become an active area of research recently, with \emph{connection pruning} emerging as one of the most successful strategies. A very natural approach is to prune connections of DNNs via pseudo-formula regularization, but recent empirical investigations have suggested that this does not work as well in the context of DNN compression. In this work, we revisit this simple strategy and analyze it rigorously, to show that: (a) any \emph{stationary point} of an pseudo-formula -regularized layerwise-pruning objective has its number of non-zero elements bounded by the number of penalized prediction logits, regardless of the strength of the regularization; (b) successful pruning highly relies on an accurate optimization solver, and there is a trade-off between compression speed and distortion of prediction accuracy, controlled by the strength of regularization. Our theoretical results thus suggest that pseudo-formula pruning could be successful provided we use an accurate optimization solver. We corroborate this in our experiments, where we show that simple pseudo-formula regularization with an Adamax-L1(cumulative) solver gives pruning ratio competitive to the state-of-the-art.""","""This paper studies the properties of L1 regularization for deep neural network. It contains some interesting results, e.g. the stationary point of an l1 regularized layer has bounded number of non-zero elements. On the other hand, the majority of reviewers has concerns on that experimental supports are weak and suggests rejection. Therefore, a final rejection is proposed."""
1071,"""Intrinsic Social Motivation via Causal Influence in Multi-Agent RL""","['multi-agent reinforcement learning', 'causal inference', 'game theory', 'social dilemma', 'intrinsic motivation', 'counterfactual reasoning', 'empowerment', 'communication']","""We derive a new intrinsic social motivation for multi-agent reinforcement learning (MARL), in which agents are rewarded for having causal influence over another agent's actions, where causal influence is assessed using counterfactual reasoning. The reward does not depend on observing another agent's reward function, and is thus a more realistic approach to MARL than taken in previous work. We show that the causal influence reward is related to maximizing the mutual information between agents' actions. We test the approach in challenging social dilemma environments, where it consistently leads to enhanced cooperation between agents and higher collective reward. Moreover, we find that rewarding influence can lead agents to develop emergent communication protocols. Therefore, we also employ influence to train agents to use an explicit communication channel, and find that it leads to more effective communication and higher collective reward. Finally, we show that influence can be computed by equipping each agent with an internal model that predicts the actions of other agents. This allows the social influence reward to be computed without the use of a centralised controller, and as such represents a significantly more general and scalable inductive bias for MARL with independent agents.""","""The reviewers raised a number of concerns including the appropriateness of the chosen application and the terms in which social dilemmas have been discussed, the lack of explanations and discussions, missing references, and the extent of the evaluation studies. The authors rebuttal addressed some of the reviewers concerns but not fully. Overall, I believe that the work is interesting and may be useful to the community (though to a small extent., in my opinion). However, the paper would benefit from additional explanations, experiments and discussions pointed in quite some detail by the reviewers. AS is, the paper is below the acceptance threshold for presentation at ICLR."""
1072,"""MahiNet: A Neural Network for Many-Class Few-Shot Learning with Class Hierarchy""","['deep learning', 'many-class few-shot', 'class hierarchy', 'meta learning']","""We study many-class few-shot (MCFS) problem in both supervised learning and meta-learning scenarios. Compared to the well-studied many-class many-shot and few-class few-shot problems, MCFS problem commonly occurs in practical applications but is rarely studied. MCFS brings new challenges because it needs to distinguish between many classes, but only a few samples per class are available for training. In this paper, we propose ``memory-augmented hierarchical-classification network (MahiNet)'' for MCFS learning. It addresses the ``many-class'' problem by exploring the class hierarchy, e.g., the coarse-class label that covers a subset of fine classes, which helps to narrow down the candidates for the fine class and is cheaper to obtain. MahiNet uses a convolutional neural network (CNN) to extract features, and integrates a memory-augmented attention module with a multi-layer perceptron (MLP) to produce the probabilities over coarse and fine classes. While the MLP extends the linear classifier, the attention module extends a KNN classifier, both together targeting the ''`few-shot'' problem. We design different training strategies of MahiNet for supervised learning and meta-learning. Moreover, we propose two novel benchmark datasets ''mcfsImageNet'' (as a subset of ImageNet) and ''mcfsOmniglot'' (re-splitted Omniglot) specifically for MCFS problem. In experiments, we show that MahiNet outperforms several state-of-the-art models on MCFS classification tasks in both supervised learning and meta-learning scenarios.""","""The reviewers raised a number of concerns including low readability/ clarity of the presented work and methodology, insufficient and at times unconvincing experimental evaluation of the proposed, and lack of discussion on pros and cons of the presented. The authors rebuttal addressed some of the reviewers comments but failed to address all concerns and reconfirmed that relatively large changes are still needed for the paper to be useful to the readers. Hence, although I believe this could be a very interesting paper, I cannot suggest it at this stage for presentation at ICLR."""
1073,"""Environment Probing Interaction Policies""",['Reinforcement Learning'],"""A key challenge in reinforcement learning (RL) is environment generalization: a policy trained to solve a task in one environment often fails to solve the same task in a slightly different test environment. A common approach to improve inter-environment transfer is to learn policies that are invariant to the distribution of testing environments. However, we argue that instead of being invariant, the policy should identify the specific nuances of an environment and exploit them to achieve better performance. In this work, we propose the Environment-Probing Interaction (EPI) policy, a policy that probes a new environment to extract an implicit understanding of that environments behavior. Once this environment-specific information is obtained, it is used as an additional input to a task-specific policy that can now perform environment-conditioned actions to solve a task. To learn these EPI-policies, we present a reward function based on transition predictability. Specifically, a higher reward is given if the trajectory generated by the EPI-policy can be used to better predict transitions. We experimentally show that EPI-conditioned task-specific policies significantly outperform commonly used policy generalization methods on novel testing environments.""","""This paper proposes an approach for probing an environment to quickly identify the dynamics. The problem is relevant to the ICLR community. The paper is well-written, and provides a detailed empirical evaluation. The main weakness of the paper is the somewhat small originality over prior methods on online system identification. Despite this, the reviewer's agreed that the paper exceeds the bar for publication at ICLR. Hence, I recommend accept. Beyond the related work mentioned by the reviewers, the approach is similar to work in meta-learning. Meta-RL and multi-task learning has typically been considered in settings where the reward is changing (e.g. see [1],[2],[3],[4], where [4] also uses an embedding-based approach). However, there is some more recent work on meta-RL across varying dynamics, e.g. see [5],[6]. The authors are encouraged to make a conceptual connection between this approach and the line of work in model-based meta-RL (particularly [5] and [6]) in the final version of the paper. [1] Duan et al. pseudo-url [2] Wang et al. CogSci '17 pseudo-url [3] Finn et al. ICML '17 pseudo-url [4] Hausman et al. ICLR '17: pseudo-url [5] Smundsson et al. pseudo-url [6] Nagabandi et al. pseudo-url """
1074,"""Phrase-Based Attentions""","['neural machine translation', 'natural language processing', 'attention', 'transformer', 'seq2seq', 'phrase-based', 'phrase', 'n-gram']","""Most state-of-the-art neural machine translation systems, despite being different in architectural skeletons (e.g., recurrence, convolutional), share an indispensable feature: the Attention. However, most existing attention methods are token-based and ignore the importance of phrasal alignments, the key ingredient for the success of phrase-based statistical machine translation. In this paper, we propose novel phrase-based attention methods to model n-grams of tokens as attention entities. We incorporate our phrase-based attentions into the recently proposed Transformer network, and demonstrate that our approach yields improvements of 1.3 BLEU for English-to-German and 0.5 BLEU for German-to-English translation tasks, and 1.75 and 1.35 BLEU points in English-to-Russian and Russian-to-English translation tasks on WMT newstest2014 using WMT16 training data. ""","""All reviewers agree in their assessment that this paper does not meet the bar for ICLR. The area chair commends the authors for their detailed responses."""
1075,"""Recycling the discriminator for improving the inference mapping of GAN""",[],"""Generative adversarial networks (GANs) have achieved outstanding success in generating the high-quality data. Focusing on the generation process, existing GANs learn a unidirectional mapping from the latent vector to the data. Later, various studies point out that the latent space of GANs is semantically meaningful and can be utilized in advanced data analysis and manipulation. In order to analyze the real data in the latent space of GANs, it is necessary to investigate the inverse generation mapping from the data to the latent vector. To tackle this problem, the bidirectional generative models introduce an encoder to establish the inverse path of the generation process. Unfortunately, this effort leads to the degradation of generation quality because the imperfect generator rather interferes the encoder training and vice versa. In this paper, we propose an effective algorithm to infer the latent vector based on existing unidirectional GANs by preserving their generation quality. It is important to note that we focus on increasing the accuracy and efficiency of the inference mapping but not influencing the GAN performance (i.e., the quality or the diversity of the generated sample). Furthermore, utilizing the proposed inference mapping algorithm, we suggest a new metric for evaluating the GAN models by measuring the reconstruction error of unseen real data. The experimental analysis demonstrates that the proposed algorithm achieves more accurate inference mapping than the existing method and provides the robust metric for evaluating GAN performance. ""","""The paper presents a method to learn inference mapping for GANs by reusing the learned discriminator's features and fitting a model over these features to reconstruct the original latent code z. R1 pointed out the connection to InfoGAN which the authors have addressed. R2 is concerned about limited novelty of the proposed method, which the AC agrees with, and lack of comparison to a related iGAN work by Zhu et al. (2016). The authors have provided the comparison in the revised version but the proposed method seems to be worse than iGAN in terms of the metrics used (PSNR and SSIM), though more efficient. The benefits of using the proposed metrics for evaluating GAN quality are also not established well, particularly in the context of other recent metrics such as FID and GILBO. """
1076,"""Variadic Learning by Bayesian Nonparametric Deep Embedding""","['meta-learning', 'metric learning', 'bayesian nonparametrics', 'few-shot learning', 'deep learning']","""Learning at small or large scales of data is addressed by two strong but divided frontiers: few-shot learning and standard supervised learning. Few-shot learning focuses on sample efficiency at small scale, while supervised learning focuses on accuracy at large scale. Ideally they could be reconciled for effective learning at any number of data points (shot) and number of classes (way). To span the full spectrum of shot and way, we frame the variadic learning regime of learning from any number of inputs. We approach variadic learning by meta-learning a novel multi-modal clustering model that connects bayesian nonparametrics and deep metric learning. Our bayesian nonparametric deep embedding (BANDE) method is optimized end-to-end with a single objective, and adaptively adjusts capacity to learn from variable amounts of supervision. We show that multi-modality is critical for learning complex classes such as Omniglot alphabets and carrying out unsupervised clustering. We explore variadic learning by measuring generalization across shot and way between meta-train and meta-test, show the first results for scaling from few-way, few-shot tasks to 1692-way Omniglot classification and 5k-shot CIFAR-10 classification, and find that nonparametric methods generalize better than parametric methods. On the standard few-shot learning benchmarks of Omniglot and mini-ImageNet, BANDE equals or improves on the state-of-the-art for semi-supervised classification.""","""All reviewers wrote strong and long reviews with good feedback but do not believe the work is currently ready for publication. I encourage the authors to update and resubmit. """
1077,"""Query-Efficient Hard-label Black-box Attack: An Optimization-based Approach""","['Adversarial example', 'Hard-label', 'Black-box attack', 'Query-efficient']","""We study the problem of attacking machine learning models in the hard-label black-box setting, where no model information is revealed except that the attacker can make queries to probe the corresponding hard-label decisions. This is a very challenging problem since the direct extension of state-of-the-art white-box attacks (e.g., C&W or PGD) to the hard-label black-box setting will require minimizing a non-continuous step function, which is combinatorial and cannot be solved by a gradient-based optimizer. The only two current approaches are based on random walk on the boundary (Brendel et al., 2017) and random trials to evaluate the loss function (Ilyas et al., 2018), which require lots of queries and lacks convergence guarantees. We propose a novel way to formulate the hard-label black-box attack as a real-valued optimization problem which is usually continuous and can be solved by any zeroth order optimization algorithm. For example, using the Randomized Gradient-Free method (Nesterov & Spokoiny, 2017), we are able to bound the number of iterations needed for our algorithm to achieve stationary points under mild assumptions. We demonstrate that our proposed method outperforms the previous stochastic approaches to attacking convolutional neural networks on MNIST, CIFAR, and ImageNet datasets. More interestingly, we show that the proposed algorithm can also be used to attack other discrete and non-continuous machine learning models, such as Gradient Boosting Decision Trees (GBDT).""","""The reviewers liked the clarity of the material and agreed the experimental study is convincing. Accept."""
1078,"""Ain't Nobody Got Time for Coding: Structure-Aware Program Synthesis from Natural Language""","['Program synthesis', 'tree2tree autoencoders', 'soft attention', 'doubly-recurrent neural networks', 'LSTM', 'nlp2tree']","""Program synthesis from natural language (NL) is practical for humans and, once technically feasible, would significantly facilitate software development and revolutionize end-user programming. We present SAPS, an end-to-end neural network capable of mapping relatively complex, multi-sentence NL specifications to snippets of executable code. The proposed architecture relies exclusively on neural components, and is built upon a tree2tree autoencoder trained on abstract syntax trees, combined with a pretrained word embedding and a bi-directional multi-layer LSTM for NL processing. The decoder features a doubly-recurrent LSTM with a novel signal propagation scheme and soft attention mechanism. When applied to a large dataset of problems proposed in a previous study, SAPS performs on par with or better than the method proposed there, producing correct programs in over 90% of cases. In contrast to other methods, it does not involve any non-neural components to post-process the resulting programs, and uses a fixed-dimensional latent representation as the only link between the NL analyzer and source code generator. ""","""The paper presents a new neural program synthesis architecture, SAPS, which seems to produce accuracy improvements in some synthesis tasks. The reviewer consensus, even after discussion with the authors, was that the paper is not acceptable at the conference. Two concerns emerge during discussion, even considering the authors efforts to improve the paper. First, the system seems to have many ""moving parts"", but there is a lack of rigorous ablation studies to demonstrate which components of the system (or combination thereof) make significant contributions to the results. I agree with this assessment: it is not sufficient to demonstrate increased scores, even if the experimental protocol and clear and sound (more on this later), but there must be some evidence as to why this increase happens, both in the discussion and in the empirical segment of the paper, by conducting a thorough ablation study. Second, all reviewers had issues with proper and fair comparison with prior work, with the consensus being that the model is not adequately compared to convincing benchmarks in the paper. The results of the paper sound like there is something promising going on, but the need for a clear presentation of what is the driving factor behind any improvement is not only a superficial stylistic requirement, but a key tenet of proper scholarship. This is one front on which the paper fails to make a successful case for the work and methods it describes, and unfortunately is not ready for publication at this time (despite having a cool title)."""
1079,"""SSoC: Learning Spontaneous and Self-Organizing Communication for Multi-Agent Collaboration""","['reinforcement learning', 'multi-agent learning', 'multi-agent communication', 'deep learning']","""Multi-agent collaboration is required by numerous real-world problems. Although distributed setting is usually adopted by practical systems, local range communication and information aggregation still matter in fulfilling complex tasks. For multi-agent reinforcement learning, many previous studies have been dedicated to design an effective communication architecture. However, existing models usually suffer from an ossified communication structure, e.g., most of them predefine a particular communication mode by specifying a fixed time frequency and spatial scope for agents to communicate regardless of necessity. Such design is incapable of dealing with multi-agent scenarios that are capricious and complicated, especially when only partial information is available. Motivated by this, we argue that the solution is to build a spontaneous and self-organizing communication (SSoC) learning scheme. By treating the communication behaviour as an explicit action, SSoC learns to organize communication in an effective and efficient way. Particularly, it enables each agent to spontaneously decide when and who to send messages based on its observed states. In this way, a dynamic inter-agent communication channel is established in an online and self-organizing manner. The agents also learn how to adaptively aggregate the received messages and its own hidden states to execute actions. Various experiments have been conducted to demonstrate that SSoC really learns intelligent message passing among agents located far apart. With such agile communications, we observe that effective collaboration tactics emerge which have not been mastered by the compared baselines.""","""The reviewers raised a number of major concerns including the incremental novelty of the proposed and a poor readability of the presented materials (lack of sufficient explanations and discussions). The authors decided to withdraw the paper."""
1080,"""PATE-GAN: Generating Synthetic Data with Differential Privacy Guarantees""","['Synthetic data generation', 'Differential privacy', 'Generative adversarial networks', 'Private Aggregation of Teacher ensembles']","""Machine learning has the potential to assist many communities in using the large datasets that are becoming more and more available. Unfortunately, much of that potential is not being realized because it would require sharing data in a way that compromises privacy. In this paper, we investigate a method for ensuring (differential) privacy of the generator of the Generative Adversarial Nets (GAN) framework. The resulting model can be used for generating synthetic data on which algorithms can be trained and validated, and on which competitions can be conducted, without compromising the privacy of the original dataset. Our method modifies the Private Aggregation of Teacher Ensembles (PATE) framework and applies it to GANs. Our modified framework (which we call PATE-GAN) allows us to tightly bound the influence of any individual sample on the model, resulting in tight differential privacy guarantees and thus an improved performance over models with the same guarantees. We also look at measuring the quality of synthetic data from a new angle; we assert that for the synthetic data to be useful for machine learning researchers, the relative performance of two algorithms (trained and tested) on the synthetic dataset should be the same as their relative performance (when trained and tested) on the original dataset. Our experiments, on various datasets, demonstrate that PATE-GAN consistently outperforms the state-of-the-art method with respect to this and other notions of synthetic data quality.""","""This paper improves upon the PATE-GAN framework for differentially-private synthetic data generation. They eliminate the need for public data samples for training the GAN, by providing a distribution which can be sampled from instead. The authors were unanimous in their vote to accept."""
1081,"""P^2IR: Universal Deep Node Representation via Partial Permutation Invariant Set Functions""","['graph embedding', 'set function', 'representation learning']","""Graph node representation learning is a central problem in social network analysis, aiming to learn the vector representation for each node in a graph. The key problem is how to model the dependence of each node to its neighbor nodes since the neighborhood can uniquely characterize a graph. Most existing approaches rely on defining the specific neighborhood dependence as the computation mechanism of representations, which may exclude important subtle structures within the graph and dependence among neighbors. Instead, we propose a novel graph node embedding method (namely P^2IR) via developing a novel notion, namely partial permutation invariant set function} to learn those subtle structures. Our method can 1) learn an arbitrary form of the representation function from the neighborhood, without losing any potential dependence structures, 2) automatically decide the significance of neighbors at different distances, and 3) be applicable to both homogeneous and heterogeneous graph embedding, which may contain multiple types of nodes. Theoretical guarantee for the representation capability of our method has been proved for general homogeneous and heterogeneous graphs. Evaluation results on benchmark data sets show that the proposed P^IR outperforms the state-of-the-art approaches on producing node vectors for classification tasks.""","""AR1 is concerned with the presentation of the paper and the complexity as well as missing discussion on recent embedding methods. AR2 is concerned about comparison to recent methods and the small size of datasets. AR3 is also concerned about limited comparisons and evaluations. Lastly, AR4 again points out the poor complexity due to the spectral decomposition. While authors argue that the sparsity can be exploited to speed up computations, AR4 still asks for results of the exact model with/without any approximation, effect of clipping spectrum, time complexity versus GCN, and more empirical results covering all these aspects. On balance, all reviewers seem to voice similar concerns which need to be resolved. However, this requires more than just a minor revision of the manuscript. Thus, at this time, the proposed paper cannot be accepted. """
1082,"""Information Regularized Neural Networks""","['supervised classification', 'information theory', 'deep learning', 'regularization']","""We formulate an information-based optimization problem for supervised classification. For invertible neural networks, the control of these information terms is passed down to the latent features and parameter matrix in the last fully connected layer, given that mutual information is invariant under invertible map. We propose an objective function and prove that it solves the optimization problem. Our framework allows us to learn latent features in an more interpretable form while improving the classification performance. We perform extensive quantitative and qualitative experiments in comparison with the existing state-of-the-art classification models.""","""This paper proposes an approach to regularizing classifiers based on invertible networks using concepts from the information bottleneck theory. Because mutual information is invariant under invertible maps, the regularizer only considers the latent representation produced by the last hidden layer in the network and the network parameters that transform that representation into a classification decision. This leads to a combined 1 regularization on the final weights, W, and 2 regularization on W^{T} F(x), where F(x) is the latent representation produced by the last hidden layer. Experiments on CIFAR-100 image classification show that the proposed regularization can improve test performance. The reviewers liked the theoretical analysis, especially proposition 2.1 and its proof, but even after discussion and revision wanted a more careful empirical comparison to established forms of regularization to establish that the proposed approach has practical merit. The authors are encouraged to continue this line of research, building on the fruitful discussions they had with the reviewers."""
1083,"""Emergent Coordination Through Competition""","['Multi-agent learning', 'Reinforcement Learning']","""We study the emergence of cooperative behaviors in reinforcement learning agents by introducing a challenging competitive multi-agent soccer environment with continuous simulated physics. We demonstrate that decentralized, population-based training with co-play can lead to a progression in agents' behaviors: from random, to simple ball chasing, and finally showing evidence of cooperation. Our study highlights several of the challenges encountered in large scale multi-agent training in continuous control. In particular, we demonstrate that the automatic optimization of simple shaping rewards, not themselves conducive to co-operative behavior, can lead to long-horizon team behavior. We further apply an evaluation scheme, grounded by game theoretic principals, that can assess agent performance in the absence of pre-defined evaluation tasks or human baselines.""","""The paper studies population-based training for MARL with co-play, in MuJoCo (continuous control) soccer. It shows that (long term) cooperative behaviors can emerge from simple rewards, shaped but not towards cooperation. The paper is overall well written and includes a thorough study/ablation. The weaknesses are the lack of strong comparisons (or at least easy to grasp baselines) on a new task, and the lack of some of the experimental details (about reward shaping, about hyperparameters). The reviewers reached an agreement. This paper is welcomed to be published at ICLR."""
1084,"""Top-Down Neural Model For Formulae""","['logic', 'formula', 'recursive neural networks', 'recurrent neural networks']","""We present a simple neural model that given a formula and a property tries to answer the question whether the formula has the given property, for example whether a propositional formula is always true. The structure of the formula is captured by a feedforward neural network recursively built for the given formula in a top-down manner. The results of this network are then processed by two recurrent neural networks. One of the interesting aspects of our model is how propositional atoms are treated. For example, the model is insensitive to their names, it only matters whether they are the same or distinct.""","""This paper presents a method for building representations of logical formulae not by propagating information upwards from leaves to root and making decisions (e.g. as to whether one formula entails another) based on the root representation, but rather by propagating information down from root to leaves. It is a somewhat curious approach, and it is interesting to see that it works so well, especially on the ""massive"" train/test split of Evans et al. (2018). This paper certainly piques my interest, and I was disappointed to see a complete absence of discussion from reviewers during the rebuttal period despite author responses. The reviewer scores are all middle-of-the-road scores lightly leaning towards accepting, so the paper is rather borderline. It would have been most helpful to hear what the reviewers thought of the rebuttal and revisions made to the paper. Having read through the paper myself, and through the reviews and rebuttal, I am hesitantly casting an extra vote in favour of acceptance: the sort of work discussed in this paper is important and under-represented in the conference, and the results are convincing. I however, share the concerns outlined by the reviewers in their first (and only) set of comments, and invite the authors to take particular heed of the points made by AnonReviewer3, although all make excellent points. There needs to be some further analysis and explanation of these results. If not in this paper, then at least in follow up work. For now, I will recommend with medium confidence that the paper be accepted."""
1085,"""Approximability of Discriminators Implies Diversity in GANs""","['Theory', 'Generative adversarial networks', 'Mode collapse', 'Generalization']","""While Generative Adversarial Networks (GANs) have empirically produced impressive results on learning complex real-world distributions, recent works have shown that they suffer from lack of diversity or mode collapse. The theoretical work of Arora et al. (2017a) suggests a dilemma about GANs statistical properties: powerful discriminators cause overfitting, whereas weak discriminators cannot detect mode collapse. By contrast, we show in this paper that GANs can in principle learn distributions in Wasserstein distance (or KL-divergence in many cases) with polynomial sample complexity, if the discriminator class has strong distinguishing power against the particular generator class (instead of against all possible generators). For various generator classes such as mixture of Gaussians, exponential families, and invertible and injective neural networks generators, we design corresponding discriminators (which are often neural nets of specific architectures) such that the Integral Probability Metric (IPM) induced by the discriminators can provably approximate the Wasserstein distance and/or KL-divergence. This implies that if the training is successful, then the learned distribution is close to the true distribution in Wasserstein distance or KL divergence, and thus cannot drop modes. Our preliminary experiments show that on synthetic datasets the test IPM is well correlated with KL divergence or the Wasserstein distance, indicating that the lack of diversity in GANs may be caused by the sub-optimality in optimization instead of statistical inefficiency.""","""The paper presents an interesting theoretical analysis by deriving polynomial sample complexity bounds for the training of GANs that depend on the approximator properties of the discriminator. Even if it is not clear if the theory will help to pick suitable discriminators in practice, it provides new and interesting theoretical insights on the properties of GAN training. """
1086,"""DeepOBS: A Deep Learning Optimizer Benchmark Suite""","['deep learning', 'optimization']","""Because the choice and tuning of the optimizer affects the speed, and ultimately the performance of deep learning, there is significant past and recent research in this area. Yet, perhaps surprisingly, there is no generally agreed-upon protocol for the quantitative and reproducible evaluation of optimization strategies for deep learning. We suggest routines and benchmarks for stochastic optimization, with special focus on the unique aspects of deep learning, such as stochasticity, tunability and generalization. As the primary contribution, we present DeepOBS, a Python package of deep learning optimization benchmarks. The package addresses key challenges in the quantitative assessment of stochastic optimizers, and automates most steps of benchmarking. The library includes a wide and extensible set of ready-to-use realistic optimization problems, such as training Residual Networks for image classification on ImageNet or character-level language prediction models, as well as popular classics like MNIST and CIFAR-10. The package also provides realistic baseline results for the most popular optimizers on these test problems, ensuring a fair comparison to the competition when benchmarking new optimizers, and without having to run costly experiments. It comes with output back-ends that directly produce LaTeX code for inclusion in academic publications. It supports TensorFlow and is available open source.""","""The field of deep learning optimization suffers from a lack of standard benchmarks, and every paper reports results on a different set of models and architectures, likely with different protocols for tuning the baselines. This paper takes the useful step of providing a single benchmark suite for neural net optimizers. The set of benchmarks seems well-designed, and covers the range of baselines with a variety of representative architectures. It seems like a useful contribution that will improve the rigor of neural net optimizer evaluation. One reviewer had a long back-and-forth with the authors about whether to provide a standard protocol for hyperparameter tuning. I side with the authors on this one: it seems like a bad idea to force a one-size-fits-all protocol here. As a lesser point, I'm a little concerned about the strength of some of the baselines. As reviewers point out, some of the baseline results are weaker than typical implementations of those methods. One explanation might be the lack of learning rate schedules, something that's critical to get reasonable performance on some of these tasks. I get that using a fixed learning rate simplifies the grid search protocol, but I'm worried it will hurt the baselines enough that effective learning rate schedules and normalization issues come to dominate the comparisons. Still, the benchmark suite seems well constructed on the whole, and will probably be useful for evaluation of neural net optimizers. I recommend acceptance. """
1087,"""Learning Diverse Generations using Determinantal Point Processes""",['Generative Adversarial Networks'],"""Generative models have proven to be an outstanding tool for representing high-dimensional probability distributions and generating realistic looking images. A fundamental characteristic of generative models is their ability to produce multi-modal outputs. However, while training, they are often susceptible to mode collapse, which means that the model is limited in mapping the input noise to only a few modes of the true data distribution. In this paper, we draw inspiration from Determinantal Point Process (DPP) to devise a generative model that alleviates mode collapse while producing higher quality samples. DPP is an elegant probabilistic measure used to model negative correlations within a subset and hence quantify its diversity. We use DPP kernel to model the diversity in real data as well as in synthetic data. Then, we devise a generation penalty term that encourages the generator to synthesize data with a similar diversity to real data. In contrast to previous state-of-the-art generative models that tend to use additional trainable parameters or complex training paradigms, our method does not change the original training scheme. Embedded in an adversarial training and variational autoencoder, our Generative DPP approach shows a consistent resistance to mode-collapse on a wide-variety of synthetic data and natural image datasets including MNIST, CIFAR10, and CelebA, while outperforming state-of-the-art methods for data-efficiency, convergence-time, and generation quality. Our code will be made publicly available.""","""The paper proposes GAN regularized by Determinantal Point Process to learn diverse data samples. The reviewers and AC commonly note the critical limitation of novelty of this paper. The authors pointed out ""To the best of our knowledge, we are the first to introduce modeling data diversity using a Point process kernel that we embed within a generative model. "" AC does not think this is convincing enough to meet the high standard of ICLR. AC decided the paper might not be ready to publish in the current form."""
1088,"""Clean-Label Backdoor Attacks""","['data poisoning', 'backdoor attacks', 'clean labels', 'adversarial examples', 'generative adversarial networks']","""Deep neural networks have been recently demonstrated to be vulnerable to backdoor attacks. Specifically, by altering a small set of training examples, an adversary is able to install a backdoor that can be used during inference to fully control the models behavior. While the attack is very powerful, it crucially relies on the adversary being able to introduce arbitrary, often clearly mislabeled, inputs to the training set and can thus be detected even by fairly rudimentary data filtering. In this paper, we introduce a new approach to executing backdoor attacks, utilizing adversarial examples and GAN-generated data. The key feature is that the resulting poisoned inputs appear to be consistent with their label and thus seem benign even upon human inspection.""","""The present work proposes to improve backdoor poisoning attacks by only using ""clean-label"" images (images whose label would be judged correct by a human), with the motivation that this would make them harder to detect. It considers two approaches to this, one based on GANs and one based on adversarial examples, and shows that the latter works better (and is in general quite effective). It also identifies an interesting phenomenon---that simply using existing back-door attacks with clean labels is substantially less effective than with incorrect labels, because the network does not need to modify itself to accommodate these additional correctly-labeled examples. The strengths of this paper are that it has a detailed empirical evaluation with multiple interesting insights (described above). It also considers efficacy against some basic defense measures based on random pre-processing. A weakness of the paper is that the justification for clean-label attacks is somewhat heuristic, based on the claim that dirty-label attacks can be recognized by hand. There is additional justification that dirty labels tend to be correlated with low confidence, but this correlation (as shown in Figure 2) is actually quite weak. On the other hand, natural defense strategies against the adversarial examples based attack (such as detecting and removing points with large loss at intermediate stages of training) are not considered. This might be fine, as we often assume that the attacker can react to the defender, but it is unclear why we should reject dirty-label attacks on the basis that they can be recognized by one detection mechanism but not give the defender the benefit of other simple detection mechanisms for clean-label attacks. A separate concern was brought up that the attack is too similar to that of Guo et al., and that the method was not run on large-scale datasets. The Guo et al. paper does somewhat diminish the novelty of the present work, but not in a way that I consider problematic; there are definitely new results in this paper, especially the interesting empirical finding that the Guo et al. attack crucially relies on dirty labels. I do not agree with the criticism about large-scale datasets; in general, not all authors have the resources to test on ImageNet, and it is not clear why this should be required unless there is a specific hypothesis that running on ImageNet would test. It is true that the GAN-based method might work more poorly on ImageNet than on CIFAR, but the adversarial attack method (which is in any case the stronger method) seems unlikely to run into scaling issues. Overall, this paper is right on the borderline of acceptance. There are interesting results, and none of the weaknesses are critical. It was unfortunately the case that there wasn't room in the program this year, so the paper was ultimately rejected. However, I think this could be a strong piece of work (and a clear accept) with some additional development. Here are some ideas that might help: (1) Further investigate the phenomenon that adding data points that are too easy to fit do not succeed in data poisoning. This is a fairly interesting point but is not emphasized in the paper. (2) Investigate natural defense mechanisms in the clean-label setting (such as filtering by loss or other such strategies). I do not think it is crucial that the clean-label attack bypasses every simple defense, but considering such defenses can provide more insight into how the attack works--e.g., does it in fact lead to substantially higher loss during training? And if so, at what stage does this occur? If not, how does it succeed in altering the model without inducing high loss?"""
1089,"""Adversarially Robust Training through Structured Gradient Regularization""","['Adversarial Training', 'Gradient Regularization', 'Deep Learning']","""We propose a novel data-dependent structured gradient regularizer to increase the robustness of neural networks vis-a-vis adversarial perturbations. Our regularizer can be derived as a controlled approximation from first principles, leveraging the fundamental link between training with noise and regularization. It adds very little computational overhead during learning and is simple to implement generically in standard deep learning frameworks. Our experiments provide strong evidence that structured gradient regularization can act as an effective first line of defense against attacks based on long-range correlated signal corruptions.""","""Reviewers are in a consensus and recommended to reject after engaging with the authors. Further, many additional questions raised in the discussion should be addressed in the submission to improve clarity. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit. """
1090,"""Computation-Efficient Quantization Method for Deep Neural Networks""","['quantization', 'binary', 'ternary', 'flat minima', 'model compression', 'deep learning']","""Deep Neural Networks, being memory and computation intensive, are a challenge to deploy in smaller devices. Numerous quantization techniques have been proposed to reduce the inference latency/memory consumption. However, these techniques impose a large overhead on the training procedure or need to change the training process. We present a non-intrusive quantization technique based on re-training the full precision model, followed by directly optimizing the corresponding binary model. The quantization training process takes no longer than the original training process. We also propose a new loss function to regularize the weights, resulting in reduced quantization error. Combining both help us achieve full precision accuracy on CIFAR dataset using binary quantization. We also achieve full precision accuracy on WikiText-2 using 2 bit quantization. Comparable results are also shown for ImageNet. We also present a 1.5 bits hybrid model exceeding the performance of TWN LSTM model for WikiText-2.""","""The authors propose a technique for quantizing neural networks, which consist of repeated quantization/de-quantization operations during training, and the second step learns scale factors. The method is simple, clearly presented, and requires no change in the training procedure. However, the authors noted that the work is somewhat incremental, and is similar to previously proposed approaches. As noted by the reviewers, the AC agrees that the work would be significantly strengthened by additional analysis of complexity in terms of computational time and memory relative to the other techniques. """
1091,"""Probabilistic Federated Neural Matching""","['Bayesian nonparametrics', 'Indian Buffet Process', 'Federated Learning']","""In federated learning problems, data is scattered across different servers and exchanging or pooling it is often impractical or prohibited. We develop a Bayesian nonparametric framework for federated learning with neural networks. Each data server is assumed to train local neural network weights, which are modeled through our framework. We then develop an inference approach that allows us to synthesize a more expressive global network without additional supervision or data pooling. We then demonstrate the efficacy of our approach on federated learning problems simulated from two popular image classification datasets.""","""While there was some support for the ideas presented, unfortunately this paper was on the borderline. Significant concerns were raised as to whether the setting studied was realistic, among others."""
1092,"""Multi-way Encoding for Robustness to Adversarial Attacks""","['Adversarial Defense', 'Robustness of Deep Convolutional Networks']","""Deep models are state-of-the-art for many computer vision tasks including image classification and object detection. However, it has been shown that deep models are vulnerable to adversarial examples. We highlight how one-hot encoding directly contributes to this vulnerability and propose breaking away from this widely-used, but highly-vulnerable mapping. We demonstrate that by leveraging a different output encoding, multi-way encoding, we can make models more robust. Our approach makes it more difficult for adversaries to find useful gradients for generating adversarial attacks. We present state-of-the-art robustness results for black-box, white-box attacks, and achieve higher clean accuracy on four benchmark datasets: MNIST, CIFAR-10, CIFAR-100, and SVHN when combined with adversarial training. The strength of our approach is also presented in the form of an attack for model watermarking, raising challenges in detecting stolen models.""","""This paper proposes a method for improving robustness to black-box adversarial attacks by replacing the cross-entropy layer with an output vector encoding scheme. The paper is well-written, and the approach appears to be novel. However, Reviewer 4 raises very relevant concerns regarding the experimental evaluation of the method, including (a) lack of robustness without AT in the whitebox case (which is very relevant as we still lack good understanding of blackbox vs whitebox robustness) (b) comparison with Kannan et al and (c) lack of some common strong attacks. Reviewer 1 echoes many of these concerns."""
1093,"""Cost-Sensitive Robustness against Adversarial Examples""","['Certified robustness', 'Adversarial examples', 'Cost-sensitive learning']","""Several recent works have developed methods for training classifiers that are certifiably robust against norm-bounded adversarial perturbations. These methods assume that all the adversarial transformations are equally important, which is seldom the case in real-world applications. We advocate for cost-sensitive robustness as the criteria for measuring the classifier's performance for tasks where some adversarial transformation are more important than others. We encode the potential harm of each adversarial transformation in a cost matrix, and propose a general objective function to adapt the robust training method of Wong & Kolter (2018) to optimize for cost-sensitive robustness. Our experiments on simple MNIST and CIFAR10 models with a variety of cost matrices show that the proposed approach can produce models with substantially reduced cost-sensitive robust error, while maintaining classification accuracy.""","""This paper studies the notion of certified cost-sensitive robustness against adversarial examples, by building from the recent [Wong & Koller'18]. Its main contribution is to adapt the robust classification objective to a 'cost-sensitive' objective, that weights labelling errors according to their potential damage. This paper received mixed reviews, with a clear champion and two skeptical reviewers. On the one hand, they all highlighted the clarity of the presentation and the relevance of the topic as strengths; on the other hand, they noted the relatively little novelty of the paper relative [W & K'18]. Reviewers also acknowledged the diligence of authors during the response phase. The AC mostly agrees with these assessments, and taking them all into consideration, he/she concludes that the potential practical benefits of cost-sensitive certified robustness outweight the limited scientific novelty. Therefore, he recommends acceptance as a poster. """
1094,"""ATTENTION INCORPORATE NETWORK: A NETWORK CAN ADAPT VARIOUS DATA SIZE""","['attention mechanism', 'various image size']","""In traditional neural networks for image processing, the inputs of the neural networks should be the same size such as 2242243. But how can we train the neural net model with different input size? A common way to do is image deformation which accompany a problem of information loss (e.g. image crop or wrap). In this paper we propose a new network structure called Attention Incorporate Network(AIN). It solve the problem of different size of input images and extract the key features of the inputs by attention mechanism, pay different attention depends on the importance of the features not rely on the data size. Experimentally, AIN achieve a higher accuracy, better convergence comparing to the same size of other network structure.""","""All reviewers agree that the paper should be rejected and there is no rebuttal."""
1095,"""Learning to remember: Dynamic Generative Memory for Continual Learning""","['Continual Learning', 'Catastrophic Forgetting', 'Dynamic Network Expansion']","""Continuously trainable models should be able to learn from a stream of data over an undefined period of time. This becomes even more difficult in a strictly incremental context, where data access to previously seen categories is not possible. To that end, we propose making use of a conditional generative adversarial model where the generator is used as a memory module through neural masking to emulate neural plasticity in the human brain. This memory module is further associated with a dynamic capacity expansion mechanism. Taken together, this method facilitates a resource efficient capacity adaption to accommodate new tasks, while retaining previously attained knowledge. The proposed approach outperforms state-of-the-art algorithms on publicly available datasets, overcoming catastrophic forgetting.""","""The authors propose to tackle the problem of catastrophic forgetting in continual learning by adopting the generative replay strategy with the generator network as an extendable memory module. While acknowledging that the proposed model is potentially useful, the reviewers raised several important concerns that were viewed by AC as critical issues: (1) poor presentation clarity of the manuscript and incremental technical contribution in light of prior work by Serra et al. (2018); (2) rigorous experiments and in-depth analysis of the baseline models in terms of accuracy, number of parameters, memory demand and model complexity would significantly strengthen the evaluation see R1s and R3s suggestions how to improve; (3) simple strategies such as storing a number of examples and memory replay should not be neglected and evaluated to assess the scope of the contribution. Additionally R1 raised a concern that preventing the generator from forgetting should be supported by an ablation study on both, the discriminator and the generator, abilities to remember and to forget. R1 and R3 provided very detailed and constructive reviews, as acknowledged by the authors. R2 expressed similar concerns about time/memory comparison of different methods, but his/her brief review did not have a substantial impact on the decision. AC suggests in its current state the manuscript is not ready for a publication. We hope the reviews are useful for improving and revising the paper. """
1096,"""Riemannian Stochastic Gradient Descent for Tensor-Train Recurrent Neural Networks""","['Riemannian Stochastic Gradient Descent', 'Tensor-Train', 'Recurrent Neural Networks']","""The Tensor-Train factorization (TTF) is an efficient way to compress large weight matrices of fully-connected layers and recurrent layers in recurrent neural networks (RNNs). However, high Tensor-Train ranks for all the core tensors of parameters need to be element-wise fixed, which results in an unnecessary redundancy of model parameters. This work applies Riemannian stochastic gradient descent (RSGD) to train core tensors of parameters in the Riemannian Manifold before finding vectors of lower Tensor-Train ranks for parameters. The paper first presents the RSGD algorithm with a convergence analysis and then tests it on more advanced Tensor-Train RNNs such as bi-directional GRU/LSTM and Encoder-Decoder RNNs with a Tensor-Train attention model. The experiments on digit recognition and machine translation tasks suggest the effectiveness of the RSGD algorithm for Tensor-Train RNNs. ""","""This paper proposes using a tensor train low rank decomposition for compressing neural network parameters. However the paper falls short on multiple fronts 1)lack of comparison with existing methods 2) no baseline experiments. Further there are concerns about correctness of the math in deriving the algorithms, convergence and computational complexity of the proposed method. I strongly suggest taking the reviews into account before submitting the paper it again. """
1097,"""Unsupervised Learning of the Set of Local Maxima""","['Unsupervised Learning', 'One-class Classification', 'Multi-player Optimization']","""This paper describes a new form of unsupervised learning, whose input is a set of unlabeled points that are assumed to be local maxima of an unknown value function pseudo-formula in an unknown subset of the vector space. Two functions are learned: (i) a set indicator pseudo-formula , which is a binary classifier, and (ii) a comparator function pseudo-formula that given two nearby samples, predicts which sample has the higher value of the unknown function pseudo-formula . Loss terms are used to ensure that all training samples pseudo-formula are a local maxima of pseudo-formula , according to pseudo-formula and satisfy pseudo-formula . Therefore, pseudo-formula and pseudo-formula provide training signals to each other: a point pseudo-formula in the vicinity of pseudo-formula satisfies pseudo-formula or is deemed by pseudo-formula to be lower in value than pseudo-formula . We present an algorithm, show an example where it is more efficient to use local maxima as an indicator function than to employ conventional classification, and derive a suitable generalization bound. Our experiments show that the method is able to outperform one-class classification algorithms in the task of anomaly detection and also provide an additional signal that is extracted in a completely unsupervised way. ""","""The paper proposes a new unsupervised learning scheme via utilizing local maxima as an indicator function. The reviewers and AC note the novelty of this paper and good empirical justifications. Hence, AC decided to recommend acceptance. However, AC thinks the readability of the paper can be improved."""
1098,"""Understanding Composition of Word Embeddings via Tensor Decomposition""","['word embeddings', 'semantic composition', 'tensor decomposition']","""Word embedding is a powerful tool in natural language processing. In this paper we consider the problem of word embedding composition \--- given vector representations of two words, compute a vector for the entire phrase. We give a generative model that can capture specific syntactic relations between words. Under our model, we prove that the correlations between three words (measured by their PMI) form a tensor that has an approximate low rank Tucker decomposition. The result of the Tucker decomposition gives the word embeddings as well as a core tensor, which can be used to produce better compositions of the word embeddings. We also complement our theoretical results with experiments that verify our assumptions, and demonstrate the effectiveness of the new composition method.""","""AR1 is concerned about lack of downstream applications which show that higher-order interactions are useful and asks why not to model higher-order interactions for all (a,b) pairs. AR2 notes that this submission is a further development of Arora et al. and is satisfied with the paper. AR3 is the most critical regarding lack of explanations, e.g. why linear addition of two word embeddings is bad and why the corrective term proposed here is a good idea. The authors suggest that linear addition is insufficient when final meaning differs from the individual meanings and show tome quantitative results to back up their corrective term. On balance, all reviewers find the theoretical contributions sufficient which warrants an accept. The authors are asked to honestly reflect all uncertain aspects of their work in the final draft to reflect legitimate concerns of reviewers."""
1099,"""Contextualized Role Interaction for Neural Machine Translation""","['Neural Machine Translation', 'Natural Language Processing']","""Word inputs tend to be represented as single continuous vectors in deep neural networks. It is left to the subsequent layers of the network to extract relevant aspects of a word's meaning based on the context in which it appears. In this paper, we investigate whether word representations can be improved by explicitly incorporating the idea of latent roles. That is, we propose a role interaction layer (RIL) that consists of context-dependent (latent) role assignments and role-specific transformations. We evaluate the RIL on machine translation using two language pairs (En-De and En-Fi) and three datasets of varying size. We find that the proposed mechanism improves translation quality over strong baselines with limited amounts of data, but that the improvement diminishes as the size of data grows, indicating that powerful neural MT systems are capable of implicitly modeling role-word interaction by themselves. Our qualitative analysis reveals that the RIL extracts meaningful context-dependent roles and that it allows us to inspect more deeply the internal mechanisms of state-of-the-art neural machine translation systems.""","""This paper proposes to improve MT with a specialized encoder component that models roles. It shows some improvements in low-resource scenarios. Overall, reviewers felt there were two issues with the paper: clarity of description of the contribution, and also the fact that the method itself was not seeing large empirical gains. On top of this, the method adds some additional complexity on top of the original model. Given that no reviewer was strongly in favor of the paper, I am not going to recommend acceptance at this time."""
1100,"""Unsupervised Neural Multi-Document Abstractive Summarization of Reviews""","['unsupervised learning', 'abstractive summarization', 'reviews', 'text generation']","""Abstractive summarization has been studied using neural sequence transduction methods with datasets of large, paired document-summary examples. However, such datasets are rare and the models trained from them do not generalize to other domains. Recently, some progress has been made in learning sequence-to-sequence mappings with only unpaired examples. In our work, we consider the setting where there are only documents (product or business reviews) with no summaries provided, and propose an end-to-end, neural model architecture to perform unsupervised abstractive summarization. Our proposed model consists of an auto-encoder trained so that the mean of the representations of the input reviews decodes to a reasonable summary-review. We consider variants of the proposed architecture and perform an ablation study to show the importance of specific components. We show through metrics and human evaluation that the generated summaries are highly abstractive, fluent, relevant, and representative of the average sentiment of the input reviews.""","""This paper introduces a method for unsupervised abstractive summarization of reviews. Strengths: (1) The direction (developing unsupervised multi-document summarization systems) is exciting (2) There are interesting aspects to the model Weaknesses: (1) The authors are clearly undecided how to position this work: either as introducing a generic document summarization framework or as an approach specific to summarization of reviews. If this is the former, the underlying assumptions, e.g., that the summary looks like a single document in a group is problematic. If this is the latter, then comparison to some more specialized methods are lacking (see comments of R1). (2) Evaluation, though improved since the first submitted version (when human evaluation was added), is still not great (see R1 / R3). The automatic metrics are not very convincing and do not seem to be very consistent with the results of human eval. I believe that instead or along with human eval, the authors should create human written summaries and evaluate against them. It has been done for extractive multi-document summarization and can be done here. Without this, it would be impossible to compare to this submission in the future work. (3) It is not very clear that generating abstractive summaries of the form proposed in the paper is an effective way to summarize documents. Basically, a good summary should reflect diversity of the opinions rather than reflect an average / most frequent opinion from tin the review collection. By generating the summary from a review LM, the authors make sure that there is no redundancy (e.g., alternative views) or contradictions. That's not really what one would want from a summary (See R3 and also non-public discussion with R1) Overall, I'd definitely like to see this work published but my take is that it is not ready yet. R1 and R2 are relatively negative and generally in agreement. R3 is very positive. I share excitement about the research direction with R3 but I believe that concerns of R1 and R2 are valid and need to be addressed before the paper gets published. """
1101,"""The GAN Landscape: Losses, Architectures, Regularization, and Normalization""","['GANs', 'empirical evaluation', 'large-scale', 'reproducibility']","""Generative adversarial networks (GANs) are a class of deep generative models which aim to learn a target distribution in an unsupervised fashion. While they were successfully applied to many problems, training a GAN is a notoriously challenging task and requires a significant amount of hyperparameter tuning, neural architecture engineering, and a non-trivial amount of ``tricks"". The success in many practical applications coupled with the lack of a measure to quantify the failure modes of GANs resulted in a plethora of proposed losses, regularization and normalization schemes, and neural architectures. In this work we take a sober view of the current state of GANs from a practical perspective. We reproduce the current state of the art and go beyond fairly exploring the GAN landscape. We discuss common pitfalls and reproducibility issues, open-source our code on Github, and provide pre-trained models on TensorFlow Hub.""","""The paper presents a large scale empirical comparison between different prominent losses, regularization and normalization schemes, and neural architectures frequently used in GAN training. Large scale comparisons in this field are rare and important and the outcome of the experimental analysis is clearly of interest for practitioners. However, as two of the reviewers point out, the significance of the new insights is limited, and after rebutal all reviewers agree that the paper would profit from a clearer write-up and presentation of the main findings. I see the paper therefore, as lying slightly under the acceptance trashhold."""
1102,"""Generative Adversarial Models for Learning Private and Fair Representations""","['Data Privacy', 'Fairness', 'Adversarial Learning', 'Generative Adversarial Networks', 'Minimax Games', 'Information Theory']","""We present Generative Adversarial Privacy and Fairness (GAPF), a data-driven framework for learning private and fair representations of the data. GAPF leverages recent advances in adversarial learning to allow a data holder to learn ""universal"" representations that decouple a set of sensitive attributes from the rest of the dataset. Under GAPF, finding the optimal decorrelation scheme is formulated as a constrained minimax game between a generative decorrelator and an adversary. We show that for appropriately chosen adversarial loss functions, GAPF provides privacy guarantees against strong information-theoretic adversaries and enforces demographic parity. We also evaluate the performance of GAPF on multi-dimensional Gaussian mixture models and real datasets, and show how a designer can certify that representations learned under an adversary with a fixed architecture perform well against more complex adversaries. ""","""While there was some support for the ideas presented, the majority of the reviewers did not think the submission was ready for presentation at ICLR. Concerns raised included that the experiments needed more work, and the paper needs to do a better job of distinguishing the contributions beyond those of past work."""
1103,"""Success at any cost: value constrained model-free continuous control""","['reinforcement learning', 'continuous control', 'robotics', 'constrained optimization', 'multi-objective optimization']","""Naively applying Reinforcement Learning algorithms to continuous control problems -- such as locomotion and robot control -- to maximize task reward often results in policies which rely on high-amplitude, high-frequency control signals, known colloquially as bang-bang control. While such policies can implement the optimal solution, particularly in simulated systems, they are often not desirable for real world systems since bang-bang control can lead to increased wear and tear and energy consumption and tends to excite undesired second-order dynamics. To counteract this issue, multi-objective optimization can be used to simultaneously optimize both the reward and some auxiliary cost that discourages undesired (e.g. high-amplitude) control. In principle, such an approach can yield the sought after, smooth, control policies. It can, however, be hard to find the correct trade-off between cost and return that results in the desired behavior. In this paper we propose a new constraint-based approach which defines a lower bound on the return while minimizing one or more costs (such as control effort). We employ Lagrangian relaxation to learn both (a) the parameters of a control policy that satisfies the desired constraints and (b) the Lagrangian multipliers for the optimization. Moreover, we demonstrate policy optimization which satisfies constraints either in expectation or in a per-step fashion, and we learn a single conditional policy that is able to dynamically change the trade-off between return and cost. We demonstrate the efficiency of our approach using a number of continuous control benchmark tasks as well as a realistic, energy-optimized quadruped locomotion task.""","""Strengths: The paper introduces a novel constrained-optimization method for RL problems. A lower-bound constraint can be imposed on the return (cumulative reward), while optimizing one or more other costs, such as control effort. The method learns multiple The paper is clearly written. Results are shown on the cart-and-pole, a humanoid, and a realistic Minitaur quadruped model. AC: Being able to learn conditional constraints is an interesting direction. Weaknesses: There are often simpler ways to solve the problem of high-amplitude, high-frequency controls in the setting of robotics. The paper removes one hyperparameter (lambda) but then introduces another (beta), although beta is likely easier to tune. The ideas have some strong connections to existing work in safe reinforcement learning. AC: Video results for the humanoid and cart-and-pole examples would have been useful to see. Summary: The paper makes progress on ideas that are fairly involved to explore and use (perhaps limiting their use in the short term), but that have potential, i.e., learning state-dependent Lagrange multipliers for constrained RL. The paper is perfectly fine technically, and does break some new ground in putting a particular set of pieces together. As articulated by two of the reviewers, from a pragmatic perspective, the results are not yet entirely compelling. I do believe that a better understanding of working with constrained RL, in ways that are somewhat different than those used in Safe RL work. Given the remaining muted enthusiasm of two of the reviewers, and in the absence of further calibration, the AC leans marginally towards a reject. Current scores: 5,6,7. Again, the paper does have novelty, although it's a pretty intricate setup. The AC would be happy to revisit upon global recalibration. """
1104,"""A Model Cortical Network for Spatiotemporal Sequence Learning and Prediction""","['cortical models', 'spatiotemporal memory', 'video prediction', 'predictive coding']","""In this paper we developed a hierarchical network model, called Hierarchical Prediction Network (HPNet) to understand how spatiotemporal memories might be learned and encoded in a representational hierarchy for predicting future video frames. The model is inspired by the feedforward, feedback and lateral recurrent circuits in the mammalian hierarchical visual system. It assumes that spatiotemporal memories are encoded in the recurrent connections within each level and between different levels of the hierarchy. The model contains a feed-forward path that computes and encodes spatiotemporal features of successive complexity and a feedback path that projects interpretation from a higher level to the level below. Within each level, the feed-forward path and the feedback path intersect in a recurrent gated circuit that integrates their signals as well as the circuit's internal memory states to generate a prediction of the incoming signals. The network learns by comparing the incoming signals with its prediction, updating its internal model of the world by minimizing the prediction errors at each level of the hierarchy in the style of {\em predictive self-supervised learning}. The network processes data in blocks of video frames rather than a frame-to-frame basis. This allows it to learn relationships among movement patterns, yielding state-of-the-art performance in long range video sequence predictions in benchmark datasets. We observed that hierarchical interaction in the network introduces sensitivity to memories of global movement patterns even in the population representation of the units in the earliest level. Finally, we provided neurophysiological evidence, showing that neurons in the early visual cortex of awake monkeys exhibit very similar sensitivity and behaviors. These findings suggest that predictive self-supervised learning might be an important principle for representational learning in the visual cortex. ""","""There was major disagreement between reviewers on this paper. Two reviewers recommend acceptance, and one firm rejection. The initial version of the manuscript was of poor quality in terms of exposition, as noted by all reviewers. However, the authors responded carefully and thoroughly to reviewer comments, and major clarity and technical issues were resolved by all authors. I ask PCs to note that the paper, as originally submitted, was not fit for acceptance, and reviewers noted major changes during the review process. I do believe this behavior should be discouraged, since it effectively requires reviewers to examine the paper twice. Regardless, the final overall score of the paper does not meet the bar for acceptance into ICLR."""
1105,"""Neural Random Projections for Language Modelling""","['neural networks', 'language modelling', 'natural language processing', 'uncertainty', 'random projections']","""Neural network-based language models deal with data sparsity problems by mapping the large discrete space of words into a smaller continuous space of real-valued vectors. By learning distributed vector representations for words, each training sample informs the neural network model about a combinatorial number of other patterns. In this paper, we exploit the sparsity in natural language even further by encoding each unique input word using a fixed sparse random representation. These sparse codes are then projected onto a smaller embedding space which allows for the encoding of word occurrences from a possibly unknown vocabulary, along with the creation of more compact language models using a reduced number of parameters. We investigate the properties of our encoding mechanism empirically, by evaluating its performance on the widely used Penn Treebank corpus. We show that guaranteeing approximately equidistant vector representations for unique discrete inputs is enough to provide the neural network model with enough information to learn --and make use-- of distributed representations for these inputs.""","""There is a clear reviewer consensus to reject this paper so I am also recommending rejecting it. The paper is about an interesting and underused technique. However, ultimately the issue here is that the paper does not do a good enough job of explaining the contribution. I hope the reviews have given the authors some ideas on how to frame and sell this work better in the future. For instance, from my own reading of the abstract, I do not understand what this paper is trying to do and why it is valuable. Phrases such as ""we exploit the sparsity"" do not tell me why the paper is important to read or what it accomplishes, only how it accomplishes the seemingly elided contribution. I am forced to make assumptions that might not be correct about the goals and motivation. It is certainly true that the implicit one-hot representation of words most common in neural language models is not the only possibility and that random sparse vectors for words will also work reasonably well. I have even tried techniques like this myself, personally, in language modeling experiments and I believe others have as well, although I do not have a nice reference close to hand (some of the various Mikolov models use random hashing of n-grams and I believe related ideas are common in the maxent LM literature and elsewhere). So when the abstract says things like ""We show that guaranteeing approximately equidistant vector representations for unique discrete inputs is enough to provide the neural network model with enough information to learn"" my immediate reaction is to ask why this would be surprising or why it would matter. Based on the reviews, I believe these sorts of issues affect other parts of the manuscript as well. There needs to be a sharper argument that either presents a problem and its solution or presents a scientific question and its answer. In the first case, the problem should be well motivated and in the second case the question should not yet have been adequately answered by previous work and should be non-obvious. I should not have to read beyond the abstract to understand the accomplishments of this work. Moving to the conclusion and future work section, I can see the appeal of the future work in the second paragraph, but this work has not been done. The first paragraph is about how it is possible to use random projections to represent words, which is not something I think most researchers would question. Missing is a clear demonstration of the potential advantages of doing so. """
1106,"""Model Compression with Generative Adversarial Networks""","['Model compression', 'distillation', 'generative adversarial network', 'GAN', 'deep neural network', 'random forest', 'ensemble', 'decision tree', 'convolutional neural network']","""More accurate machine learning models often demand more computation and memory at test time, making them difficult to deploy on CPU- or memory-constrained devices. Model compression (also known as distillation) alleviates this burden by training a less expensive student model to mimic the expensive teacher model while maintaining most of the original accuracy. However, when fresh data is unavailable for the compression task, the teacher's training data is typically reused, leading to suboptimal compression. In this work, we propose to augment the compression dataset with synthetic data from a generative adversarial network (GAN) designed to approximate the training data distribution. Our GAN-assisted model compression (GAN-MC) significantly improves student accuracy for expensive models such as deep neural networks and large random forests on both image and tabular datasets. Building on these results, we propose a comprehensive metricthe Compression Scoreto evaluate the quality of synthetic datasets based on their induced model compression performance. The Compression Score captures both data diversity and discriminability, and we illustrate its benefits over the popular Inception Score in the context of image classification.""","""The authors propose a scheme to compress models using student-teacher distillation, where training data are augmented using examples generated from a conditional GAN. The reviewers were generally in agreement that 1) that the experimental results generally support the claims made by the authors, and 2) that the paper is clearly written and easy to follow. However, the reviewers also raised a number of concerns: 1) that the experiments were conducted on small-scale tasks, 2) the use of the compression score might be impractical since it would require retraining a compressed model, and is affected by the effectiveness of the compression algorithm which is an additional confounding factor. The authors in their rebuttal address 2) by noting that the student training was not too expensive, but I believe that this cost is task specific. Overall, I think 1) is a significant concern, and the AC agrees with the reviewers that an evaluation of the techniques on large-scale datasets would strengthen the paper. """
1107,"""Empirical Bounds on Linear Regions of Deep Rectifier Networks""","['linear regions', 'approximate model counting', 'mixed-integer linear programming']","""One form of characterizing the expressiveness of a piecewise linear neural network is by the number of linear regions, or pieces, of the function modeled. We have observed substantial progress in this topic through lower and upper bounds on the maximum number of linear regions and a counting procedure. However, these bounds only account for the dimensions of the network and the exact counting may take a prohibitive amount of time, therefore making it infeasible to benchmark the expressiveness of networks. In this work, we approximate the number of linear regions of specific rectifier networks with an algorithm for probabilistic lower bounds of mixed-integer linear sets. In addition, we present a tighter upper bound that leverages network coefficients. We test both on trained networks. The algorithm for probabilistic lower bounds is several orders of magnitude faster than exact counting and the values reach similar orders of magnitude, hence making our approach a viable method to compare the expressiveness of such networks. The refined upper bound is particularly stronger on networks with narrow layers. ""","""The paper seeks to obtain faster means to count or approximately count of the number of linear regions of a neural network. The paper improves bounds and makes an interesting contribution to a long line of work. A consistent concern of the reviewers is the limited applicability of the method. The empirical evaluation can serve to better assess the accuracy of theoretical bounds that have been obtained in previous works, but the practical utility is not as clear yet. This is a borderline case. The reviewers lean towards a positive rating of the paper, but are not particularly enthusiastic about the paper. The paper makes good contributions, but is just not convincing enough. I think that the work program that the authors suggest in their responses could lead to a stronger paper in the future. In particular, the exploration of necessary and sufficient conditions for different neural networks to be equivalent and the use of number of linear regions when analyzing neural networks, seem to be very promising directions. """
1108,"""Mean Replacement Pruning ""","['pruning', 'saliency', 'neural networks', 'optimization', 'redundancy', 'model compression']","""Pruning units in a deep network can help speed up inference and training as well as reduce the size of the model. We show that bias propagation is a pruning technique which consistently outperforms the common approach of merely removing units, regardless of the architecture and the dataset. We also show how a simple adaptation to an existing scoring function allows us to select the best units to prune. Finally, we show that the units selected by the best performing scoring functions are somewhat consistent over the course of training, implying the dead parts of the network appear during the stages of training.""","""This paper proposes an approach to pruning units in a deep neural network while training is in progress. The idea is to (1) use a specific ""scoring function"" (the absolute-valued Taylor expansion of the loss) to identify the best units to prune, (2) computing the mean activations of the units to be pruned on a small sample of training data, (3) adding the mean activations multiplied by the outgoing weights into the biases of the next layer's units, and (4) removing the pruned units from the network. Extensive experiments show that this approach to pruning does less immediate damage than the more common zero-replacement approach, that this advantage remains (but is much smaller) after fine-tuning, and that the importance of units tends not to change much during training. The reviewers liked the quality of the writing and the extensive experimentation, but even after discussion and revision had concerns about the limited novelty of the approach, the fact that the proposed approach is incompatible with batch normalization (which severely limits the range of architectures to which the method may be applied), and were concerned that the proposed method has limited impact after fine-tuning."""
1109,"""Relaxed Quantization for Discretized Neural Networks""","['Quantization', 'Compression', 'Neural Networks', 'Efficiency']","""Neural network quantization has become an important research area due to its great impact on deployment of large models on resource constrained devices. In order to train networks that can be effectively discretized without loss of performance, we introduce a differentiable quantization procedure. Differentiability can be achieved by transforming continuous distributions over the weights and activations of the network to categorical distributions over the quantization grid. These are subsequently relaxed to continuous surrogates that can allow for efficient gradient-based optimization. We further show that stochastic rounding can be seen as a special case of the proposed approach and that under this formulation the quantization grid itself can also be optimized with gradient descent. We experimentally validate the performance of our method on MNIST, CIFAR 10 and Imagenet classification.""","""This paper proposes an effective method to train neural networks with quantized reduced precision. It's fairly straight-forward idea and achieved good results and solid empirical work. reviewers have a consensus on acceptance. """
1110,"""Learning Self-Imitating Diverse Policies""","['Reinforcement-learning', 'Imitation-learning', 'Ensemble-training']","""The success of popular algorithms for deep reinforcement learning, such as policy-gradients and Q-learning, relies heavily on the availability of an informative reward signal at each timestep of the sequential decision-making process. When rewards are only sparsely available during an episode, or a rewarding feedback is provided only after episode termination, these algorithms perform sub-optimally due to the difficultly in credit assignment. Alternatively, trajectory-based policy optimization methods, such as cross-entropy method and evolution strategies, do not require per-timestep rewards, but have been found to suffer from high sample complexity by completing forgoing the temporal nature of the problem. Improving the efficiency of RL algorithms in real-world problems with sparse or episodic rewards is therefore a pressing need. In this work, we introduce a self-imitation learning algorithm that exploits and explores well in the sparse and episodic reward settings. We view each policy as a state-action visitation distribution and formulate policy optimization as a divergence minimization problem. We show that with Jensen-Shannon divergence, this divergence minimization problem can be reduced into a policy-gradient algorithm with shaped rewards learned from experience replays. Experimental results indicate that our algorithm works comparable to existing algorithms in environments with dense rewards, and significantly better in environments with sparse and episodic rewards. We then discuss limitations of self-imitation learning, and propose to solve them by using Stein variational policy gradient descent with the Jensen-Shannon kernel to learn multiple diverse policies. We demonstrate its effectiveness on a challenging variant of continuous-control MuJoCo locomotion tasks.""","""This paper proposes a reinforcement learning approach that better handles sparse reward environments, by using previously-experienced roll-outs that achieve high reward. The approach is intuitive, and the results in the paper are convincing. The authors addressed nearly all of the reviewer's concerns. The reviewers all agree that the paper should be accepted."""
1111,"""Selectivity metrics can overestimate the selectivity of units: a case study on AlexNet""","['AlexNet', 'neural networks', 'selectivity', 'localist', 'distributed', 'represenataion', 'precision', 'measures of selectivity', 'object detectors', 'single directions', 'network analysis']","""Various methods of measuring unit selectivity have been developed in order to understand the representations learned by neural networks (NNs). Here we undertake a comparison of four such measures on AlexNet, namely, localist selectivity, \precision (Zhou et al, ICLR 2015), class-conditional mean activity selectivity CCMAS; (Morcos et al, ICLR 2018), and a new measure called top-class selectivity. In contrast with previous work on recurrent neural networks (RNNs), we fail to find any 100\% selective `localist units' in AlexNet, and demonstrate that the \precision and CCMAS measures provide a much higher level of selectivity than is warranted, with the most selective hidden units only responding strongly to a small minority of images from within a category. We also generated activation maximization (AM) images that maximally activated individual units and found that under (5\%) of units in fc6 and conv5 produced interpretable images of objects, whereas fc8 produced over 50\% interpretable images. Furthermore, the interpretable images in the hidden layers were not associated with highly selective units. These findings highlight the problem with current selectivity measures and show that new measures are required in order to provide a better assessment of learned representations in NNs. We also consider why localist representations are learned in RNNs and not AlexNet.""","""The paper examined the folk-knowledge that there are highly selective units in popular CNN architectures, and performed a detailed analysis of recent measures of unit selectivity, as well as introducing a novel one. The finding that units are not extremely selective in CNNs was intriguing to some (not all) reviewers. Further, they show recent measures of selectivity dramatically over-estimate selectivity. There was not tight agreement amongst the reviewers on the paper's rating, but it trended towards rejection. Weaknesses highlighted by reviewers include lack of visual clarity in their demonstrations, the use of a several-generations-old CNN architecture, as well as a lack of enthusiasm for the findings."""
1112,"""Efficient Exploration through Bayesian Deep Q-Networks""","['Deep RL', 'Exploration Exploitation', 'DQN', 'Bayesian Regret', 'Thompson Sampling']","""We propose Bayesian Deep Q-Networks (BDQN), a principled and a practical Deep Reinforcement Learning (DRL) algorithm for Markov decision processes (MDP). It combines Thompson sampling with deep-Q networks (DQN). Thompson sampling ensures more efficient exploration-exploitation tradeoff in high dimensions. It is typically carried out through posterior sampling over the model parameters, which makes it computationally expensive. To overcome this limitation, we directly incorporate uncertainty over the value (Q) function. Further, we only introduce randomness in the last layer (i.e. the output layer) of the DQN and use independent Gaussian priors on the weights. This allows us to efficiently carry out Thompson sampling through Gaussian sampling and Bayesian Linear Regression (BLR), which has fast closed-form updates. The rest of the layers of the Q network are trained through back propagation, as in a standard DQN. We apply our method to a wide range of Atari games in Arcade Learning Environments and compare BDQN to a powerful baseline: the double deep Q-network (DDQN). Since BDQN carries out more efficient exploration, it is able to reach higher rewards substantially faster: in less than 5M1M samples for almost half of the games to reach DDQN scores while a typical run of DDQN is 50-200M. We also establish theoretical guarantees for the special case when the feature representation is fixed and not learnt. We show that the Bayesian regret is bounded by O(d \sqrt(N)) after N time steps for a d-dimensional feature map, and this bound is shown to be tight up-to logarithmic factors. To the best of our knowledge, this is the first Bayesian theoretical guarantee for Markov Decision Processes (MDP) beyond the tabula rasa setting.""","""There was a significant amount of discussion on this paper, both from the reviewers and from unsolicited feedback. This is a good sign as it demonstrates interest in the work. Improving exploration in Deep Q-learning through Thompson sampling using uncertainty from the model seems sensible and the empirical results on Atari seem quite impressive. However, the reviewers and others argued that there were technical flaws in the work, particularly in the proofs. Also, reviewers noted that clarity of the paper was a significant issue, even more so than a previous submission. One reviewer noted that the authors had significantly improved the paper throughout the discussion phase. However, ultimately all reviewers agreed that the paper was not quite ready for acceptance. It seems that the paper could still use some significant editing and careful exposition and justification of the technical content. Note, one of the reviews was disregarded due to incorrectness and a fourth reviewer was brought in."""
1113,"""Capsule Graph Neural Network""","['CapsNet', 'Graph embedding', 'GNN']","""The high-quality node embeddings learned from the Graph Neural Networks (GNNs) have been applied to a wide range of node-based applications and some of them have achieved state-of-the-art (SOTA) performance. However, when applying node embeddings learned from GNNs to generate graph embeddings, the scalar node representation may not suffice to preserve the node/graph properties efficiently, resulting in sub-optimal graph embeddings. Inspired by the Capsule Neural Network (CapsNet), we propose the Capsule Graph Neural Network (CapsGNN), which adopts the concept of capsules to address the weakness in existing GNN-based graph embeddings algorithms. By extracting node features in the form of capsules, routing mechanism can be utilized to capture important information at the graph level. As a result, our model generates multiple embeddings for each graph to capture graph properties from different aspects. The attention module incorporated in CapsGNN is used to tackle graphs with various sizes which also enables the model to focus on critical parts of the graphs. Our extensive evaluations with 10 graph-structured datasets demonstrate that CapsGNN has a powerful mechanism that operates to capture macroscopic properties of the whole graph by data-driven. It outperforms other SOTA techniques on several graph classification tasks, by virtue of the new instrument.""","""AR1 asks for a clear experimental evaluation showing that capsules and dynamic routing help in the GCN setting. After rebuttal, AR1 seems satisfied that routing in CapsGNN might help generate 'more representative graph embeddings from different aspects'. AC strongly encourages the authors to improve the discussion on these 'different aspects' as currently it feels vague. AR2 is initially concerned about experimental evaluations and whether the attention mechanism works as expected, though, he/she is happy with the revised experiments. AR3 would like to see all biological datasets included in experiments. He/she is also concerned about the lack of ability to preserve fine structures by CapsGNN. The authors leave this aspect of their approach for the future work. On balance, all reviewers felt this paper is a borderline paper. After going through all questions and responses, AC sees that many requests about aspects of the proposed method have not been clarified by the authors. However, reviewers note that the authors provided more evaluations/visualisations etc. The reviewers expressed hope (numerous times) that this initial attempt to introduce capsules into GCN will result in future developments and improvements. While AC thinks this is an overoptimistic view, AC will give the authors the benefit of doubt and will advocate a weak accept. The authors are asked to incorporate all modifications requested by the reviewers. Moreover, 'Graph capsule convolutional neural networks' is not a mere ArXiV work. It is an ICML workshop paper. Kindly check all ArXiV references and update with the actual conference venues."""
1114,"""3D-RelNet: Joint Object and Relational Network for 3D Prediction""","['3D Reconstruction', '3D Scene Understanding', 'Relative Prediction']","""We propose an approach to predict the 3D shape and pose for the objects present in a scene. Existing learning based methods that pursue this goal make independent predictions per object, and do not leverage the relationships amongst them. We argue that reasoning about these relationships is crucial, and present an approach to incorporate these in a 3D prediction framework. In addition to independent per-object predictions, we predict pairwise relations in the form of relative 3D pose, and demonstrate that these can be easily incorporated to improve object level estimates. We report performance across different datasets (SUNCG, NYUv2), and show that our approach significantly improves over independent prediction approaches while also outperforming alternate implicit reasoning methods.""","""With ratings of 6, 5 & 3 the numerical scores are just not strong enough to warrant acceptance. The author rebuttal was not able to sway opinions. """
1115,"""Adaptive Posterior Learning: few-shot learning with a surprise-based memory module""","['metalearning', 'memory', 'few-shot', 'relational', 'self-attention', 'classification', 'sequential', 'reasoning', 'working memory', 'episodic memory']","""The ability to generalize quickly from few observations is crucial for intelligent systems. In this paper we introduce APL, an algorithm that approximates probability distributions by remembering the most surprising observations it has encountered. These past observations are recalled from an external memory module and processed by a decoder network that can combine information from different memory slots to generalize beyond direct recall. We show this algorithm can perform as well as state of the art baselines on few-shot classification benchmarks with a smaller memory footprint. In addition, its memory compression allows it to scale to thousands of unknown labels. Finally, we introduce a meta-learning reasoning task which is more challenging than direct classification. In this setting, APL is able to generalize with fewer than one example per class via deductive reasoning.""","""All reviewers recommend acceptance. The problem is an interesting one. THe method is interesting. Authors were responsive in the reviewing process. Good work. I recommend acceptance :)"""
1116,"""Probabilistic Neural-Symbolic Models for Interpretable Visual Question Answering""","['Neural-symbolic models', 'visual question answering', 'reasoning', 'interpretability', 'graphical models', 'variational inference']","""We propose a new class of probabilistic neural-symbolic models for visual question answering (VQA) that provide interpretable explanations of their decision making in the form of programs, given a small annotated set of human programs. The key idea of our approach is to learn a rich latent space which effectively propagates program annotations from known questions to novel questions. We do this by formalizing prior work on VQA, called module networks (Andreas, 2016) as discrete, structured, latent variable models on the joint distribution over questions and answers given images, and devise a procedure to train the model effectively. Our results on a dataset of compositional questions about SHAPES (Andreas, 2016) show that our model generates more interpretable programs and obtains better accuracy on VQA in the low-data regime than prior work. ""","""This paper proposes a latent variable approach to the neural module networks of Andreas et al, whereby the program determining the structure of a module network is a structured discrete latent variable. The authors explore inference mechanisms over such programs and evaluate them on SHAPES. This paper may seem acceptable on the basis of its scores, but R1 (in particular) and R3 did a shambolic job of reviewing: their reviews are extremely short, and offer no substance to justify their scores. R2 has admirably engaged in discussion and upped their score to 6, but continue to find the paper fairly borderline, as do I. Weighing the reviews by the confidence I have in the reviewers based on their engagement, I would have to concur with R2 that this paper is very borderline. I like the core idea, but agree that the presentation of the inference techniques for V-NMN is complex and its presentation could stand to be significantly improved. I appreciate that the authors have made some updates on the basis of R2's feedback, but unfortunately due to the competitive nature of this year's ICLR and the number of acceptable paper, I cannot fully recommend acceptance at this time. As a complete side note, it is surprising not to see the Kingma & Welling (2013) VAE paper cited here, given the topic."""
1117,"""Difference-Seeking Generative Adversarial Network""","['Generative Adversarial Network', 'Semi-Supervised Learning', 'Adversarial Training']","""We propose a novel algorithm, Difference-Seeking Generative Adversarial Network (DSGAN), developed from traditional GAN. DSGAN considers the scenario that the training samples of target distribution, pseudo-formula , are difficult to collect. Suppose there are two distributions pseudo-formula and pseudo-formula such that the density of the target distribution can be the differences between the densities of pseudo-formula and pseudo-formula . We show how to learn the target distribution pseudo-formula only via samples from pseudo-formula and pseudo-formula (relatively easy to obtain). DSGAN has the flexibility to produce samples from various target distributions (e.g. the out-of-distribution). Two key applications, semi-supervised learning and adversarial training, are taken as examples to validate the effectiveness of DSGAN. We also provide theoretical analyses about the convergence of DSGAN.""","""The paper presents a GAN for learning a target distribution that is defined as the difference between two other distributions. The reviewers and AC note the critical limitation of novelty and appealing results of this paper to meet the high standard of ICLR. AC thinks the proposed method has potential and is interesting, but decided that the authors need more works to publish."""
1118,"""Exploiting Environmental Variation to Improve Policy Robustness in Reinforcement Learning""","['Reinforcement Learning', 'Policy Robustness', 'Policy generalization', 'Automated Curriculum']","""Conventional reinforcement learning rarely considers how the physical variations in the environment (eg. mass, drag, etc.) affect the policy learned by the agent. In this paper, we explore how changes in the environment affect policy generalization. We observe experimentally that, for each task we considered, there exists an optimal environment setting that results in the most robust policy that generalizes well to future environments. We propose a novel method to exploit this observation to develop robust actor policies, by automatically developing a sampling curriculum over environment settings to use in training. Ours is a model-free approach and experiments demonstrate that the performance of our method is on par with the best policies found by an exhaustive grid search, while bearing a significantly lower computational cost.""","""The paper presents a strategy for randomizing the underlying physical hyper-parameters of RL environments to improve policy's robustness. The paper has a simple and effective idea, however, the machine learning content is minimal. I agree with the reviewers that in order for the paper to pass the bar at ICLR, either the proposed ideas need to be extended theoretically or it should be backed with much more convincing results. Please take the reviewers' feedback into account and improve the paper."""
1119,"""D-GAN: Divergent generative adversarial network for positive unlabeled learning and counter-examples generation""",['Representation learning. Generative Adversarial Network (GAN). Positive Unlabeled learning. Image classification'],"""Positive Unlabeled (PU) learning consists in learning to distinguish samples of our class of interest, the positive class, from the counter-examples, the negative class, by using positive labeled and unlabeled samples during the training. Recent approaches exploit the GANs abilities to address the PU learning problem by generating relevant counter-examples. In this paper, we propose a new GAN-based PU learning approach named Divergent-GAN (D-GAN). The key idea is to incorporate a standard Positive Unlabeled learning risk inside the GAN discriminator loss function. In this way, the discriminator can ask the generator to converge towards the unlabeled samples distribution while diverging from the positive samples distribution. This enables the generator convergence towards the unlabeled counter-examples distribution without using prior knowledge, while keeping the standard adversarial GAN architecture. In addition, we discuss normalization techniques in the context of the proposed framework. Experimental results show that the proposed approach overcomes previous GAN-based PU learning methods issues, and it globally outperforms two-stage state of the art PU learning performances in terms of stability and prediction on both simple and complex image datasets.""","""With positive unlabeled learning the paper targets an interesting problem and proposes a new GAN based method to tackle it. All reviewers however agree that the write-up and the motivation behind the method could be made more clear and that novelty compared to other GAN based methods is limited. Also the experimental analysis does not show a strong clear performance advantage over existing models. """
1120,"""Excitation Dropout: Encouraging Plasticity in Deep Neural Networks""","['Dropout', 'Saliency', 'Deep Neural Networks']","""We propose a guided dropout regularizer for deep networks based on the evidence of a network prediction: the firing of neurons in specific paths. In this work, we utilize the evidence at each neuron to determine the probability of dropout, rather than dropping out neurons uniformly at random as in standard dropout. In essence, we dropout with higher probability those neurons which contribute more to decision making at training time. This approach penalizes high saliency neurons that are most relevant for model prediction, i.e. those having stronger evidence. By dropping such high-saliency neurons, the network is forced to learn alternative paths in order to maintain loss minimization, resulting in a plasticity-like behavior, a characteristic of human brains too. We demonstrate better generalization ability, an increased utilization of network neurons, and a higher resilience to network compression using several metrics over four image/video recognition benchmarks.""","""The reviewers overall agree that excitation dropout is a novel idea that seems to produce good empirical performance. However, they remain optimistic, but unconvinced by the experiments in their current form. The authors have done an admiral job of addressing this through more experiments, including providing error bars, however it seems as though the reviewers still require more. I would recommend creating tables of architecture x dropout technique, where dropout technique includes information dropout, adaptive dropout, curriculum dropout, and standard dropout, across several standard datasets. Alternatively, the authors could try to be more ambitious and classify Imagenet. Essentially, it seems as though the current small-scale datasets have become somewhat saturated, and therefore the bar for gauging a new method on them is higher in terms of experimental rigor. This means the best strategy is to either try more difficult benchmarks, or be extremely thorough and complete in your experiments. Regarding the wide resnet result, while I can appreciate that the original version published with higher errors, the later draft should still be taken into account as it has a) been out for a while now and b) can been reproduced in open source implementations (e.g., pseudo-url)."""
1121,"""Meta-Learning Update Rules for Unsupervised Representation Learning""","['Meta-learning', 'unsupervised learning', 'representation learning']","""A major goal of unsupervised learning is to discover data representations that are useful for subsequent tasks, without access to supervised labels during training. Typically, this involves minimizing a surrogate objective, such as the negative log likelihood of a generative model, with the hope that representations useful for subsequent tasks will arise as a side effect. In this work, we propose instead to directly target later desired tasks by meta-learning an unsupervised learning rule which leads to representations useful for those tasks. Specifically, we target semi-supervised classification performance, and we meta-learn an algorithm -- an unsupervised weight update rule -- that produces representations useful for this task. Additionally, we constrain our unsupervised update rule to a be a biologically-motivated, neuron-local function, which enables it to generalize to different neural network architectures, datasets, and data modalities. We show that the meta-learned update rule produces useful features and sometimes outperforms existing unsupervised learning techniques. We further show that the meta-learned unsupervised update rule generalizes to train networks with different widths, depths, and nonlinearities. It also generalizes to train on data with randomly permuted input dimensions and even generalizes from image datasets to a text task.""","""The reviewers all agree that the idea is interesting, the writing clear and the experiments sufficient. To improve the paper, the authors should consider better discussing their meta-objective and some of the algorithmic choices. """
1122,"""Hindsight policy gradients""","['reinforcement learning', 'policy gradients', 'multi-goal reinforcement learning']","""A reinforcement learning agent that needs to pursue different goals across episodes requires a goal-conditional policy. In addition to their potential to generalize desirable behavior to unseen goals, such policies may also enable higher-level planning based on subgoals. In sparse-reward environments, the capacity to exploit information about the degree to which an arbitrary goal has been achieved while another goal was intended appears crucial to enable sample efficient learning. However, reinforcement learning agents have only recently been endowed with such capacity for hindsight. In this paper, we demonstrate how hindsight can be introduced to policy gradient methods, generalizing this idea to a broad class of successful algorithms. Our experiments on a diverse selection of sparse-reward environments show that hindsight leads to a remarkable increase in sample efficiency.""","""The paper generalizes the concept of ""hindsight"", i.e. the recycling of data from trajectories in a goal-based system based on the goal state actually achieved, to policy gradient methods. This was an interesting paper in that it scored quite highly despite all three reviewers mentioning incrementality or a relative lack of novelty. Although the authors naturally took some exception to this, AC personally believes that properly executed, contributions that seem quite straightforward in hindsight (pun partly intended) can be valuable in moving the field forward: a clean and didactic presentation of theory backed by well-designed and extensive empirical investigation (both of which are adjectives used by reviewers to describe the empirical work in this paper) can be as valuable, or moreso, than a poorly executed but higher-novelty works. To quote AnonReviewer3, ""HPG is almost certainly going to end up being a widely used addition to the RL toolbox"". Feedback from reviewers prompted extensive discussion and a direct comparison with Hindsight Experience Replay which reviewers agreed added significant value to the manuscript, earning it a post-rebuttal unanimous rating of 7. It is therefore my pleasure to recommend acceptance."""
1123,"""DynCNN: An Effective Dynamic Architecture on Convolutional Neural Network for Surveillance Videos""","['CNN optimization', 'Reduction on convolution calculation', 'dynamic convolution', 'surveillance video']","""The large-scale surveillance video analysis becomes important as the development of intelligent city. The heavy computation resources neccessary for state-of-the-art deep learning model makes the real-time processing hard to be implemented. This paper exploits the characteristic of high scene similarity generally existing in surveillance videos and proposes dynamic convolution reusing the previous feature map to reduce the computation amount. We tested the proposed method on 45 surveillance videos with various scenes. The experimental results show that dynamic convolution can reduce up to 75.7% of FLOPs while preserving the precision within 0.7% mAP. Furthermore, the dynamic convolution can enhance the processing time up to 2.2 times.""","""The paper proposes a method for saving computation in surveillance videos (videos without camera motion) by re-using features from parts of the image that do not change. The results show that this significantly saves computation time, which is a big benefit, given also the amount of surveillance video input available for processing nowadays. Reviewers request comparisons to obvious baselines, e.g., selecting a subset of frames for processing or performing a low level pixel matching to select the pixels to compute new features on. Such experiments would make this paper much stronger. There is no rebuttal and thus no ground for discussion or acceptance. """
1124,"""Canonical Correlation Analysis with Implicit Distributions""","['Canonical Correlation Analysis', 'implicit probabilistic model', 'cross-view structure output prediction']","""Canonical Correlation Analysis (CCA) is a ubiquitous technique that shows promising performance in multi-view learning problems. Due to the conjugacy of the prior and the likelihood, probabilistic CCA (PCCA) presents the posterior with an analytic solution, which provides probabilistic interpretation for classic linear CCA. As the multi-view data are usually complex in practice, nonlinear mappings are adopted to capture nonlinear dependency among the views. However, the interpretation provided in PCCA cannot be generalized to this nonlinear setting, as the distribution assumptions on the prior and the likelihood makes it restrictive to capture nonlinear dependency. To overcome this bottleneck, in this paper, we provide a novel perspective for CCA based on implicit distributions. Specifically, we present minimum Conditional Mutual Information (CMI) as a new criteria to capture nonlinear dependency for multi-view learning problem. To eliminate the explicit distribution requirement in direct estimation of CMI, we derive an objective whose minimization implicitly leads to the proposed criteria. Based on this objective, we present an implicit probabilistic formulation for CCA, named Implicit CCA (ICCA), which provides a flexible framework to design CCA extensions with implicit distributions. As an instantiation, we present adversarial CCA (ACCA), a nonlinear CCA variant which benefits from consistent encoding achieved by adversarial learning. Quantitative correlation analysis and superior performance on cross-view generation task demonstrate the superiority of the proposed ACCA.""","""This manuscript proposes an implicit generative modeling approach for the non-linear CCA problem. One contribution is the proposal of Conditional Mutual Information (CMI) as a criterion to capture nonlinear dependency, resulting in an objective that can be solved using implicit distributions. The work seems to be well motivated and of interest to the community. The reviewers and AC opinions were mixed, and the rebuttal did not completely address the concerns. In particular, a reviewer pointed out an issue with a derivation in the paper, and the issue was not satisfactorily resolved by the authors. Some additional reading suggests that the misunderstanding may be partially due to incomplete notation and other issues with clarity of writing."""
1125,"""Multi-step Retriever-Reader Interaction for Scalable Open-domain Question Answering""","['Open domain Question Answering', 'Reinforcement Learning', 'Query reformulation']","""This paper introduces a new framework for open-domain question answering in which the retriever and the reader \emph{iteratively interact} with each other. The framework is agnostic to the architecture of the machine reading model provided it has \emph{access} to the token-level hidden representations of the reader. The retriever uses fast nearest neighbor search that allows it to scale to corpora containing millions of paragraphs. A gated recurrent unit updates the query at each step conditioned on the \emph{state} of the reader and the \emph{reformulated} query is used to re-rank the paragraphs by the retriever. We conduct analysis and show that iterative interaction helps in retrieving informative paragraphs from the corpus. Finally, we show that our multi-step-reasoning framework brings consistent improvement when applied to two widely used reader architectures (\drqa and \bidaf) on various large open-domain datasets ---\tqau, \quasart, \searchqa, and \squado\footnote{Code and pretrained models are available at \url{pseudo-url}}.""",""" pros: - novel idea for multi-step QA which rewrites the query in embedding space - good comparison with related work - reasonable evaluation and improved results cons: There were concerns about missing training details, insufficient evaluation, and presentation. These have been largely addressed in revision and I am recommending acceptance."""
1126,"""CGNF: Conditional Graph Neural Fields""","['graph neural networks', 'energy models', 'conditional random fields', 'label correlation']","""Graph convolutional networks have achieved tremendous success in the tasks of graph node classification. These models could learn a better node representation through encoding the graph structure and node features. However, the correlation between the node labels are not considered. In this paper, we propose a novel architecture for graph node classification, named conditional graph neural fields (CGNF). By integrating the conditional random fields (CRF) in the graph convolutional networks, we explicitly model a joint probability of the entire set of node labels, thus taking advantage of neighborhood label information in the node label prediction task. Our model could have both the representation capacity of graph neural networks and the prediction power of CRFs. Experiments on several graph datasets demonstrate effectiveness of CGNF.""",""" This paper introduces conditional graph neural fields, an approach that combines label compatibility scoring of conditional random fields with deep neural representations of nodes provided by graph convolutional networks. The intuition behind the proposed work is promising, and the results are strong. The reviewers and the AC note the following as the primary concerns of the paper: (1) The novelty of this work is limited, since a number of approaches have recently combined CRFs and neural networks, and it is unclear whether the application of those ideas to GCNs is sufficiently interesting, (2) the losses, especially EBM, and the use of greedy/beam-search inference was found to be quite simple, especially given these have been studied extensively in the literature, and (3) analysis and adequate discussion of the results is missing (only a single table of numbers is provided). Amongst other concerns, the reviewers identified issues with writing quality, lack of clear motivation for CRFs, and the selection of the benchmarks. Given the feedback, the authors responded with comments, and a revision that removes the use of EBM loss from the paper, which the reviewers appreciated. However, most of the concerns remain unaddressed. Reviewer 2 maintains that CRFs+NNs still need to be motivated better, since hidden representations already take the neighborhood into account, as demonstrated by the fact that CRF+NNs are not state-of-art in other applications. Reviewer 2 also points out the lack of a detailed analysis of the results. Reviewer 2 focuses on the simplicity of the loss and inference algorithms, which is also echoed by reviewer 2 and reviewer 1. Finally, reviewer 1 also notes that the datasets are quite simple, and not ideal evaluation for label consistency given most of them are single-label (and thus need only few transition probabilities). Based on this discussion, the reviewers and the AC agree that the paper is not ready for acceptance."""
1127,"""Learning to Design RNA""","['matter engineering', 'bioinformatics', 'rna design', 'reinforcement learning', 'meta learning', 'neural architecture search', 'hyperparameter optimization']","""Designing RNA molecules has garnered recent interest in medicine, synthetic biology, biotechnology and bioinformatics since many functional RNA molecules were shown to be involved in regulatory processes for transcription, epigenetics and translation. Since an RNA's function depends on its structural properties, the RNA Design problem is to find an RNA sequence which satisfies given structural constraints. Here, we propose a new algorithm for the RNA Design problem, dubbed LEARNA. LEARNA uses deep reinforcement learning to train a policy network to sequentially design an entire RNA sequence given a specified target structure. By meta-learning across 65000 different RNA Design tasks for one hour on 20 CPU cores, our extension Meta-LEARNA constructs an RNA Design policy that can be applied out of the box to solve novel RNA Design tasks. Methodologically, for what we believe to be the first time, we jointly optimize over a rich space of architectures for the policy network, the hyperparameters of the training procedure and the formulation of the decision process. Comprehensive empirical results on two widely-used RNA Design benchmarks, as well as a third one that we introduce, show that our approach achieves new state-of-the-art performance on the former while also being orders of magnitudes faster in reaching the previous state-of-the-art performance. In an ablation study, we analyze the importance of our method's different components. ""","""After a healthy discussion between reviewers and authors, the reviewers' consensus is to recommend acceptance to ICLR. The authors thoroughly addressed reviewer concerns, and all reviewers noted the quality of the paper, methodological innovations and SotA results."""
1128,"""Backpropamine: training self-modifying neural networks with differentiable neuromodulated plasticity""","['meta-learning', 'reinforcement learning', 'plasticity', 'neuromodulation', 'Hebbian learning', 'recurrent neural networks']","""The impressive lifelong learning in animal brains is primarily enabled by plastic changes in synaptic connectivity. Importantly, these changes are not passive, but are actively controlled by neuromodulation, which is itself under the control of the brain. The resulting self-modifying abilities of the brain play an important role in learning and adaptation, and are a major basis for biological reinforcement learning. Here we show for the first time that artificial neural networks with such neuromodulated plasticity can be trained with gradient descent. Extending previous work on differentiable Hebbian plasticity, we propose a differentiable formulation for the neuromodulation of plasticity. We show that neuromodulated plasticity improves the performance of neural networks on both reinforcement learning and supervised learning tasks. In one task, neuromodulated plastic LSTMs with millions of parameters outperform standard LSTMs on a benchmark language modeling task (controlling for the number of parameters). We conclude that differentiable neuromodulation of plasticity offers a powerful new framework for training neural networks.""","""The authors consider the problem of active plasticity in the mammalian brain, seen as being a means to enable lifelong learning. Building on the recent paper on differentiable plasticity, the authors propose a learnt, neuro-modulated differentiable plasticity that can be trained with gradient descent but is more flexible than fixed plasticity. The paper is clearly motivated and written, and the tasks are constructed to validate the method by demonstrating clear cases where non-modulated plasticity fails completely but where the proposed approach succeeds. On a large, general language modeling task (PTB) there is a small but consistent improvement over LSTMS. The reviewers were very split on this submission, with two reviewers focusing on the lack of large improvements on large benchmarks, and the other reviewer focusing on the novelty and success of the method on simple tasks. The AC tends to side with the positive review because of the following observations: the method is novel and potentially will have long term impact on the field, the language modeling task seems like a poor fit to demonstrate the advantages of the dynamic plasticity, so focusing on that benchmark overly much is misleading, and the paper is high-quality and interesting to the community. """
1129,"""Uncertainty in Multitask Transfer Learning""","['Multi Task', 'Transfer Learning', 'Hierarchical Bayes', 'Variational Bayes', 'Meta Learning', 'Few Shot learning']","""Using variational Bayes neural networks, we develop an algorithm capable of accumulating knowledge into a prior from multiple different tasks. This results in a rich prior capable of few-shot learning on new tasks. The posterior can go beyond the mean field approximation and yields good uncertainty on the performed experiments. Analysis on toy tasks show that it can learn from significantly different tasks while finding similarities among them. Experiments on Mini-Imagenet reach state of the art with 74.5% accuracy on 5 shot learning. Finally, we provide two new benchmarks, each showing a failure mode of existing meta learning algorithms such as MAML and prototypical Networks.""","""This paper presents a meta-learning approach which relies on a learned prior over neural networks for different tasks. The reviewers found this work to be well-motivated and timely. While there are some concerns regarding experiments, the results in the miniImageNet one seem to have impressed some reviewers. However, all reviewers found the presentation to be inaccurate in more than one points. R1 points out to ""issues with presentation"" for the hierarchical Bayes motivation, R2 mentions that the motivation and derivation in Section 2 is ""misleading"" and R3 talks about ""short presentation shortcomings"". R3 also raises important concerns about correctness of the derivation. The authors have replied to the correctness critique by explaining that the paper has been proofread by strong mathematicians, however they do not specifically rebut R3's points. The authors requested R3 to more specifically point to the location of the error, however it seems that R3 had already explained in a very detailed manner the source of the concern, including detailed equations. There have been other raised issues, such as concerns about experimental evaluation. However, the reviewers' almost complete agreement in the presentation issue is a clear signal that this paper needs to be substantially re-worked. """
1130,"""Feature Attribution As Feature Selection""","['feature attribution', 'feature selection']","""Feature attribution methods identify ""relevant"" features as an explanation of a complex machine learning model. Several feature attribution methods have been proposed; however, only a few studies have attempted to define the ""relevance"" of each feature mathematically. In this study, we formalize the feature attribution problem as a feature selection problem. In our proposed formalization, there arise two possible definitions of relevance. We name the feature attribution problems based on these two relevances as Exclusive Feature Selection (EFS) and Inclusive Feature Selection (IFS). We show that several existing feature attribution methods can be interpreted as approximation algorithms for EFS and IFS. Moreover, through exhaustive experiments, we show that IFS is better suited as the formalization for the feature attribution problem than EFS.""","""All in all, while the reviewers found that the problem at hand is interesting to study, the submission's contributions in terms of significance/novelty did not rise to the standards for acceptance. The reasoning is most succinctly discussed by R3 who argues that IFS and EFS are basically feature selection and applying them to feature attribution is not particularly novel from a methodological point of view. """
1131,"""Discriminative out-of-distribution detection for semantic segmentation""","['out-of-distribution detection', 'semantic segmentation']","""Most classification and segmentation datasets assume a closed-world scenario in which predictions are expressed as distribution over a predetermined set of visual classes. However, such assumption implies unavoidable and often unnoticeable failures in presence of out-of-distribution (OOD) input. These failures are bound to happen in most real-life applications since current visual ontologies are far from being comprehensive. We propose to address this issue by discriminative detection of OOD pixels in input data. Different from recent approaches, we avoid to bring any decisions by only observing the training dataset of the primary model trained to solve the desired computer vision task. Instead, we train a dedicated OOD model which discriminates the primary training set from a much larger ""background"" dataset which approximates the variety of the visual world. We perform our experiments on high resolution natural images in a dense prediction setup. We use several road driving datasets as our training distribution, while we approximate the background distribution with the ILSVRC dataset. We evaluate our approach on WildDash test, which is currently the only public test dataset with out-of-distribution images. The obtained results show that the proposed approach succeeds to identify out-of-distribution pixels while outperforming previous work by a wide margin.""","""The paper addresses the problem of out-of-distribution detection for helping the segmentation process. The reviewers and AC note the critical limitation of novelty of this paper to meet the high standard of ICLR. AC also thinks the authors should avoid using explicit OOD datasets (e.g., ILVRC) due to the nature of this problem. Otherwise, this is a toy binary classification problem. AC thinks the proposed method has potential and is interesting, but decided that the authors need more works to publish."""
1132,"""Variational recurrent models for representation learning""","['Representation learning', 'variational model']","""We study the problem of learning representations of sequence data. Recent work has built on variational autoencoders to develop variational recurrent models for generation. Our main goal is not generation but rather representation learning for downstream prediction tasks. Existing variational recurrent models typically use stochastic recurrent connections to model the dependence among neighboring latent variables, while generation assumes independence of generated data per time step given the latent sequence. In contrast, our models assume independence among all latent variables given non-stochastic hidden states, which speeds up inference, while assuming dependence of observations at each time step on all latent variables, which improves representation quality. In addition, we propose and study extensions for improving downstream performance, including hierarchical auxiliary latent variables and prior updating during training. Experiments show improved performance on several speech and language tasks with different levels of supervision, as well as in a multi-view learning setting.""","""This paper heavily modifies standard time-series-VAE models to improve their representation learning abilities. However, the resulting model seems like an ad-hoc combination of tricks that lose most of the nice properties of VAEs. The resulting method does not appear to be useful enough to justify itself, and it's not clear that the same ends couldn't be pursued using simpler, more general, and computationally cheaper approaches."""
1133,"""Probabilistic Model-Based Dynamic Architecture Search""","['architecture search', 'stochastic natural gradient', 'convolutional neural networks']","""The architecture search methods for convolutional neural networks (CNNs) have shown promising results. These methods require significant computational resources, as they repeat the neural network training many times to evaluate and search the architectures. Developing the computationally efficient architecture search method is an important research topic. In this paper, we assume that the structure parameters of CNNs are categorical variables, such as types and connectivities of layers, and they are regarded as the learnable parameters. Introducing the multivariate categorical distribution as the underlying distribution for the structure parameters, we formulate a differentiable loss for the training task, where the training of the weights and the optimization of the parameters of the distribution for the structure parameters are coupled. They are trained using the stochastic gradient descent, leading to the optimization of the structure parameters within a single training. We apply the proposed method to search the architecture for two computer vision tasks: image classification and inpainting. The experimental results show that the proposed architecture search method is fast and can achieve comparable performance to the existing methods.""","""The paper presents an architecture search method which jointly optimises the architecture and its weights. As noted by reviewers, the method is very close to Shirakawa et al., with the main innovation being the use of categorical distributions to model the architecture. This is a minor innovation, and while the results are promising, they are not strong enough to justify acceptance based on the results alone."""
1134,"""Variational Autoencoder with Arbitrary Conditioning""","['unsupervised learning', 'generative models', 'conditional variational autoencoder', 'variational autoencoder', 'missing features multiple imputation', 'inpainting']","""We propose a single neural probabilistic model based on variational autoencoder that can be conditioned on an arbitrary subset of observed features and then sample the remaining features in ""one shot"". The features may be both real-valued and categorical. Training of the model is performed by stochastic variational Bayes. The experimental evaluation on synthetic data, as well as feature imputation and image inpainting problems, shows the effectiveness of the proposed approach and diversity of the generated samples.""","""This paper proposes a VAE model with arbitrary conditioning. It is a novel idea, and the model derivation and training approach are technically sound. Experiments are thoughtfully designed and include comparison with latest related works. R1 and R3 suggested the original version of the paper was lack of comparison with relevant work and the authors provided new experiments in the revision. The rebuttal also addressed a few other concerns about the novelty and clarity raised by R3. Based on the novel contribution in handling missing feature imputation with VAE, I would recommend to accept the paper. It is worth noticing that there is another submission to ICLR (pseudo-url) that shares a similar idea of constructing the inference network with binary masking, although it is designed for a pre-trained VAE model. There are still two weaknesses pointed out by R3 that would help improve the paper by addressing them: 1. The paper does not handle different kinds of missingness beyond missing at random. 2. VAE model makes the trade-off between computational complexity and accuracy. Point 1 would be a good direction for future research, and point 2 is a common problem for all VAE approaches. While the latter should not become a reason to reject the paper, I encourage the authors to take MCMC methods into account in the evaluation section. """
1135,"""Escaping Flat Areas via Function-Preserving Structural Network Modifications""","['deep learning', 'cnn', 'structural modification', 'optimization', 'saddle point']","""Hierarchically embedding smaller networks in larger networks, e.g.~by increasing the number of hidden units, has been studied since the 1990s. The main interest was in understanding possible redundancies in the parameterization, as well as in studying how such embeddings affect critical points. We take these results as a point of departure to devise a novel strategy for escaping from flat regions of the error surface and to address the slow-down of gradient-based methods experienced in plateaus of saddle points. The idea is to expand the dimensionality of a network in a way that guarantees the existence of new escape directions. We call this operation the opening of a tunnel. One may then continue with the larger network either temporarily, i.e.~closing the tunnel later, or permanently, i.e.~iteratively growing the network, whenever needed. We develop our method for fully-connected as well as convolutional layers. Moreover, we present a practical version of our algorithm that requires no network structure modification and can be deployed as plug-and-play into any current deep learning framework. Experimentally, our method shows significant speed-ups.""","""The paper proposes a method to escape saddle points by adding and removing units during training. The method does so by preserving the function when the unit is added while increasing the gradient norm to move away from the critical point. The experimental evaluation shows that the proposed method does escape when positioned at a saddle point - as found by the Newton method. The reviewers find the theoretical ideas interesting and novel, but they raised concerns about the method's applicability for typical initializations, the experimental setup, as well as the terminology used in the paper. The title and terminology were improved with the revision, but the other issues were not sufficiently addressed."""
1136,"""REVISTING NEGATIVE TRANSFER USING ADVERSARIAL LEARNING""","['Negative Transfer', 'Adversarial Learning']","""An unintended consequence of feature sharing is the model fitting to correlated tasks within the dataset, termed negative transfer. In this paper, we revisit the problem of negative transfer in multitask setting and find that its corrosive effects are applicable to a wide range of linear and non-linear models, including neural networks. We first study the effects of negative transfer in a principled way and show that previously proposed counter-measures are insufficient, particularly for trainable features. We propose an adversarial training approach to mitigate the effects of negative transfer by viewing the problem in a domain adaptation setting. Finally, empirical results on attribute prediction multi-task on AWA and CUB datasets further validate the need for correcting negative sharing in an end-to-end manner.""","""This paper proposes reducing so called ""negative transfer"" through adversarial feature learning. The application of DANN for this task is new. However, the problem setting and particular assumptions are not sufficiently justified. As commented by the reviewers and acknowledged by the authors there is miscommunication about the basic premise of negative transfer and the main assumptions about the target distribution and it's label distribution need further justification. The authors are advised to restructure their manuscript so as to clarify the main contribution, assumptions, and motivation for their problem statement. In addition, the paper in it's current form is lacking sufficient experimental evidence to conclude that the proposed approach is preferable compared to prior work (such as Li 2018 and Zhang 2018) and lacks the proper ablation to conclude that the elimination of negative transfer is the main source of improvements. We encourage the authors to improve these aspects of the work and resubmit to a future venue. """
1137,"""What Information Does a ResNet Compress?""","['Deep Learning', 'Information Bottleneck', 'Residual Neural Networks', 'Information Theory']","""The information bottleneck principle (Shwartz-Ziv & Tishby, 2017) suggests that SGD-based training of deep neural networks results in optimally compressed hidden layers, from an information theoretic perspective. However, this claim was established on toy data. The goal of the work we present here is to test these claims in a realistic setting using a larger and deeper convolutional architecture, a ResNet model. We trained PixelCNN++ models as inverse representation decoders to measure the mutual information between hidden layers of a ResNet and input image data, when trained for (1) classification and (2) autoencoding. We find that two stages of learning happen for both training regimes, and that compression does occur, even for an autoencoder. Sampling images by conditioning on hidden layers activations offers an intuitive visualisation to understand what a ResNets learns to forget.""","""This paper explores an approach to testing the information bottleneck hypothesis of deep learning, specifically the idea that layers in a deep model successively discard information about the input which is irrelevant to the task being performed by the model, in full-scale ResNet models that are too large to admit the more standard binning-based estimators used in other work. Instead, to lower-bound I(x;h), the authors propose using the log-likelihood of a generative model (PixelCNN++). They also attempt visualize what sort of information is lost and what is retained by examining PixelCNN++ reconstructions from the hidden representation at different positions in a ResNet trained to perform image classification on the CINIC-10 task. To lower-bound I(y;h), they perform classification. In the experiments, the evolution of the bounds on I(x;h) and I(y;h) are tracked as a function of training epoch, and visualizations (reconstructions of the input) are shown to support the argument that color-invariance and diversity of samples increases during the compression phase of training. These tests are done on models trained to perform either image classification or autoencoding. This paper enjoyed a good discussion between the reviewers and the authors. The reviewers liked the quantitative analysis of ""usable information"" using PixelCNN++, though R2 wanted additional experiments to better quantify the limitations of the PixelCNN++ model to provide the reader with a better understanding of plots in Fig. 3, as well as more points sampled during training. Both R2 and R3 had reservations about the qualitative analysis based on the visualizations, which constitute the bulk of the paper. Unfortunately, the PixelCNN++ training is computationally intensive enough that these requests could not be fulfilled during the ICLR discussion phase. While the AC recommends that this submission be rejected from ICLR, this is a promising line of research. The authors should address the constructive suggestions of R2 and R3 and submit this work elsewhere."""
1138,"""Beyond Winning and Losing: Modeling Human Motivations and Behaviors with Vector-valued Inverse Reinforcement Learning""",[],"""In recent years, reinforcement learning methods have been applied to model gameplay with great success, achieving super-human performance in various environments, such as Atari, Go and Poker. However, those studies mostly focus on winning the game and have largely ignored the rich and complex human motivations, which are essential for understanding the agents' diverse behavior. In this paper, we present a multi-motivation behavior modeling which investigates the multifaceted human motivations and models the underlying value structure of the agents. Our approach extends inverse RL to the vectored-valued setting which imposes a much weaker assumption than previous studies. The vectorized rewards incorporate Pareto optimality, which is a powerful tool to explain a wide range of behavior by its optimality. For practical assessment, our algorithm is tested on the World of Warcraft Avatar History dataset spanning three years of the gameplay. Our experiments demonstrate the improvement over the scalarization-based methods on real-world problem settings.""","""Pros: - new multi-objective approach to IRL - new algorithm - strong results - real-world dataset Cons: - straightforward theoretical extensions - unclear motivation - inappropriate empirical assessment metrics - weak rebuttal All the reviewers feel that the paper needs further improvements, and while the authors comment on some of these concerns, their rebuttal and revised paper does not address them sufficiently. So at this stage it is a (borderline) reject."""
1139,"""Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware""","['Trusted hardware', 'integrity', 'privacy', 'secure inference', 'SGX']","""As Machine Learning (ML) gets applied to security-critical or sensitive domains, there is a growing need for integrity and privacy for outsourced ML computations. A pragmatic solution comes from Trusted Execution Environments (TEEs), which use hardware and software protections to isolate sensitive computations from the untrusted software stack. However, these isolation guarantees come at a price in performance, compared to untrusted alternatives. This paper initiates the study of high performance execution of Deep Neural Networks (DNNs) in TEEs by efficiently partitioning DNN computations between trusted and untrusted devices. Building upon an efficient outsourcing scheme for matrix multiplication, we propose Slalom, a framework that securely delegates execution of all linear layers in a DNN from a TEE (e.g., Intel SGX or Sanctum) to a faster, yet untrusted, co-located processor. We evaluate Slalom by running DNNs in an Intel SGX enclave, which selectively delegates work to an untrusted GPU. For canonical DNNs (VGG16, MobileNet and ResNet variants) we obtain 6x to 20x increases in throughput for verifiable inference, and 4x to 11x for verifiable and private inference.""","""The authors propose a new method of securely evaluating neural networks. The reviewers were unanimous in their vote to accept. The paper is very well written, the idea is relatively simple, and so it is likely that this would make a nice presentation."""
1140,"""NECST: Neural Joint Source-Channel Coding""","['joint source-channel coding', 'deep generative models', 'unsupervised learning']","""For reliable transmission across a noisy communication channel, classical results from information theory show that it is asymptotically optimal to separate out the source and channel coding processes. However, this decomposition can fall short in the finite bit-length regime, as it requires non-trivial tuning of hand-crafted codes and assumes infinite computational power for decoding. In this work, we propose Neural Error Correcting and Source Trimming (NECST) codes to jointly learn the encoding and decoding processes in an end-to-end fashion. By adding noise into the latent codes to simulate the channel during training, we learn to both compress and error-correct given a fixed bit-length and computational budget. We obtain codes that are not only competitive against several capacity-approaching channel codes, but also learn useful robust representations of the data for downstream tasks such as classification. Finally, we learn an extremely fast neural decoder, yielding almost an order of magnitude in speedup compared to standard decoding methods based on iterative belief propagation. ""","""This paper proposes a principled solution to the problem of joint source-channel coding. The reviewers find the perspectives put forward in the paper refreshing and that the paper is well written. The background and motivation is explained really well. However, reviewers found the paper limited in terms of modeling choices and evaluation methodology. One major flaw is that the experiments are limited to unrealistic datasets, and does not evaluate the method on a realistic benchmarks. It is also questioned whether the error-correcting aspect is practically relevant. """
1141,"""Execution-Guided Neural Program Synthesis""",[],"""Neural program synthesis from input-output examples has attracted an increasing interest from both the machine learning and the programming language community. Most existing neural program synthesis approaches employ an encoder-decoder architecture, which uses an encoder to compute the embedding of the given input-output examples, as well as a decoder to generate the program from the embedding following a given syntax. Although such approaches achieve a reasonable performance on simple tasks such as FlashFill, on more complex tasks such as Karel, the state-of-the-art approach can only achieve an accuracy of around 77%. We observe that the main drawback of existing approaches is that the semantic information is greatly under-utilized. In this work, we propose two simple yet principled techniques to better leverage the semantic information, which are execution-guided synthesis and synthesizer ensemble. These techniques are general enough to be combined with any existing encoder-decoder-style neural program synthesizer. Applying our techniques to the Karel dataset, we can boost the accuracy from around 77% to more than 90%.""","""This paper presents a system which exploits semantic information of partial programs during program synthesis, and ensembling of synthesisers. The idea is general, and admirably simple. The explanation is clear, and the results are impressive. The reviewers, some after significant discussion, agree that this paper makes an import contribution and is one of the stronger papers in the conference. While some possible improvements to the method and experiment were discussed with the reviewers, it seems these are more suitable for future research, and that the paper is clearly publishable in its current form."""
1142,"""textTOvec: DEEP CONTEXTUALIZED NEURAL AUTOREGRESSIVE TOPIC MODELS OF LANGUAGE WITH DISTRIBUTED COMPOSITIONAL PRIOR""","['neural topic model', 'natural language processing', 'text representation', 'language modeling', 'information retrieval', 'deep learning']","""We address two challenges of probabilistic topic modelling in order to better estimate the probability of a word in a given context, i.e., P(wordjcontext) : (1) No Language Structure in Context: Probabilistic topic models ignore word order by summarizing a given context as a bag-of-word and consequently the semantics of words in the context is lost. In this work, we incorporate language structure by combining a neural autoregressive topic model (TM) with a LSTM based language model (LSTM-LM) in a single probabilistic framework. The LSTM-LM learns a vector-space representation of each word by accounting for word order in local collocation patterns, while the TM simultaneously learns a latent representation from the entire document. In addition, the LSTM-LM models complex characteristics of language (e.g., syntax and semantics), while the TM discovers the underlying thematic structure in a collection of documents. We unite two complementary paradigms of learning the meaning of word occurrences by combining a topic model and a language model in a unified probabilistic framework, named as ctx-DocNADE. (2) Limited Context and/or Smaller training corpus of documents: In settings with a small number of word occurrences (i.e., lack of context) in short text or data sparsity in a corpus of few documents, the application of TMs is challenging. We address this challenge by incorporating external knowledge into neural autoregressive topic models via a language modelling approach: we use word embeddings as input of a LSTM-LM with the aim to improve the wordtopic mapping on a smaller and/or short-text corpus. The proposed DocNADE extension is named as ctx-DocNADEe. We present novel neural autoregressive topic model variants coupled with neural language models and embeddings priors that consistently outperform state-of-theart generative topic models in terms of generalization (perplexity), interpretability (topic coherence) and applicability (retrieval and classification) over 6 long-text and 8 short-text datasets from diverse domains.""","""This paper presents an extension of an existing topic model, DocNADE. Compared to DocNADE and other existing bag-of-word topic models, the primary contribution of this work is to integrate neural language models into the topic model in order to address two limitations of the bag-of-word topic models: expressiveness and interpretability. In addtion, the paper presents an approach to integrate external knowledge into the neural topic models to address the empirical challenges of the application scenarios where there might be only a small training corpus or limited context available. Pros: The paper presents strong and extensive empirical results. The authors went above and beyond to strengthen their paper during the rebuttal and address all the reviewers' questions and suggestions (e.g., the submitted version had 7 baselines, and the revised version has 6 additional baselines per reviewers' requests). Cons: The paper builds on an earlier paper that introduced the DocNADE model. Thus, the modeling contribution is relatively marginal. On the other hand, the extended model, albeit based on a relatively simple idea, is still new and demonstrates strong empirical results. Verdict: Probably accept. While not groundbreaking, the proposed model is new and the empirical results are strong. """
1143,"""Self-Tuning Networks: Bilevel Optimization of Hyperparameters using Structured Best-Response Functions""","['hyperparameter optimization', 'game theory', 'optimization']","""Hyperparameter optimization can be formulated as a bilevel optimization problem, where the optimal parameters on the training set depend on the hyperparameters. We aim to adapt regularization hyperparameters for neural networks by fitting compact approximations to the best-response function, which maps hyperparameters to optimal weights and biases. We show how to construct scalable best-response approximations for neural networks by modeling the best-response as a single network whose hidden units are gated conditionally on the regularizer. We justify this approximation by showing the exact best-response for a shallow linear network with L2-regularized Jacobian can be represented by a similar gating mechanism. We fit this model using a gradient-based hyperparameter optimization algorithm which alternates between approximating the best-response around the current hyperparameters and optimizing the hyperparameters using the approximate best-response function. Unlike other gradient-based approaches, we do not require differentiating the training loss with respect to the hyperparameters, allowing us to tune discrete hyperparameters, data augmentation hyperparameters, and dropout probabilities. Because the hyperparameters are adapted online, our approach discovers hyperparameter schedules that can outperform fixed hyperparameter values. Empirically, our approach outperforms competing hyperparameter optimization methods on large-scale deep learning problems. We call our networks, which update their own hyperparameters online during training, Self-Tuning Networks (STNs).""","""The paper proposes an approach to hyperparameter tuning based on bilevel optimization, and demonstrates promising empirical results. Reviewer's concerns seem to be addressed well in rebuttals and extended version of the paper."""
1144,"""Domain Adaptation for Structured Output via Disentangled Patch Representations""","['Domain Adaptation', 'Feature Representation Learning', 'Semantic Segmentation']","""Predicting structured outputs such as semantic segmentation relies on expensive per-pixel annotations to learn strong supervised models like convolutional neural networks. However, these models trained on one data domain may not generalize well to other domains unequipped with annotations for model finetuning. To avoid the labor-intensive process of annotation, we develop a domain adaptation method to adapt the source data to the unlabeled target domain. To this end, we propose to learn discriminative feature representations of patches based on label histograms in the source domain, through the construction of a disentangled space. With such representations as guidance, we then use an adversarial learning scheme to push the feature representations in target patches to the closer distributions in source ones. In addition, we show that our framework can integrate a global alignment process with the proposed patch-level alignment and achieve state-of-the-art performance on semantic segmentation. Extensive ablation studies and experiments are conducted on numerous benchmark datasets with various settings, such as synthetic-to-real and cross-city scenarios.""","""The paper explores unsupervised domain adaptation when the output is structured. Here they focus experimentally on semantic segmentation in driving scenes and use the spatial structure of the scene to produce two losses for adaptation: one global and one patch based. The method tackles an important problem and proposes a first attempt at a new solution. While the the experiments are missing ablations and some comparisons to prior work as noted by the reviewers, the authors have provided comments in their rebuttal explaining the relation to the prior work and promising to include more in the revised manuscript. The paper is borderline, but falls short on the necessary updates requested by reviewers. The use of the structured output which is available in semantic segmentation of driving scenes is a useful direction. The paper is missing enough key results and analysis in it's current form to be accepted. """
1145,"""Empirically Characterizing Overparameterization Impact on Convergence""","['gradient descent', 'optimization', 'convergence time', 'halting time', 'characterization']","""A long-held conventional wisdom states that larger models train more slowly when using gradient descent. This work challenges this widely-held belief, showing that larger models can potentially train faster despite the increasing computational requirements of each training step. In particular, we study the effect of network structure (depth and width) on halting time and show that larger models---wider models in particular---take fewer training steps to converge. We design simple experiments to quantitatively characterize the effect of overparametrization on weight space traversal. Results show that halting time improves when growing model's width for three different applications, and the improvement comes from each factor: The distance from initialized weights to converged weights shrinks with a power-law-like relationship, the average step size grows with a power-law-like relationship, and gradient vectors become more aligned with each other during traversal. ""","""This paper studies the behavior of training of over parametrized models. All the reviewers agree that the questions studied in this paper are important. However the experiments in the paper are fairly preliminary and the paper does not offer any answers to the questions it studies. Further the writing is very loose and the paper is not ready for publication. I advise authors to take the reviews seriously into account before submitting the paper again. """
1146,"""TimbreTron: A WaveNet(CycleGAN(CQT(Audio))) Pipeline for Musical Timbre Transfer""","['Generative models', 'Timbre Transfer', 'Wavenet', 'CycleGAN']","""In this work, we address the problem of musical timbre transfer, where the goal is to manipulate the timbre of a sound sample from one instrument to match another instrument while preserving other musical content, such as pitch, rhythm, and loudness. In principle, one could apply image-based style transfer techniques to a time-frequency representation of an audio signal, but this depends on having a representation that allows independent manipulation of timbre as well as high-quality waveform generation. We introduce TimbreTron, a method for musical timbre transfer which applies image domain style transfer to a time-frequency representation of the audio signal, and then produces a high-quality waveform using a conditional WaveNet synthesizer. We show that the Constant Q Transform (CQT) representation is particularly well-suited to convolutional architectures due to its approximate pitch equivariance. Based on human perceptual evaluations, we confirmed that TimbreTron recognizably transferred the timbre while otherwise preserving the musical content, for both monophonic and polyphonic samples. We made an accompanying demo video here: pseudo-url which we strongly encourage you to watch before reading the paper.""","""Strengths: This paper is ""thorough and well written"", exploring the timbre transfer problem in a novel way. There is a video accompanying the work and some reviewers assessed the quality of the results as being good relative to other approaches. Two of the reviewers were quite positive about the work. Weaknesses: Reviewer 2 (the lowest scoring reviewer) felt that the paper was a little too far from solving the problem to be of high significance and that there was: - too much focus on STFT vs. CQT - too little focus on getting WaveNet synthesis right - too limited experimental validation (too restricted choice of instruments) - poor resulting audio quality - feels too much of combining black boxes AMT listening tests were performed, but better baselines could have been used. The author response addressed some of these points. Contention: An anonymous commenter noted that the revised manuscript added some names in the acknowledgements, thereby violating double blind review guidelines. However, the aggregated initial scores for this work were past the threshold for acceptance. Reviewer 2 was the most critical of the work but did not engage in dialog or comment on the author response. Consensus: The two positive reviewers felt that this work is worth of presentation at ICLR. The AC recommends accept as poster unless the PC feel the issue of names in the Acknowledgements in an updated draft is too serious of an issue. """
1147,"""Hierarchically-Structured Variational Autoencoders for Long Text Generation""","['Natural Language Processing', 'Text Generation', 'Variational Autoencoders']","""Variational autoencoders (VAEs) have received much attention recently as an end-to-end architecture for text generation. Existing methods primarily focus on synthesizing relatively short sentences (with less than twenty words). In this paper, we propose a novel framework, hierarchically-structured variational autoencoder (hier-VAE), for generating long and coherent units of text. To enhance the models plan-ahead ability, intermediate sentence representations are introduced into the generative networks to guide the word-level predictions. To alleviate the typical optimization challenges associated with textual VAEs, we further employ a hierarchy of stochastic layers between the encoder and decoder networks. Extensive experiments are conducted to evaluate the proposed method, where hier-VAE is shown to make effective use of the latent codes and achieve lower perplexity relative to language models. Moreover, the generated samples from hier-VAE also exhibit superior quality according to both automatic and human evaluations. ""","""Strengths: Interesting work on using latent variables for generating long text sequences. The paper shows convincing results on perplexity, N-gram based and human qualitative evaluation. Weaknesses: More extensive comparisons with hierarchical VAEs and the approach in Serban et. al in terms of language generation quality and perplexity would have been helpful. Another point of reference for which additional comparisons were desired was: ""A Hierarchical Latent Structure for Variational Conversation Modeling"" by Park et al. Some additional substantive experiments were added during the discussion period. Contention: Authors differentiated their work from Park et al. and the reviewer bringing this work up ended up upgrading their score to a 7. The other reviewers kept their scores at 5. Consensus: The positive reviewer raised their score to a 7 through the author rebuttal and discussion period. One negative reviewer was not responsive, but the other reviewer giving a 5 asserts that they maintain their position. The AC recommends rejection. Situating this work with respect to other prior work and properly comparing with it seems to be the contentious issue. Authors are encouraged to revise and re-submit elsewhere."""
1148,"""Human-level Protein Localization with Convolutional Neural Networks""","['Convolutional Neural Networks', 'High-resolution images', 'Multiple-Instance Learning', 'Microscopy Imaging', 'Protein Localization']","""Localizing a specific protein in a human cell is essential for understanding cellular functions and biological processes of underlying diseases. A promising, low-cost,and time-efficient biotechnology for localizing proteins is high-throughput fluorescence microscopy imaging (HTI). This imaging technique stains the protein of interest in a cell with fluorescent antibodies and subsequently takes a microscopic image. Together with images of other stained proteins or cell organelles and the annotation by the Human Protein Atlas project, these images provide a rich source of information on the protein location which can be utilized by computational methods. It is yet unclear how precise such methods are and whether they can compete with human experts. We here focus on deep learning image analysis methods and, in particular, on Convolutional Neural Networks (CNNs)since they showed overwhelming success across different imaging tasks. We pro-pose a novel CNN architecture GapNet-PL that has been designed to tackle the characteristics of HTI data and uses global averages of filters at different abstraction levels. We present the largest comparison of CNN architectures including GapNet-PL for protein localization in HTI images of human cells. GapNet-PL outperforms all other competing methods and reaches close to perfect localization in all 13 tasks with an average AUC of 98% and F1 score of 78%. On a separate test set the performance of GapNet-PL was compared with three human experts and 25 scholars. GapNet-PL achieved an accuracy of 91%, significantly (p-value 1.1e6) outperforming the best human expert with an accuracy of 72%.""","""The reviewers all agreed that the problem application is interesting, and that there is little new methodology, but disagreed as to how that should translate into a score. The highest rating seemed to heavily weight the importance of the method to biological application, whereas the lowest rating heavily weighted the lack of technical novelty. However, because the ICLR call for papers clearly calls out applications in biology, and all reviewers agreed on its strength in that regard, and it was well-written and executed, I would recommend it for acceptance."""
1149,"""Neural Predictive Belief Representations""","['belief states', 'representation learning', 'contrastive predictive coding', 'reinforcement learning', 'predictive state representations', 'deep reinforcement learning']","""Unsupervised representation learning has succeeded with excellent results in many applications. It is an especially powerful tool to learn a good representation of environments with partial or noisy observations. In partially observable domains it is important for the representation to encode a belief state---a sufficient statistic of the observations seen so far. In this paper, we investigate whether it is possible to learn such a belief representation using modern neural architectures. Specifically, we focus on one-step frame prediction and two variants of contrastive predictive coding (CPC) as the objective functions to learn the representations. To evaluate these learned representations, we test how well they can predict various pieces of information about the underlying state of the environment, e.g., position of the agent in a 3D maze. We show that all three methods are able to learn belief representations of the environment---they encode not only the state information, but also its uncertainty, a crucial aspect of belief states. We also find that for CPC multi-step predictions and action-conditioning are critical for accurate belief representations in visually complex environments. The ability of neural representations to capture the belief information has the potential to spur new advances for learning and planning in partially observable domains, where leveraging uncertainty is essential for optimal decision making.""","""This paper proposed an unsupervised learning algorithm for predictive modeling. The key idea of using NCE/CPC for predictive modeling is interesting. However, major concerns were raised by reviewers on the experimental design/empirical comparisons and paper writing. Overall, this paper cannot be published in its current form, but I think it may be dramatically improved for a future publication. """
1150,"""Morpho-MNIST: Quantitative Assessment and Diagnostics for Representation Learning""","['quantitative evaluation', 'diagnostics', 'generative models', 'representation learning', 'morphometrics', 'image perturbations']","""Revealing latent structure in data is an active field of research, having introduced exciting technologies such as variational autoencoders and adversarial networks, and is essential to push machine learning towards unsupervised knowledge discovery. However, a major challenge is the lack of suitable benchmarks for an objective and quantitative evaluation of learned representations. To address this issue we introduce Morpho-MNIST, a framework that aims to answer: ""to what extent has my model learned to represent specific factors of variation in the data?"" We extend the popular MNIST dataset by adding a morphometric analysis enabling quantitative comparison of trained models, identification of the roles of latent variables, and characterisation of sample diversity. We further propose a set of quantifiable perturbations to assess the performance of unsupervised and supervised methods on challenging tasks such as outlier detection and domain adaptation.""","""This paper presents a dataset for measuring disentanglement in learned representations. It consists of MNIST digits, sometimes transformed in various ways, and labeled with a variety of attributes. This dataset is used to measure statistics of various learned models. Measuring disentanglement is certainly an important problem in our field. This dataset seems to be well designed, and I would recommend its use for papers studying disentanglement. The experiments are well-designed. While the reviewers seem bothered by the fact that it's limited to MNIST, this doesn't strike me as a problem. We continue to learn a lot from MNIST, even today. But producing a useful dataset isn't by itself a significant enough research contribution for an ICLR paper. I'd recommend publication if (a) it were very different from currently existing datasets, (b) constructing it required overcoming significant technical obstacles, or (c) the dataset led to particularly interesting findings. Regarding (a), there are already datasets of similar complexity which have ground-truth attributes useful for measuring disentanglement, such as dSprites and 3D Faces. Regarding (b), the construction seems technically straightforward. Regarding (c), the experimental findings are plausible and consistent with past findings (which is a good validation of the dataset) but not obviously interesting in their own right. So overall, this seems like a useful dataset, but I cannot recommend publication at ICLR. """
1151,"""Riemannian Adaptive Optimization Methods""","['Riemannian optimization', 'adaptive', 'hyperbolic', 'curvature', 'manifold', 'adam', 'amsgrad', 'adagrad', 'rsgd', 'convergence']","""Several first order stochastic optimization methods commonly used in the Euclidean domain such as stochastic gradient descent (SGD), accelerated gradient descent or variance reduced methods have already been adapted to certain Riemannian settings. However, some of the most popular of these optimization tools - namely Adam, Adagrad and the more recent Amsgrad - remain to be generalized to Riemannian manifolds. We discuss the difficulty of generalizing such adaptive schemes to the most agnostic Riemannian setting, and then provide algorithms and convergence proofs for geodesically convex objectives in the particular case of a product of Riemannian manifolds, in which adaptivity is implemented across manifolds in the cartesian product. Our generalization is tight in the sense that choosing the Euclidean space as Riemannian manifold yields the same algorithms and regret bounds as those that were already known for the standard algorithms. Experimentally, we show faster convergence and to a lower train loss value for Riemannian adaptive methods over their corresponding baselines on the realistic task of embedding the WordNet taxonomy in the Poincare ball.""","""Dear authors, All reviewers agreed that your work sheds new light on a popular class of algorithms and should thus be presented at ICLR. Please make sure to implement all their comments in the final version."""
1152,"""Policy Transfer with Strategy Optimization""","['transfer learning', 'reinforcement learning', 'modeling error', 'strategy optimization']","""Computer simulation provides an automatic and safe way for training robotic control policies to achieve complex tasks such as locomotion. However, a policy trained in simulation usually does not transfer directly to the real hardware due to the differences between the two environments. Transfer learning using domain randomization is a promising approach, but it usually assumes that the target environment is close to the distribution of the training environments, thus relying heavily on accurate system identification. In this paper, we present a different approach that leverages domain randomization for transferring control policies to unknown environments. The key idea that, instead of learning a single policy in the simulation, we simultaneously learn a family of policies that exhibit different behaviors. When tested in the target environment, we directly search for the best policy in the family based on the task performance, without the need to identify the dynamic parameters. We evaluate our method on five simulated robotic control problems with different discrepancies in the training and testing environment and demonstrate that our method can overcome larger modeling errors compared to training a robust policy or an adaptive policy.""","""The paper presents quite a simple idea to transfer a policy between domains by conditioning the orginal learned policy on the physical parameter used in dynamics randomization. CMA-ES then finds the best parameters in the target domain. Importantly, it is shown to work well, for examples where the dynamics randomization parameters do not span the parameters that are actually changed, i.e., as is likely common in reality-gap problems. A weakness is the size of the contribution beyond UPOSI (Yu et al. 2017), the closest work. The authors now explicitly benchmark against this, with (generally) positive results. AC: It would be ideal to see that the method does truly help span the reality gap, by seeing working sim2real transfer. Overall, the reviewers and AC are in agreement that this is a good idea that is likely to have impact. Its fundamental simplicity means that it can also readily be used as a benchmark in future sim2real work. The AC recommend it be considered for oral presentation based on its simplicity, the importance of the sim2real problem, and particularly if it can be demonstrated to work well on actual sim2real transfer tasks (not yet shown in the current results). """
1153,"""FFJORD: Free-Form Continuous Dynamics for Scalable Reversible Generative Models""","['generative models', 'density estimation', 'approximate inference', 'ordinary differential equations']","""A promising class of generative models maps points from a simple distribution to a complex distribution through an invertible neural network. Likelihood-based training of these models requires restricting their architectures to allow cheap computation of Jacobian determinants. Alternatively, the Jacobian trace can be used if the transformation is specified by an ordinary differential equation. In this paper, we use Hutchinsons trace estimator to give a scalable unbiased estimate of the log-density. The result is a continuous-time invertible generative model with unbiased density estimation and one-pass sampling, while allowing unrestricted neural network architectures. We demonstrate our approach on high-dimensional density estimation, image generation, and variational inference, achieving the state-of-the-art among exact likelihood methods with efficient sampling.""","""This paper proposes the use of recently propose neural ODEs in a flow-based generative model. As the paper shows, a big advantage of a neural ODE in a generative flow is that an unbiased estimator of the log-determinant of the mapping is straightforward to construct. Another advantage, compared to earlier published flows, is that all variables can be updated in parallel, as the method does not require ""chopping up"" the variables into blocks. The paper shows significant improvements on several benchmarks, and seems to be a promising venue for further research. A disadvantage of the method is that the authors were unable to show that the method could produce results that were similar (of better than) the SOTA on the more challenging benchmark of CIFAR-10. Another downside is its computational cost. Since neural ODEs are relatively new, however, these problems might resolved with further refinements to the method. """
1154,"""Constrained Bayesian Optimization for Automatic Chemical Design""","['Bayesian Optimization', 'Generative Models']","""Automatic Chemical Design provides a framework for generating novel molecules with optimized molecular properties. The current model suffers from the pathology that it tends to produce invalid molecular structures. By reformulating the search procedure as a constrained Bayesian optimization problem, we showcase improvements in both the validity and quality of the generated molecules. We demonstrate that the model consistently produces novel molecules ranking above the 90th percentile of the distribution over training set scores across a range of objective functions. Importantly, our method suffers no degradation in the complexity or the diversity of the generated molecules.""","""This paper proposes to use constrained Bayesian optimization to improve the chemical compound generation. Unfortunately, the reviewers raises a range of critical issues which are not responded by authors' rebuttal. """
1155,""" Large-Scale Visual Speech Recognition""","['visual speech recognition', 'speech recognition', 'lipreading']","""This work presents a scalable solution to continuous visual speech recognition. To achieve this, we constructed the largest existing visual speech recognition dataset, consisting of pairs of text and video clips of faces speaking (3,886 hours of video). In tandem, we designed and trained an integrated lipreading system, consisting of a video processing pipeline that maps raw video to stable videos of lips and sequences of phonemes, a scalable deep neural network that maps the lip videos to sequences of phoneme distributions, and a production-level speech decoder that outputs sequences of words. The proposed system achieves a word error rate (WER) of 40.9% as measured on a held-out set. In comparison, professional lipreaders achieve either 86.4% or 92.9% WER on the same dataset when having access to additional types of contextual information. Our approach significantly improves on previous lipreading approaches, including variants of LipNet and of Watch, Attend, and Spell (WAS), which are only capable of 89.8% and 76.8% WER respectively.""","""This paper describes the development of a large-scale continuous visual speech recognition (lipreading) system, including an audiovisual processing pipeline that is used to extract stabilized videos of lips and corresponding phone sequences from YouTube videos, a deep network architecture trained with CTC loss that maps video sequences to sequences of distributions over phones, and an FST-based decoder that produces word sequences from the phone score sequences. A performance evaluation shows that the proposed system outperforms other models described in the literature, as well as professional lipreaders. A number of ablation experiments compare the performance of the proposed architecture to the previously proposed LipNet and ""Watch, Attend, and Spell"" architectures, explore the performance differences caused by using phone- or character-based CTC models, and some variations on the proposed architecture. This paper was extremely controversial and received a robust discussion between the authors and reviewers, with the primary point of contention being the suitability of the paper for ICLR. All reviewers agree that the quality of the work in the paper is excellent and that the reported results are impressive, but there was strong disagreement on whether or not this was sufficient for an ICLR paper. One reviewer thought so, while the other two reviewers argued that this is insufficient, and that to appear in ICLR the paper either (1) should have focused more on the preparation of the dataset, included public release of the data so other researchers could build on the work, and put forth the V2P model as a (very) strong baseline for the task; or (2) done a more in-depth exploration of the representation learning aspects of the work by comparing phoneme and viseme units and providing more (admittedly costly) ablation experiments to shed more light on what aspects of the V2P architecture lead to the reported improvements in performance. The AC finds the arguments of the two negative reviewers to be persuasive. It is quite clear at this point that many supervised classification tasks (even structured classification tasks like lipreading) can be effectively tackled by a combination of a sufficiently flexible learning architecture and collection of a massive, annotated dataset, and the modeling techniques used in this paper are not new, per se, even if their application to lipreading is. Moreover, if the dataset is not publicly available, it is impossible for anyone else to build on this work. The paper, as it currently stands, would be appropriate in a more applications-oriented venue."""
1156,"""Learning to Drive by Observing the Best and Synthesizing the Worst""","['Imitation Learning', 'End-to-End Driving', 'Learning to drive', 'Autonomous Driving']","""Our goal is to train a policy for autonomous driving via imitation learning that is robust enough to drive a real vehicle. We find that standard behavior cloning is insufficient for handling complex driving scenarios, even when we leverage a perception system for preprocessing the input and a controller for executing the output on the car: 30 million examples are still not enough. We propose exposing the learner to synthesized data in the form of perturbations to the expert's driving, which creates interesting situations such as collisions and/or going off the road. Rather than purely imitating all data, we augment the imitation loss with additional losses that penalize undesirable events and encourage progress -- the perturbations then provide an important signal for these losses and lead to robustness of the learned model. We show that the model can handle complex situations in simulation, and present ablation experiments that emphasize the importance of each of our proposed changes and show that the model is responding to the appropriate causal factors. Finally, we demonstrate the model driving a car in the real world ( pseudo-url ).""","""The authors present a method for training a policy for a self-driving car. The inputs to the policy are map-based perceptual features and the outputs are waypoints on a trajectory, and the method is an augmented imitation learning framework that uses perturbations and additional losses to make the policy more robust and effective in rare events. The paper is clear and well-written and the authors do demonstrate that it can be used to control a real vehicle. However, the reviewers all had concerns about the oracle feature representation which is the input and also concerns about the lack of baselines such as optimization based methods. They also felt that the approach was limited to self-driving cars and thus would have limited interest for the community."""
1157,"""Universal Attacks on Equivariant Networks""","['adversarial', 'equivariance', 'universal', 'rotation', 'translation', 'CNN', 'GCNN']","""Adversarial attacks on neural networks perturb the input at test time in order to fool trained and deployed neural network models. Most attacks such as gradient-based Fast Gradient Sign Method (FGSM) by Goodfellow et al. 2015 and DeepFool by Moosavi-Dezfooli et al. 2016 are input-dependent, small, pixel-wise perturbations, and they give different attack directions for different inputs. On the other hand, universal adversarial attacks are input-agnostic and the same attack works for most inputs. Translation or rotation-equivariant neural network models provide one approach to prevent universal attacks based on simple geometric transformations. In this paper, we observe an interesting spectral property shared by all of the above input-dependent, pixel-wise adversarial attacks on translation and rotation-equivariant networks. We exploit this property to get a single universal attack direction that fools the model on most inputs. Moreover, we show how to compute this universal attack direction using principal components of the existing input-dependent attacks on a very small sample of test inputs. We complement our empirical results by a theoretical justification, using matrix concentration inequalities and spectral perturbation bounds. We also empirically observe that the top few principal adversarial attack directions are nearly orthogonal to the top few principal invariant directions. ""","""The topic of universal adversarial perturbation is quite intriguing and fairly poorly studied and the paper provides a mix of new insights, both theoretical and empirical in nature. However, the significant presentation issues make it hard to properly understand and evaluate them. In particular, the theoretical part feels rushed and not sufficiently rigorous, and it is unclear why focusing on the case of equivariant network is crucial. Also, it would be useful if the authors put more effort in explaining how their contributions fit into the context of prior work in the area. Overall, this paper has a potential of becoming a solid contribution, once the above shortcomings are addressed."""
1158,"""On Inductive Biases in Deep Reinforcement Learning""",[],"""Many deep reinforcement learning algorithms contain inductive biases that sculpt the agent's objective and its interface to the environment. These inductive biases can take many forms, including domain knowledge and pretuned hyper-parameters. In general, there is a trade-off between generality and performance when we use such biases. Stronger biases can lead to faster learning, but weaker biases can potentially lead to more general algorithms that work on a wider class of problems. This trade-off is relevant because these inductive biases are not free; substantial effort may be required to obtain relevant domain knowledge or to tune hyper-parameters effectively. In this paper, we re-examine several domain-specific components that modify the agent's objective and environmental interface. We investigated whether the performance deteriorates when all these fixed components are replaced with adaptive solutions from the literature. In our experiments, performance sometimes decreased with the adaptive components, as one might expect when comparing to components crafted for the domain, but sometimes the adaptive components performed better. We then investigated the main benefit of having fewer domain-specific components, by comparing the learning performance of the two systems on a different set of continuous control problems, without additional tuning of either system. As hypothesized, the system with adaptive components performed better on many of the tasks.""","""The paper studies inductive biases in DRL, by comparing with different reward shaping, and curriculums. The authors performed comparative experiments where they replace domain specific heuristics by such adaptive components. The paper includes very little (new) scientific contributions, and, as such, is not suitable for publication at ICLR."""
1159,"""Learning Protein Structure with a Differentiable Simulator""","['generative models', 'simulators', 'molecular modeling', 'proteins', 'structured prediction']","""The Boltzmann distribution is a natural model for many systems, from brains to materials and biomolecules, but is often of limited utility for fitting data because Monte Carlo algorithms are unable to simulate it in available time. This gap between the expressive capabilities and sampling practicalities of energy-based models is exemplified by the protein folding problem, since energy landscapes underlie contemporary knowledge of protein biophysics but computer simulations are challenged to fold all but the smallest proteins from first principles. In this work we aim to bridge the gap between the expressive capacity of energy functions and the practical capabilities of their simulators by using an unrolled Monte Carlo simulation as a model for data. We compose a neural energy function with a novel and efficient simulator based on Langevin dynamics to build an end-to-end-differentiable model of atomic protein structure given amino acid sequence information. We introduce techniques for stabilizing backpropagation under long roll-outs and demonstrate the model's capacity to make multimodal predictions and to, in some cases, generalize to unobserved protein fold types when trained on a large corpus of protein structures.""","""This paper presents a differentiable simulator for protein structure prediction that can be trained end-to-end. It makes several contributions to this research area. Particularly training a differentiable sampling simulator could be of interest to a wider community. The main criticism comes from the clarity for the machine learning community and empirical comparison with the state-of-the-art methods. The authors' feedback addressed a few confusions in the description, and I recommend the authors to further polish the text for better readability. R4 argues that a good comparison with the state-of-the-art method in this field would be difficult and the comparison with an RNN baseline is rigorously carried out. After discussion, all reviewers agree that this paper deserves a publication at ICLR."""
1160,"""Composing Entropic Policies using Divergence Correction""","['maximum entropy RL', 'policy composition', 'deep rl']","""Deep reinforcement learning (RL) algorithms have made great strides in recent years. An important remaining challenge is the ability to quickly transfer existing skills to novel tasks, and to combine existing skills with newly acquired ones. In domains where tasks are solved by composing skills this capacity holds the promise of dramatically reducing the data requirements of deep RL algorithms, and hence increasing their applicability. Recent work has studied ways of composing behaviors represented in the form of action-value functions. We analyze these methods to highlight their strengths and weaknesses, and point out situations where each of them is susceptible to poor performance. To perform this analysis we extend generalized policy improvement to the max-entropy framework and introduce a method for the practical implementation of successor features in continuous action spaces. Then we propose a novel approach which, in principle, recovers the optimal policy during transfer. This method works by explicitly learning the (discounted, future) divergence between policies. We study this approach in the tabular case and propose a scalable variant that is applicable in multi-dimensional continuous action spaces. We compare our approach with existing ones on a range of non-trivial continuous control problems with compositional structure, and demonstrate qualitatively better performance despite not requiring simultaneous observation of all task rewards.""","""Multiple reviewers had concerns about the clarity of the presentation and the significance of the results. """
1161,"""Off-Policy Evaluation and Learning from Logged Bandit Feedback: Error Reduction via Surrogate Policy""","['Causal inference', 'Policy Optimization', 'Non-asymptotic analysis']",""" When learning from a batch of logged bandit feedback, the discrepancy between the policy to be learned and the off-policy training data imposes statistical and computational challenges. Unlike classical supervised learning and online learning settings, in batch contextual bandit learning, one only has access to a collection of logged feedback from the actions taken by a historical policy, and expect to learn a policy that takes good actions in possibly unseen contexts. Such a batch learning setting is ubiquitous in online and interactive systems, such as ad platforms and recommendation systems. Existing approaches based on inverse propensity weights, such as Inverse Propensity Scoring (IPS) and Policy Optimizer for Exponential Models (POEM), enjoy unbiasedness but often suffer from large mean squared error. In this work, we introduce a new approach named Maximum Likelihood Inverse Propensity Scoring (MLIPS) for batch learning from logged bandit feedback. Instead of using the given historical policy as the proposal in inverse propensity weights, we estimate a maximum likelihood surrogate policy based on the logged action-context pairs, and then use this surrogate policy as the proposal. We prove that MLIPS is asymptotically unbiased, and moreover, has a smaller nonasymptotic mean squared error than IPS. Such an error reduction phenomenon is somewhat surprising as the estimated surrogate policy is less accurate than the given historical policy. Results on multi-label classification problems and a large-scale ad placement dataset demonstrate the empirical effectiveness of MLIPS. Furthermore, the proposed surrogate policy technique is complementary to existing error reduction techniques, and when combined, is able to consistently boost the performance of several widely used approaches.""","""This is an interesting paper that shows how improved off-policy estimation (and optimization) can be improved by explicitly estimating the data logging policy. It is remarkable that the estimation variance can be reduced over using the original logging policy for IPW, although this result depends on the (somewhat impractical) assumption that the parametric form for the true logging policy is known. The reviewers unanimously recommended the paper be accepted. However, there remain criticisms of the theoretical analysis that the authors should take into account in preparing a final version (namely, motivating the assumptions needed to obtain the results, and providing stronger intuitions behind the reduced variance)."""
1162,"""Unlabeled Disentangling of GANs with Guided Siamese Networks""","['GAN', 'disentange', 'siamese networks', 'semantic']","""Disentangling underlying generative factors of a data distribution is important for interpretability and generalizable representations. In this paper, we introduce two novel disentangling methods. Our first method, Unlabeled Disentangling GAN (UD-GAN, unsupervised), decomposes the latent noise by generating similar/dissimilar image pairs and it learns a distance metric on these pairs with siamese networks and a contrastive loss. This pairwise approach provides consistent representations for similar data points. Our second method (UD-GAN-G, weakly supervised) modifies the UD-GAN with user-defined guidance functions, which restrict the information that goes into the siamese networks. This constraint helps UD-GAN-G to focus on the desired semantic variations in the data. We show that both our methods outperform existing unsupervised approaches in quantitative metrics that measure semantic accuracy of the learned representations. In addition, we illustrate that simple guidance functions we use in UD-GAN-G allow us to directly capture the desired variations in the data.""","""The paper received mixed reviews. It proposes a variant of Siamese network objective function, which is interesting. However, its unclear if the performance of the unguided method is much better than other baselines (e.g., InfoGAN). The guided version of the method seems to require much domain-specific knowledge and design of the feature function, which makes the paper difficult to apply to broader cases. """
1163,"""Bayesian Deep Convolutional Networks with Many Channels are Gaussian Processes""","['Deep Convolutional Neural Networks', 'Gaussian Processes', 'Bayesian']","""There is a previously identified equivalence between wide fully connected neural networks (FCNs) and Gaussian processes (GPs). This equivalence enables, for instance, test set predictions that would have resulted from a fully Bayesian, infinitely wide trained FCN to be computed without ever instantiating the FCN, but by instead evaluating the corresponding GP. In this work, we derive an analogous equivalence for multi-layer convolutional neural networks (CNNs) both with and without pooling layers, and achieve state of the art results on CIFAR10 for GPs without trainable kernels. We also introduce a Monte Carlo method to estimate the GP corresponding to a given neural network architecture, even in cases where the analytic form has too many terms to be computationally feasible. Surprisingly, in the absence of pooling layers, the GPs corresponding to CNNs with and without weight sharing are identical. As a consequence, translation equivariance, beneficial in finite channel CNNs trained with stochastic gradient descent (SGD), is guaranteed to play no role in the Bayesian treatment of the infinite channel limit - a qualitative difference between the two regimes that is not present in the FCN case. We confirm experimentally, that while in some scenarios the performance of SGD-trained finite CNNs approaches that of the corresponding GPs as the channel count increases, with careful tuning SGD-trained CNNs can significantly outperform their corresponding GPs, suggesting advantages from SGD training compared to fully Bayesian parameter estimation.""","""There has been a recent focus on proving the convergence of Bayesian fully connected networks to GPs. This work takes these ideas one step further, by proving the equivalence in the convolutional case. All reviewers and the AC are in agreement that this is interesting and impactful work. The nature of the topic is such that experimental evaluations and theoretical proofs are difficult to carry out in a convincing manner, however the authors have done a good job at it, especially after carefully taking into account the reviewers comments. """
1164,"""Learning to encode spatial relations from natural language""","['generative model', 'grounded language', 'scene understanding', 'natural language']","""Natural language processing has made significant inroads into learning the semantics of words through distributional approaches, however representations learnt via these methods fail to capture certain kinds of information implicit in the real world. In particular, spatial relations are encoded in a way that is inconsistent with human spatial reasoning and lacking invariance to viewpoint changes. We present a system capable of capturing the semantics of spatial relations such as behind, left of, etc from natural language. Our key contributions are a novel multi-modal objective based on generating images of scenes from their textual descriptions, and a new dataset on which to train it. We demonstrate that internal representations are robust to meaning preserving transformations of descriptions (paraphrase invariance), while viewpoint invariance is an emergent property of the system.""","""Strengths: Execution of paper well received. Results on new dataset. Convincing demonstration that the proposed approach learns good semantic representations. Weaknesses: Reviewers felt the positioning with prior work was not as strong as it could be. Reviewers wanted to have seen an ablation study. Contention: Some general agreement among both the one positive reviewer and negative reviewer that the representation of prior work is skewed. Consensus: With two 5s and one 6 the numerical average of 5.33 is representative of the aggregated consensus opinion which is that the work is just below threshold in its current form."""
1165,"""Deep Perm-Set Net: Learn to predict sets with unknown permutation and cardinality using deep neural networks""","['Set learning', 'Permutation invariant', 'Object detection', 'CAPTCHA test']","""Many real-world problems, e.g. object detection, have outputs that are naturally expressed as sets of entities. This creates a challenge for traditional deep neural networks which naturally deal with structured outputs such as vectors, matrices or tensors. We present a novel approach for learning to predict sets with unknown permutation and cardinality using deep neural networks. Specifically, in our formulation we incorporate the permutation as unobservable variable and estimate its distribution during the learning process using alternating optimization. We demonstrate the validity of this new formulation on two relevant vision problems: object detection, for which our formulation outperforms state-of-the-art detectors such as Faster R-CNN and YOLO, and a complex CAPTCHA test, where we observe that, surprisingly, our set based network acquired the ability of mimicking arithmetics without any rules being coded.""","""Strengths: The method extends [21], which proposes an unordered set prediction model for multi-class classification. The submission proposes a formulation to learn the distribution over unobservable permutation variables based on deep networks and uses a MAP estimator for inference. While the failure of NMS to detect overlapping objects is expected, the experiments showing that perm-set prediction handles them well is interesting and promising. Weaknesses: Reviewer 1: ""I find the paper still too scattered, trying to solve diverse problems with a hammer without properly motivating / analyzing key details of this hammer. So I keep my rating."" Reviewer 2: ""I'm glad that the authors are seeing good performance and seem to have an effective method for matching outputs to fixed predictions, however the quality of the paper is too poor for publication."" Points of contention: Although there was one reviewer who gave a high rating, they were not responsive in the rebuttal phase. The other two reviewers took into account the author responses, and a contributed comment by an unaffiliated reviewer, and both concluded that the paper still had serious issues. The main issues were: lack of clear methodology and poor clarity (AnonReviewer2), and poor organization and lack of motivation for modeling choices (AnonReviewer1)."""
1166,"""Zero-shot Dual Machine Translation""","['unsupervised', 'machine translation', 'dual learning', 'zero-shot']","""Neural Machine Translation (NMT) systems rely on large amounts of parallel data.This is a major challenge for low-resource languages. Building on recent work onunsupervised and semi-supervised methods, we present an approach that combineszero-shot and dual learning. The latter relies on reinforcement learning, to exploitthe duality of the machine translation task, and requires only monolingual datafor the target language pair. Experiments on the UN corpus show that a zero-shotdual system, trained on English-French and English-Spanish, outperforms by largemargins a standard NMT system in zero-shot translation performance on Spanish-French (both directions). We also evaluate onnewstest2014. These experimentsshow that the zero-shot dual method outperforms the LSTM-based unsupervisedNMT system proposed in (Lample et al., 2018b), on the enfr task, while onthe fren task it outperforms both the LSTM-based and the Transformers-basedunsupervised NMT systems.""","""This paper is essentially an application of dual learning to multilingual NMT. The results are reasonable. However, reviewers noted that the methodological novelty is minimal, and there are not a large number of new insights to be gained from the main experiments. Thus, I am not recommending the paper for acceptance at this time."""
1167,"""Diverse Machine Translation with a Single Multinomial Latent Variable""","['machine translation', 'latent variable models', 'diverse decoding']","""There are many ways to translate a sentence into another language. Explicit modeling of such uncertainty may enable better model fitting to the data and it may enable users to express a preference for how to translate a piece of content. Latent variable models are a natural way to represent uncertainty. Prior work investigated the use of multivariate continuous and discrete latent variables, but their interpretation and use for generating a diverse set of hypotheses have been elusive. In this work, we drastically simplify the model, using just a single multinomial latent variable. The resulting mixture of experts model can be trained efficiently via hard-EM and can generate a diverse set of hypothesis by parallel greedy decoding. We perform extensive experiments on three WMT benchmark datasets that have multiple human references, and we show that our model provides a better trade-off between quality and diversity of generations compared to all baseline methods.\footnote{Code to reproduce this work is available at: anonymized URL.}""","""+ a simple method + producing diverse translation is an important problem - technical contribution is limited / work is incremental - R1 finds writing not precise and claims not supported, also discussion of related work is considered weak by R3 - claims of modeling uncertainty are not well supported There is no consensus among reviewers. R4 provides detailed arguments why (at the very least) certain aspects of presentations are misleading (e.g., claiming that a uniform prior promotes diversity). R1 is also negative, his main concerns are limited contribution and he also questions the task (from their perspective producing diverse translation is not a valid task; I would disagree with this). R2 likes the paper and believes it is interesting, simple to use and the paper should be accepted. R3 is more lukewarm. """
1168,"""Inferring Reward Functions from Demonstrators with Unknown Biases""","['Inverse reinforcement learning', 'differentiable planning']","""Our goal is to infer reward functions from demonstrations. In order to infer the correct reward function, we must account for the systematic ways in which the demonstrator is suboptimal. Prior work in inverse reinforcement learning can account for specific, known biases, but cannot handle demonstrators with unknown biases. In this work, we explore the idea of learning the demonstrator's planning algorithm (including their unknown biases), along with their reward function. What makes this challenging is that any demonstration could be explained either by positing a term in the reward function, or by positing a particular systematic bias. We explore what assumptions are sufficient for avoiding this impossibility result: either access to tasks with known rewards which enable estimating the planner separately, or that the demonstrator is sufficiently close to optimal that this can serve as a regularizer. In our exploration with synthetic models of human biases, we find that it is possible to adapt to different biases and perform better than assuming a fixed model of the demonstrator, such as Boltzmann rationality.""","""The authors study an inverse reinforcement learning problem where the goal is to infer an underlying reward function from demonstration with bias. To achieve this, the authors learn the planners and the reward functions from demonstrations. As this is in general impossible, the authors consider two special cases in which either the reward function is observed on a subset of tasks or in which the observations are assumed to be close to optimal. They propose algorithms for both cases and evaluate these in basic experiments. The problem considered is important and challenging. One issue is that in order to make progress the authors need to make strong and restrictive assumptions (e.g., assumption 3, the well-suited inductive bias). It is not clear if the assumptions made are reasonable. Experimentally, it would be important to see how results change if the model for the planner changes and to evaluate what the inferred biases would be. Overall, there is consensus among the reviewers that the paper is interesting but not ready for publication. """
1169,"""Are Generative Classifiers More Robust to Adversarial Attacks?""","['generative models', 'adversarial attack', 'defence', 'detection', ""Bayes' rule""]","""There is a rising interest in studying the robustness of deep neural network classifiers against adversaries, with both advanced attack and defence techniques being actively developed. However, most recent work focuses on discriminative classifiers, which only model the conditional distribution of the labels given the inputs. In this paper, we propose and investigate the deep Bayes classifier, which improves classical naive Bayes with conditional deep generative models. We further develop detection methods for adversarial examples, which reject inputs with low likelihood under the generative model. Experimental results suggest that deep Bayes classifiers are more robust than deep discriminative classifiers, and that the proposed detection methods are effective against many recently proposed attacks.""","""Adversarial defense is a tricky subject, and the authors are to be commended for their novel approach to this problem. The reviewers all agree that there is promise in this approach. However, after reviewing the discussion, they have all come to the conclusion that the robustness of your generative model needs to be more thoroughly explored. Regarding gradient masking, there are other attacks like a manifold attack that use gradients that can be explored as well. Regarding SPSA, it would be helpful perhaps to also include other numerical gradient attacks to ensure that SPSA is stronger and working as intended. Essentially, the reviewers would all like to see a more streamlined version of this paper that removes any doubt about the efficacy of the generative approach. Once that is established, additional properties and features can be explored. Also note that for the purposes of these reviews and discussion, Schott et al. was considered as concurrent work and not prior work. """
1170,"""Recurrent Experience Replay in Distributed Reinforcement Learning""","['RNN', 'LSTM', 'experience replay', 'distributed training', 'reinforcement learning']","""Building on the recent successes of distributed training of RL agents, in this paper we investigate the training of RNN-based RL agents from distributed prioritized experience replay. We study the effects of parameter lag resulting in representational drift and recurrent state staleness and empirically derive an improved training strategy. Using a single network architecture and fixed set of hyper-parameters, the resulting agent, Recurrent Replay Distributed DQN, quadruples the previous state of the art on Atari-57, and matches the state of the art on DMLab-30. It is the first agent to exceed human-level performance in 52 of the 57 Atari games.""","""The paper proposes a new distributed DQN algorithm that combines recurrent neural networks with distributed prioritized replay memory. The authors systematically compare three types of initialization strategies for training the recurrent models. The thorough investigation is cited as a valuable contribution by all reviewers, with reviewer 1 noting that the study would be of interest to ""anyone using recurrent networks on RL tasks"". Empirical results on Atari and DMLab are impressive. The reviewers noted several weaknesses in their original reviews. These included issues of clarity, a need for more detailed ablation studies, and need to more carefully document the empirical setup. A further question was raised on whether the empirical results could be complemented with theoretical or conceptual insights. The authors carefully addressed all concerns raised during the reviewing and rebuttal period. They took exceptional care to clarify their writing, document experiment details, and ran a large set of additional experiments as suggested by the reviewers. The AC feels that the review period for the paper was particularly productive and would like to thank the reviewers and authors. The reviewers and AC agree that the paper makes a significant contribution to the field and should be accepted."""
1171,"""Deep Denoising: Rate-Optimal Recovery of Structured Signals with a Deep Prior""","['non-convex optimization', 'denoising', 'generative neural network']","""Deep neural networks provide state-of-the-art performance for image denoising, where the goal is to recover a near noise-free image from a noisy image. The underlying principle is that neural networks trained on large datasets have empirically been shown to be able to generate natural images well from a low-dimensional latent representation of the image. Given such a generator network, or prior, a noisy image can be denoised by finding the closest image in the range of the prior. However, there is little theory to justify this success, let alone to predict the denoising performance as a function of the networks parameters. In this paper we consider the problem of denoising an image from additive Gaussian noise, assuming the image is well described by a deep neural network with ReLu activations functions, mapping a k-dimensional latent space to an n-dimensional image. We state and analyze a simple gradient-descent-like iterative algorithm that minimizes a non-convex loss function, and provably removes a fraction of (1 - O(k/n)) of the noise energy. We also demonstrate in numerical experiments that this denoising performance is, indeed, achieved by generative priors learned from data.""","""The paper analyzes the interesting problem of image denoising with neural networks by imposing simplifying assumptions on the Gaussianity and independence of the prior. A bound is established from the analysis of (Hand & Voroninksi, 2018) that can be algorithmically achieved through a small tweak to gradient descent. Unfortunately, the contribution of this paper is incremental given the recent works of (Hand & Voroninksi, 2018) and (Bora et al., 2017); an opinion the reviewers unanimously shared. Reviewer opinion differed on whether they found the overall contribution to be barely acceptable or simply insufficient. No reviewer detected a major advance, and there seems to be a question of whether the achievement is significant given the strength of the assumptions required to achieve the modest additions. After scrutiny, the main theoretical contributions of the paper appear to be a bit overstated. For example, the bound in Theorem 1 is quite weak: it does not establish convergence to a global minimizer (even under the strong assumptions given), but only that Algorithm 1 eventually remains in a neighborhood of the global minimizer. It is true that this neighborhood can be made arbitrarily small by increasing the strength of the assumptions made on epsilon and omega, but epsilon remains a constant with respect to iteration count. The subsequent claim that the algorithm achieves a denoising rate of sigma^2 k/n is not an accurate interpretation of Theorem 1, given that this claim would require require (at the very least) that epsilon can be made arbitrarily small, which it cannot be. More precision is required in stating supportable conclusions from the given results. The algorithmic motivation itself is rather weak, in the sense that this paper only provides an anecdotal demonstration that there are no spurious critical points beyond the negation of the global minimizer---the theoretical support for this claim already resides in (Hand & Voroninski, 2018). The provenance of such a central observation was not made sufficiently clear in the paper nor in the discussion. An additional quibble about the experimental evaluation is that it does not compare to plain gradient descent (or other baseline optimization techniques), which the authors observe almost always works in the scenario considered. It seems that the ""negation tweak"" embedded in Algorithm 1 has no real impact on the experimental results, raising the question of whether the contributions do indeed have any practical import. The descriptions offered in the current paper suggest that a serious algorithmic advantage has yet to be demonstrated in any real experiment. The paper requires a far better evaluation of Algorithm 1 in comparison to standard baseline optimizers, to support the case that the proposed algorithmic tweak has practical significance. This paper remained in a weak borderline position after the review and discussion period. In the end, this was a very difficult decision to make, but I think the paper would benefit from further strengthening before it can constitute a solid publication."""
1172,"""The loss landscape of overparameterized neural networks""",[],"""We explore some mathematical features of the loss landscape of overparameterized neural networks. A priori one might imagine that the loss function looks like a typical function from pseudo-formula to pseudo-formula - in particular, nonconvex, with discrete global minima. In this paper, we prove that in at least one important way, the loss function of an overparameterized neural network does not look like a typical function. If a neural net has pseudo-formula parameters and is trained on pseudo-formula data points, with pseudo-formula , we show that the locus pseudo-formula of global minima of pseudo-formula is usually not discrete, but rather an pseudo-formula dimensional submanifold of pseudo-formula . In practice, neural nets commonly have orders of magnitude more parameters than data points, so this observation implies that pseudo-formula is typically a very high-dimensional subset of pseudo-formula . ""","""The paper proves that the locus of the global minima of an over-parameterized neural nets objective forms a low-dimensional manifold. The reviewers and AC note the following potential weaknesses: --- it's not clear why the proved result is significant: it neither implies the SGD can find a global minimum, nor that the found solution can generalize. (Very likely, most of the global minima on the manifold cannot generalize.) --- the results seem very intuitive and are a straightforward application of certain topological theorem. """
1173,"""Understanding & Generalizing AlphaGo Zero""","['reinforcement learning', 'AlphaGo Zero']","""AlphaGo Zero (AGZ) introduced a new {\em tabula rasa} reinforcement learning algorithm that has achieved superhuman performance in the games of Go, Chess, and Shogi with no prior knowledge other than the rules of the game. This success naturally begs the question whether it is possible to develop similar high-performance reinforcement learning algorithms for generic sequential decision-making problems (beyond two-player games), using only the constraints of the environment as the ``rules.'' To address this challenge, we start by taking steps towards developing a formal understanding of AGZ. AGZ includes two key innovations: (1) it learns a policy (represented as a neural network) using {\em supervised learning} with cross-entropy loss from samples generated via Monte-Carlo Tree Search (MCTS); (2) it uses {\em self-play} to learn without training data. We argue that the self-play in AGZ corresponds to learning a Nash equilibrium for the two-player game; and the supervised learning with MCTS is attempting to learn the policy corresponding to the Nash equilibrium, by establishing a novel bound on the difference between the expected return achieved by two policies in terms of the expected KL divergence (cross-entropy) of their induced distributions. To extend AGZ to generic sequential decision-making problems, we introduce a {\em robust MDP} framework, in which the agent and nature effectively play a zero-sum game: the agent aims to take actions to maximize reward while nature seeks state transitions, subject to the constraints of that environment, that minimize the agent's reward. For a challenging network scheduling domain, we find that AGZ within the robust MDP framework provides near-optimal performance, matching one of the best known scheduling policies that has taken the networking community three decades of intensive research to develop. ""","""This work examines the AlphaGo Zero algorithm, a self-play reinforcement learning algorithm that has been shown to learn policies with superhuman performance on 2 player perfect information games. The main result of the paper is that the policy learned by AGZ corresponds to a Nash equilibrium, that and that the cross-entropy minimization in the supervised learning-inspired part of the algorithm converges to this Nash equillibrium, proves a bound on the expected returns of two policies under the and introduces a ""robust MDP"" view of a 2 player zero-sum game played between the agent and nature. R3 found the paper well-structured and the results presented therein interesting. R2 complained of overly heavy notation and questioned the applicability of the results, as well as the utility of the robust MDP perspective (though did raise their score following revisions). The most detailed critique came from R1, who suggested that the bound on the convergence of returns of two policies as the KL divergence between their induced distributions decreases is unsurprising, that using it to argue for AGZ's convergence to the optimal policy ignores the effects introduced by the suboptimality of the MCTS policy (while really interesting part being understanding how AGZ deals with, and whether or not it closes, this gap), and that the ""robust MDP"" view is less novel than the authors claim based on the known relationships between 2 player zero-sum games and minimax robust control. I find R1's complaints, in particular with respect to ""robust MDPs"" (a criticism which went completely unaddressed by the authors in their rebuttal), convincing enough that I would narrowly recommend rejection at this time, while also agreeing with R3 that this is an interesting subject and that the results within could serve as the bedrock for a stronger future paper."""
1174,"""Dynamic Sparse Graph for Efficient Deep Learning""","['Sparsity', 'compression', 'training', 'acceleration']","""We propose to execute deep neural networks (DNNs) with dynamic and sparse graph (DSG) structure for compressive memory and accelerative execution during both training and inference. The great success of DNNs motivates the pursuing of lightweight models for the deployment onto embedded devices. However, most of the previous studies optimize for inference while neglect training or even complicate it. Training is far more intractable, since (i) the neurons dominate the memory cost rather than the weights in inference; (ii) the dynamic activation makes previous sparse acceleration via one-off optimization on fixed weight invalid; (iii) batch normalization (BN) is critical for maintaining accuracy while its activation reorganization damages the sparsity. To address these issues, DSG activates only a small amount of neurons with high selectivity at each iteration via a dimensionreduction search and obtains the BN compatibility via a double-mask selection. Experiments show significant memory saving (1.7-4.5x) and operation reduction (2.3-4.4x) with little accuracy loss on various benchmarks.""","""This paper proposes a novel approach for network pruning in both training and inference. This paper received a consensus of acceptance. Compared with previous work that focus and model compression on training, this paper saves memory and accelerates both training and inference. It is activation, rather than weight that dominates the training memory. Reviewer1 posed a valid concern about the efficient implementation on GPUs, and authors agreed that practical speedup on GPU is difficult. It'll be great if the authors can give practical insights on how to achieve real speedup in the final draft. """
1175,"""Learning to Augment Influential Data""","['data augmentation', 'influence function', 'generative adversarial network']","""Data augmentation is a technique to reduce overfitting and to improve generalization by increasing the number of labeled data samples by performing label preserving transformations; however, it is currently conducted in a trial and error manner. A composition of predefined transformations, such as rotation, scaling and cropping, is performed on training samples, and its effect on performance over test samples can only be empirically evaluated and cannot be predicted. This paper considers an influence function which predicts how generalization is affected by a particular augmented training sample in terms of validation loss. The influence function provides an approximation of the change in validation loss without comparing the performance which includes and excludes the sample in the training process. A differentiable augmentation model that generalizes the conventional composition of predefined transformations is also proposed. The differentiable augmentation model and reformulation of the influence function allow the parameters of the augmented model to be directly updated by backpropagation to minimize the validation loss. The experimental results show that the proposed method provides better generalization over conventional data augmentation methods.""","""This paper proposes and end-to-end trainable architecture for data augmentation, by defining a parametric model for data augmentation (using spatial transformers and GANs) and optimizing validation classification error through the notion of influence functions. Experiments are reported on MNIST and CIfar-10. This is a borderline submission. Reviewers found the theoretical framework and problem setup to be solid and promising, but were also concerned about the experimental setup and the lack of clarity in the manuscript. In particular, one would like to evaluate this model against similar baselines (e.g. Ratner et al) on a large-scale classification problem. The AC, after taking these comments into account and making his/her own assessment, recommends rejection at this time, encouraging the authors to address the above comments and resubmit this promising work in the next conference cycle. """
1176,"""A Statistical Approach to Assessing Neural Network Robustness""","['neural network verification', 'multi-level splitting', 'formal verification']","""We present a new approach to assessing the robustness of neural networks based on estimating the proportion of inputs for which a property is violated. Specifically, we estimate the probability of the event that the property is violated under an input model. Our approach critically varies from the formal verification framework in that when the property can be violated, it provides an informative notion of how robust the network is, rather than just the conventional assertion that the network is not verifiable. Furthermore, it provides an ability to scale to larger networks than formal verification approaches. Though the framework still provides a formal guarantee of satisfiability whenever it successfully finds one or more violations, these advantages do come at the cost of only providing a statistical estimate of unsatisfiability whenever no violation is found. Key to the practical success of our approach is an adaptation of multi-level splitting, a Monte Carlo approach for estimating the probability of rare events, to our statistical robustness framework. We demonstrate that our approach is able to emulate formal verification procedures on benchmark problems, while scaling to larger networks and providing reliable additional information in the form of accurate estimates of the violation probability.""","""* Strengths The paper addresses an important topic: how to bound the probability that a given bad event occurs for a neural network under some distribution of inputs. This could be relevant, for instance, in autonomous robotics settings where there is some environment model and we would like to bound the probability of an adverse outcome (e.g. for an autonomous aircraft, the time to crash under a given turbulence model). The desired failure probabilities are often low enough that direct Monte Carlo simulation is too expensive. The present work provides some preliminary but meaningful progress towards better methods of estimating such low-probability events, and provides some evidence that the methods can scale up to larger networks. It is well-written and of high technical quality. * Weaknesses In the initial submission, one reviewer was concerned that the term verification was misleading, as the methods had no formal guarantees that the estimated probability was correct. The authors proposed to revise the paper to remove reference to verification in the title and the text, and afterwards all reviewers agreed the work should be accepted. The paper also may slightly overstate the generality of the method. For instance, the claim that this can be used to show that adversarial examples do not exist is probably wrong---adversarial examples often occupy a negligibly small portion of the input space. There was also concern that most comparisons were limited to naive Monte Carlo. * Discussion While there was initial disagreement among reviewers, after the discussion all reviewers agree the paper should be accepted. However, we remind the authors to implement the changes promised during the discussion period."""
1177,"""Large-Scale Study of Curiosity-Driven Learning""","['exploration', 'curiosity', 'intrinsic reward', 'no extrinsic reward', 'unsupervised', 'no-reward', 'skills']","""Reinforcement learning algorithms rely on carefully engineered rewards from the environment that are extrinsic to the agent. However, annotating each environment with hand-designed, dense rewards is difficult and not scalable, motivating the need for developing reward functions that are intrinsic to the agent. Curiosity is such intrinsic reward function which uses prediction error as a reward signal. In this paper: (a) We perform the first large-scale study of purely curiosity-driven learning, i.e. {\em without any extrinsic rewards}, across pseudo-formula standard benchmark environments, including the Atari game suite. Our results show surprisingly good performance as well as a high degree of alignment between the intrinsic curiosity objective and the hand-designed extrinsic rewards of many games. (b) We investigate the effect of using different feature spaces for computing prediction error and show that random features are sufficient for many popular RL game benchmarks, but learned features appear to generalize better (e.g. to novel game levels in Super Mario Bros.). (c) We demonstrate limitations of the prediction-based rewards in stochastic setups. Game-play videos and code are at pseudo-url.""","""The authors have extended previous publications on curiosity driven, intrinsically motivated RL with this broad empirical study on the effectiveness of the curiosity algorithm on many game environments, the merits of different feature sets, and limitations of the approach. The paper is well-written and should be of interest to the community. The experiments are well conceived and seem to validate the general effectiveness of curiosity. However, the paper does not actually have any novel contribution compared against prior work, and there are no great insights or takeaways from the empirical study. Therefore, the reviewers were somewhat divided on how confident they were that the paper should be accepted. Overall, the AC agrees that it is a valuable paper that should be accepted even though it does not deliver any algorithmic novelty."""
1178,"""Deep Graph Infomax""","['Unsupervised Learning', 'Graph Neural Networks', 'Graph Convolutions', 'Mutual Information', 'Infomax', 'Deep Learning']","""We present Deep Graph Infomax (DGI), a general approach for learning node representations within graph-structured data in an unsupervised manner. DGI relies on maximizing mutual information between patch representations and corresponding high-level summaries of graphs---both derived using established graph convolutional network architectures. The learnt patch representations summarize subgraphs centered around nodes of interest, and can thus be reused for downstream node-wise learning tasks. In contrast to most prior approaches to unsupervised learning with GCNs, DGI does not rely on random walk objectives, and is readily applicable to both transductive and inductive learning setups. We demonstrate competitive performance on a variety of node classification benchmarks, which at times even exceeds the performance of supervised learning.""","""Because of strong support from two of the reviewers I am recommending accepting this paper. However, I believe reviewer 1's concerns should be taken seriously. Although I disagree with the reviewer that a general ""framework"" method is a bad thing, I agree with them that additional experiments would be valuable."""
1179,"""COCO-GAN: Conditional Coordinate Generative Adversarial Network""",[],"""Recent advancements on Generative Adversarial Network (GAN) have inspired a wide range of works that generate synthetic images. However, the current processes have to generate an entire image at once, and therefore resolutions are limited by memory or computational constraints. In this work, we propose COnditional COordinate GAN (COCO-GAN), which generates a specific patch of an image conditioned on a spatial position rather than the entire image at a time. The generated patches are later combined together to form a globally coherent full-image. With this process, we show that the generated image can achieve competitive quality to state-of-the-arts and the generated patches are locally smooth between consecutive neighbors. One direct implication of the COCO-GAN is that it can be applied onto any coordinate systems including the cylindrical systems which makes it feasible for generating panorama images. The fact that the patch generation process is independent to each other inspires a wide range of new applications: firstly, ""Patch-Inspired Image Generation"" enables us to generate the entire image based on a single patch. Secondly, ""Partial-Scene Generation"" allows us to generate images within a customized target region. Finally, thanks to COCO-GAN's patch generation and massive parallelism, which enables combining patches for generating a full-image with higher resolution than state-of-the-arts.""","""The paper introduces a GAN architecture for generating small patches of an image and subsequently combining them. Following the rebuttal and discussion, reviewers still rate the paper as marginally above or below the acceptance threshold. In response to updates, AnonReviewer3 comments that ""ablation experiments do make the paper stronger"" but it ""still lacks convincing experiments for its main motivating use case: generating outputs at a resolution that won't fit in memory within a single forward pass"". AnonReviewer2 points to the major shortcoming that ""throughout the exposition it is never really clear why COCO-GAN is a good idea beyond the fact that it somehow works. I was missing a concrete use case where COCO-GAN performs much better."" Though authors provide additional experiments and reference high-resolution output during the discussion phase, they caution that these results are preliminary and could likely benefit from more time/work devoted to training. On balance, the AC agrees with the reviewers that the paper contains some interesting ideas, but also believes that experimental validation simply needs more work, and as a result the paper does not meet the bar for acceptance. """
1180,"""Neural Message Passing for Multi-Label Classification""","['Multi-label Classification', 'Graph Neural Networks', 'Attention', 'Graph Attention']","""Multi-label classification (MLC) is the task of assigning a set of target labels for a given sample. Modeling the combinatorial label interactions in MLC has been a long-haul challenge. Recurrent neural network (RNN) based encoder-decoder models have shown state-of-the-art performance for solving MLC. However, the sequential nature of modeling label dependencies through an RNN limits its ability in parallel computation, predicting dense labels, and providing interpretable results. In this paper, we propose Message Passing Encoder-Decoder (MPED) Networks, aiming to provide fast, accurate, and interpretable MLC. MPED networks model the joint prediction of labels by replacing all RNNs in the encoder-decoder architecture with message passing mechanisms and dispense with autoregressive inference entirely. The proposed models are simple, fast, accurate, interpretable, and structure-agnostic (can be used on known or unknown structured data). Experiments on seven real-world MLC datasets show the proposed models outperform autoregressive RNN models across five different metrics with a significant speedup during training and testing time.""","""The reviewers highlighted aspects of the work that were interesting, particularly on the chosen topic of multi-label output of graph neural networks. However, no reviewer was willing to champion the paper, and in aggregate all reviewers trend towards rejection."""
1181,"""Riemannian TransE: Multi-relational Graph Embedding in Non-Euclidean Space""","['Riemannian TransE', 'graph embedding', 'multi-relational graph', 'Riemannian manifold', 'TransE', 'hyperbolic space', 'sphere', 'knowledge base']","""Multi-relational graph embedding which aims at achieving effective representations with reduced low-dimensional parameters, has been widely used in knowledge base completion. Although knowledge base data usually contains tree-like or cyclic structure, none of existing approaches can embed these data into a compatible space that in line with the structure. To overcome this problem, a novel framework, called Riemannian TransE, is proposed in this paper to embed the entities in a Riemannian manifold. Riemannian TransE models each relation as a move to a point and defines specific novel distance dissimilarity for each relation, so that all the relations are naturally embedded in correspondence to the structure of data. Experiments on several knowledge base completion tasks have shown that, based on an appropriate choice of manifold, Riemannian TransE achieves good performance even with a significantly reduced parameters.""","""This paper proposes a generalization of the translation-style embedding approaches for link prediction to Riemannian manifolds. The reviewers feel this is an important contribution to the recent work on embedding graphs into non-Euclidean spaces, especially since this work focuses on multi-relational links, thus supporting knowledge graph completion. The results on WN11 and FB13 are also promising. The reviewers and AC note the following potential weaknesses: (1) the primary concern is the low performance on the benchmarks, especially WN18 and FB15k, and not using the appropriate versions (WN18-RR and FB15k-237), (2) use of hyperbolic embedding for an entity shared across all relations, and (3) lack of discussion/visualization of the learned geometry. During the discussion phase, the authors clarified reviewer 1's concern regarding the difference in performance between HolE and ComplEx, along with providing a revision that addressed some of the clarity issues raised by reviewer 3. The authors also justified the lower performance due to (1) they are focusing on low-dimensionality setting, and (2) not all datasets will fit the space of the proposed model (like FB15k). However, reviewers 2 and 3 still maintain that the results provide insufficient evidence for the need for Riemannian spaces over Euclidean ones, especially for larger, and more realistic, knowledge graphs. The reviewers and the AC agree that the paper should not be accepted in the current state. """
1182,"""NATTACK: A STRONG AND UNIVERSAL GAUSSIAN BLACK-BOX ADVERSARIAL ATTACK""","['adversarial attack', 'black-box', 'evolutional strategy', 'policy gradient']","""Recent works find that DNNs are vulnerable to adversarial examples, whose changes from the benign ones are imperceptible and yet lead DNNs to make wrong predictions. One can find various adversarial examples for the same input to a DNN using different attack methods. In other words, there is a population of adversarial examples, instead of only one, for any input to a DNN. By explicitly modeling this adversarial population with a Gaussian distribution, we propose a new black-box attack called NATTACK. The adversarial attack is hence formalized as an optimization problem, which searches the mean of the Gaussian under the guidance of increasing the target DNN's prediction error. NATTACK achieves 100% attack success rate on six out of eleven recently published defense methods (and greater than 90% for four), all using the same algorithm. Such results are on par with or better than powerful state-of-the-art white-box attacks. While the white-box attacks are often model-specific or defense-specific, the proposed black-box NATTACK is universally applicable to different defenses. ""","""Although one review is favorable, it does not make a strong enough case for accepting this paper. Thus there is not sufficient support in the reviews to accept this paper. I am recommending rejecting this submission for multiple reasons. Given that this is a ""black box"" attack formalized as an optimization problem, the method must be compared to other approaches in the large field of derivative-free optimization. There are many techniques including: Bayesian optimization, (other) evolutionary algorithms, simulated annealing, Nelder-Mead, coordinate descent, etc. Since the method of the paper does not use anything about the structure of the problem it can be applied to other derivative-free optimization problems that had the same search constraint. However, the paper does not provide evidence that it has advanced the state of the art in derivative-free optimization. The method the paper describes does not need a new name and is an obvious variation of existing evolutionary algorithms. Someone facing the same problem could easily reinvent the exact method of the paper without reading it and this limits the value of the contribution. Finally, this paper amounts to breaking already broken defenses, which is not an activity of high value to the community at this stage and also limits the contribution of this work. """
1183,"""From Language to Goals: Inverse Reinforcement Learning for Vision-Based Instruction Following""","['inverse reinforcement learning', 'language grounding', 'instruction following', 'language-based learning']","""Reinforcement learning is a promising framework for solving control problems, but its use in practical situations is hampered by the fact that reward functions are often difficult to engineer. Specifying goals and tasks for autonomous machines, such as robots, is a significant challenge: conventionally, reward functions and goal states have been used to communicate objectives. But people can communicate objectives to each other simply by describing or demonstrating them. How can we build learning algorithms that will allow us to tell machines what we want them to do? In this work, we investigate the problem of grounding language commands as reward functions using inverse reinforcement learning, and argue that language-conditioned rewards are more transferable than language-conditioned policies to new environments. We propose language-conditioned reward learning (LC-RL), which grounds language commands as a reward function represented by a deep neural network. We demonstrate that our model learns rewards that transfer to novel tasks and environments on realistic, high-dimensional visual environments with natural language commands, whereas directly learning a language-conditioned policy leads to poor performance.""","""This paper generated a lot of discussion (not all of it visible to the authors or the public). R1 initially requested reasonable comparisons, but after the authors provided a response (and new results), R1 continued to recommend rejecting the paper simply because they personally did not find the manuscript insightful. Despite several requests for clarification, we could not converge on a specific problem with the manuscript. Ungrounded gut feelings are not grounds for rejection. After an extensive discussion, R2 and R3 both recommend accepting the paper and the AC agrees. Paper makes interesting contributions and will be a welcome addition to the literature. """
1184,"""Bias-Reduced Uncertainty Estimation for Deep Neural Classifiers""","['Uncertainty estimation', 'Deep learning']","""We consider the problem of uncertainty estimation in the context of (non-Bayesian) deep neural classification. In this context, all known methods are based on extracting uncertainty signals from a trained network optimized to solve the classification problem at hand. We demonstrate that such techniques tend to introduce biased estimates for instances whose predictions are supposed to be highly confident. We argue that this deficiency is an artifact of the dynamics of training with SGD-like optimizers, and it has some properties similar to overfitting. Based on this observation, we develop an uncertainty estimation algorithm that selectively estimates the uncertainty of highly confident points, using earlier snapshots of the trained model, before their estimates are jittered (and way before they are ready for actual classification). We present extensive experiments indicating that the proposed algorithm provides uncertainty estimates that are consistently better than all known methods.""","""The paper proposes an improved method for uncertainty estimation in deep neural networks. Reviewer 2 and AC note that the paper is a bit isolated in terms of comparing the literature. However, as all of reviewers and AC found, the paper is well written and the proposed idea is clearly new/interesting."""
1185,"""Quaternion Recurrent Neural Networks""","['Quaternion recurrent neural networks', 'quaternion numbers', 'recurrent neural networks', 'speech recognition']","""Recurrent neural networks (RNNs) are powerful architectures to model sequential data, due to their capability to learn short and long-term dependencies between the basic elements of a sequence. Nonetheless, popular tasks such as speech or images recognition, involve multi-dimensional input features that are characterized by strong internal dependencies between the dimensions of the input vector. We propose a novel quaternion recurrent neural network (QRNN), alongside with a quaternion long-short term memory neural network (QLSTM), that take into account both the external relations and these internal structural dependencies with the quaternion algebra. Similarly to capsules, quaternions allow the QRNN to code internal dependencies by composing and processing multidimensional features as single entities, while the recurrent operation reveals correlations between the elements composing the sequence. We show that both QRNN and QLSTM achieve better performances than RNN and LSTM in a realistic application of automatic speech recognition. Finally, we show that QRNN and QLSTM reduce by a maximum factor of 3.3x the number of free parameters needed, compared to real-valued RNNs and LSTMs to reach better results, leading to a more compact representation of the relevant information.""","""The authors derive and experiment with quaternion-based recurrent neural networks, and demonstrate their effectiveness on speech recognition tasks (TIMIT and WSJ), where the authors demonstrate that the proposed models can achieve the same accuracy with fewer parameters than conventional models. The reviewers were unanimous in recommending that the paper be accepted."""
1186,"""Characterizing Attacks on Deep Reinforcement Learning""",[],"""Deep Reinforcement learning (DRL) has achieved great success in various applications, such as playing computer games and controlling robotic manipulation. However, recent studies show that machine learning models are vulnerable to adversarial examples, which are carefully crafted instances that aim to mislead learning models to make arbitrarily incorrect prediction, and raised severe security concerns. DRL has been attacked by adding perturbation to each observed frame. However, such observation based attacks are not quite realistic considering that it would be hard for adversaries to directly manipulate pixel values in practice. Therefore, we propose to understand the vulnerabilities of DRL from various perspectives and provide a throughout taxonomy of adversarial perturbation against DRL, and we conduct the first experiments on unexplored parts of this taxonomy. In addition to current observation based attacks against DRL, we propose attacks based on the actions and environment dynamics. Among these experiments, we introduce a novel sequence-based attack to attack a sequence of frames for real-time scenarios such as autonomous driving, and the first targeted attack that perturbs environment dynamics to let the agent fail in a specific way. We show empirically that our sequence-based attack can generate effective perturbations in a blackbox setting in real time with a small number of queries, independent of episode length. We conduct extensive experiments to compare the effectiveness of different attacks with several baseline attack methods in several game playing, robotics control, and autonomous driving environments.""","""The authors have delivered an extensive examination of deep RL attacks, placing them within a taxonomy, proposing new attacks, and giving empirical evidence to compare the effectiveness of the attacks. The reviewers and AC appreciate the broad effort, comprising 14 different attacks, and the well-written taxonomic discussion. However, the reviewers were concerned that the paper had significant problems with clarity of technical presentation and that the attacks were not well grounded in any sort of real world scenario. Although the authors addressed many concerns with their revision and rebuttal, the reviewers were not convinced. The AC believes that R1 ought to have increased their score given their comments and the resulting rebuttal, but the paper remains a borderline reject even with a corrected R1 score."""
1187,"""Generative Adversarial Networks for Extreme Learned Image Compression""","['Learned compression', 'generative adversarial networks', 'extreme compression']","""We propose a framework for extreme learned image compression based on Generative Adversarial Networks (GANs), obtaining visually pleasing images at significantly lower bitrates than previous methods. This is made possible through our GAN formulation of learned compression combined with a generator/decoder which operates on the full-resolution image and is trained in combination with a multi-scale discriminator. Additionally, if a semantic label map of the original image is available, our method can fully synthesize unimportant regions in the decoded image such as streets and trees from the label map, therefore only requiring the storage of the preserved region and the semantic label map. A user study confirms that for low bitrates, our approach is preferred to state-of-the-art methods, even when they use more than double the bits.""","""This paper proposes a GAN-based framework for image compression. The reviewers and AC note a critical limitation on novelty of the paper i.e., such a conditional GAN framework is now standard. The authors mentioned that they apply GAN for extreme compression for the first time in the literature, but this is not enough to justify the novelty issue. AC thinks the proposed method has potential and is interesting, but decided that the authors need new ideas to publish the work. """
1188,"""Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networks""","['natural image model', 'image prior', 'under-determined neural networks', 'untrained network', 'non-convolutional network', 'denoising', 'inverse problem']","""Deep neural networks, in particular convolutional neural networks, have become highly effective tools for compressing images and solving inverse problems including denoising, inpainting, and reconstruction from few and noisy measurements. This success can be attributed in part to their ability to represent and generate natural images well. Contrary to classical tools such as wavelets, image-generating deep neural networks have a large number of parameters---typically a multiple of their output dimension---and need to be trained on large datasets. In this paper, we propose an untrained simple image model, called the deep decoder, which is a deep neural network that can generate natural images from very few weight parameters. The deep decoder has a simple architecture with no convolutions and fewer weight parameters than the output dimensionality. This underparameterization enables the deep decoder to compress images into a concise set of network weights, which we show is on par with wavelet-based thresholding. Further, underparameterization provides a barrier to overfitting, allowing the deep decoder to have state-of-the-art performance for denoising. The deep decoder is simple in the sense that each layer has an identical structure that consists of only one upsampling unit, pixel-wise linear combination of channels, ReLU activation, and channelwise normalization. This simplicity makes the network amenable to theoretical analysis, and it sheds light on the aspects of neural networks that enable them to form effective signal representations.""","""In this work, the authors propose a simple, under parameterized network architecture which can fit natural images well, when fed with a fixed random input signal. This allows the model to be used for a number of tasks without requiring that the model be trained on a dataset. Further, unlike a recently proposed related method (DIP; [Ulyanov et al., 18]), the method does not require regularization such as early-stopping as with DIP. The reviewers noted the simplicity and experimental validation, and were unanimous in recommending acceptance. """
1189,"""The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks""","['Neural networks', 'sparsity', 'pruning', 'compression', 'performance', 'architecture search']","""Neural network pruning techniques can reduce the parameter counts of trained networks by over 90%, decreasing storage requirements and improving computational performance of inference without compromising accuracy. However, contemporary experience is that the sparse architectures produced by pruning are difficult to train from the start, which would similarly improve training performance. We find that a standard pruning technique naturally uncovers subnetworks whose initializations made them capable of training effectively. Based on these results, we articulate the ""lottery ticket hypothesis:"" dense, randomly-initialized, feed-forward networks contain subnetworks (""winning tickets"") that - when trained in isolation - reach test accuracy comparable to the original network in a similar number of iterations. The winning tickets we find have won the initialization lottery: their connections have initial weights that make training particularly effective. We present an algorithm to identify winning tickets and a series of experiments that support the lottery ticket hypothesis and the importance of these fortuitous initializations. We consistently find winning tickets that are less than 10-20% of the size of several fully-connected and convolutional feed-forward architectures for MNIST and CIFAR10. Above this size, the winning tickets that we find learn faster than the original network and reach higher test accuracy.""","""The authors posit and investigate a hypothesis -- the lottery ticket hypothesis -- which aims to explain why overparameterized neural networks are easier to train than their sparse counterparts. Under this hypothesis, randomly initialized dense networks are easier to train because they contain a larger number of winning tickets. This paper received very favorable reviews, though there were some notable points of concern. The reviewers and the AC appreciated the detailed and careful experimentation and analysis. However, there were a couple of points of concern raised by the reviewers: 1) the lack of experiments conducted on large-scale tasks and models, and 2) the lack of a clear application of the idea beyond what has been proposed previously. Overall, this is a very interesting paper with convincing experimental validation and as such the AC is happy to accept the work."""
1190,"""Caveats for information bottleneck in deterministic scenarios""","['information bottleneck', 'supervised learning', 'deep learning', 'information theory']","""Information bottleneck (IB) is a method for extracting information from one random variable X that is relevant for predicting another random variable Y. To do so, IB identifies an intermediate ""bottleneck"" variable T that has low mutual information I(X;T) and high mutual information I(Y;T). The ""IB curve"" characterizes the set of bottleneck variables that achieve maximal I(Y;T) for a given I(X;T), and is typically explored by maximizing the ""IB Lagrangian"", I(Y;T) - I(X;T). In some cases, Y is a deterministic function of X, including many classification problems in supervised learning where the output class Y is a deterministic function of the input X. We demonstrate three caveats when using IB in any situation where Y is a deterministic function of X: (1) the IB curve cannot be recovered by maximizing the IB Lagrangian for different values of ; (2) there are ""uninteresting"" trivial solutions at all points of the IB curve; and (3) for multi-layer classifiers that achieve low prediction error, different layers cannot exhibit a strict trade-off between compression and prediction, contrary to a recent proposal. We also show that when Y is a small perturbation away from being a deterministic function of X, these three caveats arise in an approximate way. To address problem (1), we propose a functional that, unlike the IB Lagrangian, can recover the IB curve in all cases. We demonstrate the three caveats on the MNIST dataset.""","""This paper considers the information bottleneck Lagrangian as a tool for studying deep networks in the common case of supervised learning (predicting label Y from features X) with a deterministic model, and identifies a number of troublesome issues. (1) The information bottleneck curve cannot be recovered by optimizing the Lagrangian for different values of because in the deterministic case, the IB curve is piecewise linear, not strictly concave. (2) Uninteresting representations can lie on the IB curve, so information bottleneck optimality does not imply that a representation is useful. (3) In a multilayer model with a low probability of error, the only tradeoff that successive layers can make between compression and prediction is that deeper layers may compress more. Experiments on MNIST illustrate these issues, and supplementary material shows that these issues also apply to the deterministic information bottleneck and to stochastic models that are nearly deterministic. There was a substantial degree of disagreement between the reviewers of this paper. One reviewer (R3) suggested that all the conclusions of the paper are the consequence of P(X,Y) being degenerate. The authors responded to this criticism in their response and revision quite effectively, in the opinion of the AC. Because R3 failed to participate in the discussion, this review has been discounted in the final decision. The other two reviewers were considerably more positive about the paper, with one (R1) having basically no criticisms and the other (R2) expression some doubts about the novelty of the observations being made in the paper and their importance for practical machine learning scenarios. Following the revision and discussion, R2 expressed general satisfaction with the paper, so the AC is recommending acceptance. The AC thinks that the final paper would be clearer if the authors were to carefully distinguish between ground-truth labels used in training and the labels estimated by the model for a given input. At the moment, the symbol Y appears to be overloaded, standing for both. Perhaps the authors should place a hat over Y when it is standing for estimated labels?"""
1191,"""Integral Pruning on Activations and Weights for Efficient Neural Networks""","['activation pruning', 'weight pruning', 'computation cost reduction', 'efficient DNNs']","""With the rapidly scaling up of deep neural networks (DNNs), extensive research studies on network model compression such as weight pruning have been performed for efficient deployment. This work aims to advance the compression beyond the weights to the activations of DNNs. We propose the Integral Pruning (IP) technique which integrates the activation pruning with the weight pruning. Through the learning on the different importance of neuron responses and connections, the generated network, namely IPnet, balances the sparsity between activations and weights and therefore further improves execution efficiency. The feasibility and effectiveness of IPnet are thoroughly evaluated through various network models with different activation functions and on different datasets. With <0.5% disturbance on the testing accuracy, IPnet saves 71.1% ~ 96.35% of computation cost, compared to the original dense models with up to 5.8x and 10x reductions in activation and weight numbers, respectively. ""","""This paper proposes to compress the deep learning model using both activation pruning and weight pruning. The reviewers have a consensus on rejection due to lack of novelty. """
1192,"""Learning Localized Generative Models for 3D Point Clouds via Graph Convolution""","['GAN', 'graph convolution', 'point clouds']","""Point clouds are an important type of geometric data and have widespread use in computer graphics and vision. However, learning representations for point clouds is particularly challenging due to their nature as being an unordered collection of points irregularly distributed in 3D space. Graph convolution, a generalization of the convolution operation for data defined over graphs, has been recently shown to be very successful at extracting localized features from point clouds in supervised or semi-supervised tasks such as classification or segmentation. This paper studies the unsupervised problem of a generative model exploiting graph convolution. We focus on the generator of a GAN and define methods for graph convolution when the graph is not known in advance as it is the very output of the generator. The proposed architecture learns to generate localized features that approximate graph embeddings of the output geometry. We also study the problem of defining an upsampling layer in the graph-convolutional generator, such that it learns to exploit a self-similarity prior on the data distribution to sample more effectively.""","""All reviewers gave an accept rating: 9, 7 &6. A clear accept -- just not strong enough reviewer support for an oral."""
1193,"""Deep Layers as Stochastic Solvers""","['deep networks', 'optimization']","""We provide a novel perspective on the forward pass through a block of layers in a deep network. In particular, we show that a forward pass through a standard dropout layer followed by a linear layer and a non-linear activation is equivalent to optimizing a convex objective with a single iteration of a pseudo-formula -nice Proximal Stochastic Gradient method. We further show that replacing standard Bernoulli dropout with additive dropout is equivalent to optimizing the same convex objective with a variance-reduced proximal method. By expressing both fully-connected and convolutional layers as special cases of a high-order tensor product, we unify the underlying convex optimization problem in the tensor setting and derive a formula for the Lipschitz constant pseudo-formula used to determine the optimal step size of the above proximal methods. We conduct experiments with standard convolutional networks applied to the CIFAR-10 and CIFAR-100 datasets and show that replacing a block of layers with multiple iterations of the corresponding solver, with step size set via pseudo-formula , consistently improves classification accuracy.""","""This paper relates deep learning to convex optimization by showing that the forward pass though a dropout layer, linear layer (either convolutional or fully connected), and a nonlinear activation function is equivalent to taking one -nice proximal gradient descent step on a a convex optimization objective. The paper shows (1) how different activation functions correspond to different proximal operators, (2) that replacing Bernoulli dropout with additive dropout corresponds to replacing the -nice proximal gradient descent method with a variance-reduced proximal method, and (3) how to compute the Lipschitz constant required to set the optimal step size in the proximal step. The practical value of this perspective is illustrated in experiments that replace various layers in ConvNet architectures with proximal solvers, leading to performance improvements on CIFAR-10 and CIFAR-100. The reviewers felt that most of their concerns were adequately addressed in the discussion and revision, and that the paper should be accepted."""
1194,"""Scalable Neural Theorem Proving on Knowledge Bases and Natural Language""","['Machine Reading', 'Natural Language Processing', 'Neural Theorem Proving', 'Representation Learning', 'First Order Logic']","""Reasoning over text and Knowledge Bases (KBs) is a major challenge for Artificial Intelligence, with applications in machine reading, dialogue, and question answering. Transducing text to logical forms which can be operated on is a brittle and error-prone process. Operating directly on text by jointly learning representations and transformations thereof by means of neural architectures that lack the ability to learn and exploit general rules can be very data-inefficient and not generalise correctly. These issues are addressed by Neural Theorem Provers (NTPs) (Rocktschel & Riedel, 2017), neuro-symbolic systems based on a continuous relaxation of Prologs backward chaining algorithm, where symbolic unification between atoms is replaced by a differentiable operator computing the similarity between their embedding representations. In this paper, we first propose Neighbourhood-approximated Neural Theorem Provers (NaNTPs) consisting of two extensions toNTPs, namely a) a method for drastically reducing the previously prohibitive time and space complexity during inference and learning, and b) an attention mechanism for improving the rule learning process, deeming them usable on real-world datasets. Then, we propose a novel approach for jointly reasoning over KB facts and textual mentions, by jointly embedding them in a shared embedding space. The proposed method is able to extract rules and provide explanationsinvolving both textual patterns and KB relationsfrom large KBs and text corpora. We show that NaNTPs perform on par with NTPs at a fraction of a cost, and can achieve competitive link prediction results on challenging large-scale datasets, including WN18, WN18RR, and FB15k-237 (with and without textual mentions) while being able to provide explanations for each prediction and extract interpretable rules.""","""This paper focuses on scaling up neural theorem provers, a link prediction system that combines backward chaining with neural embedding of facts, but does not scale to most real-world knowledge bases. The authors introduce a nearest-neighbor search-based method to reduce the time/space complexity, along with an attention mechanism that improves the training. With these extensions, they scale NTP to modern benchmarks for the task, including ones that combine text and knowledge bases, thus providing explanations for such models. The reviewers and the AC note the following as the primary concerns of the paper: (1) the novelty of the contributions is somewhat limited, as nearest neighbor search and attention are both well-known strategies, as is embedding text+facts jointly, (2) there are several issues in the evaluation, in particular around analysis of benefits of the proposed work on new datasets. There were a number of other potential weaknesses, such the performance on some benchmarks (Fb15k) and clarity and writing quality of a few sections. The authors provided significant revisions to the paper that addressed many of the clarity and evaluation concerns, along with providing sufficient comments to better contextualize some of the concerns. However, the concerns with novelty and analysis of the results still hold. Reviewer 3 mentions that it is still unclear in the discussion why the accuracy of the proposed approach matches/outperforms that of NTP, i.e. why is there not a tradeoff. Reviewer 4 also finds the analysis lacking, and feels that the differences between the proposed work and the single-link approaches, in terms of where each excels, are described in insufficient detail. Reviewer 4 focused more on the simplicity of the text encoding, which restricts the novelty as more sophisticated text embeddings approaches are commonplace. Overall, the reviewers raised different concerns, and although all of them appreciated the need for this work and the revisions provided by the authors, ultimately feel that the paper did not quite meet the bar."""
1195,"""SGD Converges to Global Minimum in Deep Learning via Star-convex Path""","['SGD', 'deep learning', 'global minimum', 'convergence']","""Stochastic gradient descent (SGD) has been found to be surprisingly effective in training a variety of deep neural networks. However, there is still a lack of understanding on how and why SGD can train these complex networks towards a global minimum. In this study, we establish the convergence of SGD to a global minimum for nonconvex optimization problems that are commonly encountered in neural network training. Our argument exploits the following two important properties: 1) the training loss can achieve zero value (approximately), which has been widely observed in deep learning; 2) SGD follows a star-convex path, which is verified by various experiments in this paper. In such a context, our analysis shows that SGD, although has long been considered as a randomized algorithm, converges in an intrinsically deterministic manner to a global minimum. ""","""The proposed notion of star convexity is interesting and the empirical work done to provide evidence that it is indeed present in real-world neural network training is appreciated. The reviewers raise a number of concerns. The authors were able to convince some of the reviewers with new experiments under MSE loss and experiments showing how robust the method was to the reference point. The most serious concerns relate to novelty and the assumptions that individual functions share a global minima with respect to which the path of iterates generated by SGD satisfies the star convexity property. I'm inclined to accept the authors rebuttal, although it would have been nicer had the reviewer re-engaged. Overall, the paper is on the borderline."""
1196,"""Critical Learning Periods in Deep Networks""","['Critical Period', 'Deep Learning', 'Information Theory', 'Artificial Neuroscience', 'Information Plasticity']","""Similar to humans and animals, deep artificial neural networks exhibit critical periods during which a temporary stimulus deficit can impair the development of a skill. The extent of the impairment depends on the onset and length of the deficit window, as in animal models, and on the size of the neural network. Deficits that do not affect low-level statistics, such as vertical flipping of the images, have no lasting effect on performance and can be overcome with further training. To better understand this phenomenon, we use the Fisher Information of the weights to measure the effective connectivity between layers of a network during training. Counterintuitively, information rises rapidly in the early phases of training, and then decreases, preventing redistribution of information resources in a phenomenon we refer to as a loss of ""Information Plasticity"". Our analysis suggests that the first few epochs are critical for the creation of strong connections that are optimal relative to the input data distribution. Once such strong connections are created, they do not appear to change during additional training. These findings suggest that the initial learning transient, under-scrutinized compared to asymptotic behavior, plays a key role in determining the outcome of the training process. Our findings, combined with recent theoretical results in the literature, also suggest that forgetting (decrease of information in the weights) is critical to achieving invariance and disentanglement in representation learning. Finally, critical periods are not restricted to biological systems, but can emerge naturally in learning systems, whether biological or artificial, due to fundamental constrains arising from learning dynamics and information processing.""","""Irrespective of their taste for comparisons of neural networks to biological organisms, all reviewers agree that the empirical observations in this paper are quite interesting and well presented. While some reviewers note that the paper is not making theoretical contributions, the empirical results in themselves are intriguing enough to be of interest to ICLR audiences."""
1197,"""Temporal Difference Variational Auto-Encoder""","['generative models', 'variational auto-encoders', 'state space models', 'temporal difference learning']","""To act and plan in complex environments, we posit that agents should have a mental simulator of the world with three characteristics: (a) it should build an abstract state representing the condition of the world; (b) it should form a belief which represents uncertainty on the world; (c) it should go beyond simple step-by-step simulation, and exhibit temporal abstraction. Motivated by the absence of a model satisfying all these requirements, we propose TD-VAE, a generative sequence model that learns representations containing explicit beliefs about states several steps into the future, and that can be rolled out directly without single-step transitions. TD-VAE is trained on pairs of temporally separated time points, using an analogue of temporal difference learning used in reinforcement learning.""","""The reviewers agree that this is a novel paper with a convincing evaluation."""
1198,"""Relational Graph Attention Networks""","['RGCN', 'attention', 'graph convolutional networks', 'semi-supervised learning', 'graph classification', 'molecules']","""We investigate Relational Graph Attention Networks, a class of models that extends non-relational graph attention mechanisms to incorporate relational information, opening up these methods to a wider variety of problems. A thorough evaluation of these models is performed, and comparisons are made against established benchmarks. To provide a meaningful comparison, we retrain Relational Graph Convolutional Networks, the spectral counterpart of Relational Graph Attention Networks, and evaluate them under the same conditions. We find that Relational Graph Attention Networks perform worse than anticipated, although some configurations are marginally beneficial for modelling molecular properties. We provide insights as to why this may be, and suggest both modifications to evaluation strategies, as well as directions to investigate for future work.""","""The authors propose an architecture for learning and predicting graphs with relations between nodes. The approach is a combination of recent research efforts into Graph Attention Networks and Relational Graph Convolutional Networks. The authors are commended for their clear and direct writing and presentation and their honest claims and their empirical setup. However, the paper simply doesn't have much to offer to the community, since the algorithmic contributions are marginal and the results unimpressive. While the authors justify the submission in terms of the difficult implementation and the extensive experiments, this is not enough to support its publication at a top conference. Rather, this could be a technical report."""
1199,"""Bounce and Learn: Modeling Scene Dynamics with Real-World Bounces""","['intuitive physics', 'visual prediction', 'surface normal', 'restitution', 'bounces']","""We introduce an approach to model surface properties governing bounces in everyday scenes. Our model learns end-to-end, starting from sensor inputs, to predict post-bounce trajectories and infer two underlying physical properties that govern bouncing - restitution and effective collision normals. Our model, Bounce and Learn, comprises two modules -- a Physics Inference Module (PIM) and a Visual Inference Module (VIM). VIM learns to infer physical parameters for locations in a scene given a single still image, while PIM learns to model physical interactions for the prediction task given physical parameters and observed pre-collision 3D trajectories. To achieve our results, we introduce the Bounce Dataset comprising 5K RGB-D videos of bouncing trajectories of a foam ball to probe surfaces of varying shapes and materials in everyday scenes including homes and offices. Our proposed model learns from our collected dataset of real-world bounces and is bootstrapped with additional information from simple physics simulations. We show on our newly collected dataset that our model out-performs baselines, including trajectory fitting with Newtonian physics, in predicting post-bounce trajectories and inferring physical properties of a scene.""","""This paper proposes a novel dataset of bouncing balls and a way to learn the dynamics of the balls when colliding. The reviewers found the paper well-written, tackling an interesting and hard problem in a novel way. The main concern (that I share with one of the reviewers) is about the fact that the paper proposes both a new dataset/environment *and* a solution for the problem. This made it difficult the for the authors to provide baselines to compare to. The ensuing back and forth had the authors relax some of the assumptions from the environment and made it possible to evaluate with interaction nets. The main weakness of the paper is the relatively contrived setup that the authors have come up with. I will summarize some of the discussion that happened as a result of this point: it is relatively difficult to see how this setup that the authors have and have studied (esp. knowing the groundtruth impact locations and the timing of the impact) can generalize outside of the proposed approach. There is some concern that the comparison with interaction nets was not entirely fair. I would recommend the authors redo the comparisons with interaction nets in a careful way, with the right ablations, and understand if the methods have access to the same input data (e.g. are interaction nets provided with the bounce location?). Despite the relatively high average score, I think of this paper as quite borderline, specifically because of the issues related to the setup being too niche. Nonetheless, the work does have a lot of scientific value to it, in addition to a new simulation environment/dataset that other researchers can then use. Assuming the baselines are done in a way that is trustworthy, the ablation experiments and discussion will be something interesting to the ICLR community."""
1200,"""ON BREIMANS DILEMMA IN NEURAL NETWORKS: SUCCESS AND FAILURE OF NORMALIZED MARGINS""","[""Bregman's Dilemma"", 'Generalization Error', 'Margin', 'Spectral normalization']","""A belief persists long in machine learning that enlargement of margins over training data accounts for the resistance of models to overfitting by increasing the robustness. Yet Breiman shows a dilemma (Breiman, 1999) that a uniform improvement on margin distribution \emph{does not} necessarily reduces generalization error. In this paper, we revisit Breiman's dilemma in deep neural networks with recently proposed normalized margins using Lipschitz constant bound by spectral norm products. With both simplified theory and extensive experiments, Breiman's dilemma is shown to rely on dynamics of normalized margin distributions, that reflects the trade-off between model expression power and data complexity. When the complexity of data is comparable to the model expression power in the sense that training and test data share similar phase transitions in normalized margin dynamics, two efficient ways are derived via classic margin-based generalization bounds to successfully predict the trend of generalization error. On the other hand, over-expressed models that exhibit uniform improvements on training normalized margins may lose such a prediction power and fail to prevent the overfitting. ""","""The reviewers reached a consensus that the paper is not ready for publication in ICLR. (see more details in the reviews below. )"""
1201,"""Guided Evolutionary Strategies: Escaping the curse of dimensionality in random search""","['evolutionary strategies', 'optimization', 'gradient estimators', 'biased gradients']","""Many applications in machine learning require optimizing a function whose true gradient is unknown, but where surrogate gradient information (directions that may be correlated with, but not necessarily identical to, the true gradient) is available instead. This arises when an approximate gradient is easier to compute than the full gradient (e.g. in meta-learning or unrolled optimization), or when a true gradient is intractable and is replaced with a surrogate (e.g. in certain reinforcement learning applications or training networks with discrete variables). We propose Guided Evolutionary Strategies, a method for optimally using surrogate gradient directions along with random search. We define a search distribution for evolutionary strategies that is elongated along a subspace spanned by the surrogate gradients. This allows us to estimate a descent direction which can then be passed to a first-order optimizer. We analytically and numerically characterize the tradeoffs that result from tuning how strongly the search distribution is stretched along the guiding subspace, and use this to derive a setting of the hyperparameters that works well across problems. Finally, we apply our method to example problems including truncated unrolled optimization and training neural networks with discrete variables, demonstrating improvement over both standard evolutionary strategies and first-order methods (that directly follow the surrogate gradient). We provide a demo of Guided ES at: redacted URL""","""This paper proposes a guided evolution strategy method where the past surrogate gradients are used to construct a covariance matrix from which future perturbations are sampled. The bias-variance tradeoff is analyzed and the method is applied to real-world examples. The method is not entirely new, and discussion of related work as well as comparison with them is missing. The main contribution is in the analysis and application to real-world examples, and the paper should be rewritten focusing on these contributions, while discussing existing work on this topic thoroughly. Due to these issue, I recommend to reject this paper. """
1202,"""EDDI: Efficient Dynamic Discovery of High-Value Information with Partial VAE""","['active variable selection', 'missing data', 'amortized inference']","""Making decisions requires information relevant to the task at hand. Many real-life decision-making situations allow acquiring further relevant information at a specific cost. For example, in assessing the health status of a patient we may decide to take additional measurements such as diagnostic tests or imaging scans before making a final assessment. More information that is relevant allows for better decisions but it may be costly to acquire all of this information. How can we trade off the desire to make good decisions with the option to acquire further information at a cost? To this end, we propose a principled framework, named EDDI (Efficient Dynamic Discovery of high-value Information), based on the theory of Bayesian experimental design. In EDDI we propose a novel partial variational autoencoder (Partial VAE), to efficiently handle missing data over varying subsets of known information. EDDI combines this Partial VAE with an acquisition function that maximizes expected information gain on a set of target variables. EDDI is efficient and demonstrates that dynamic discovery of high-value information is possible; we show cost reduction at the same decision quality and improved decision quality at the same cost in benchmarks and in two health-care applications.. We believe there is great potential for realizing these gains in real-world decision support systems.""","""This paper develops an active variable selection framework that couples a partial variational autoencoder capable of handling missing data with an information acquisition criteria derived from Bayesian experimental design. The paper is generally well written and the formulation appears to be natural, with a compelling real world healthcare application. The topic is relatively under-explored in deep learning and the paper appears to attempt to set a valuable baseline. However, the AC cannot recommend acceptance based on the fact that reviewer 2 has brought up concerns about the competitiveness of the approach relative to alternative methods reported in the experimental section, and all reviewers have found various parts of the paper to have room for improvement with regards to technical clarity. As such the paper would benefit from a revision and a stronger resubmission."""
1203,"""Toward Understanding the Impact of Staleness in Distributed Machine Learning""",[],"""Most distributed machine learning (ML) systems store a copy of the model parameters locally on each machine to minimize network communication. In practice, in order to reduce synchronization waiting time, these copies of the model are not necessarily updated in lock-step, and can become stale. Despite much development in large-scale ML, the effect of staleness on the learning efficiency is inconclusive, mainly because it is challenging to control or monitor the staleness in complex distributed environments. In this work, we study the convergence behaviors of a wide array of ML models and algorithms under delayed updates. Our extensive experiments reveal the rich diversity of the effects of staleness on the convergence of ML algorithms and offer insights into seemingly contradictory reports in the literature. The empirical findings also inspire a new convergence analysis of SGD in non-convex optimization under staleness, matching the best-known convergence rate of O(1/\sqrt{T}).""","""The reviewers that provided extensive and technically well-justified reviews agreed that the paper is of high quality. The authors are encouraged to make sure all concerns of these reviewers are properly addressed in the paper."""
1204,"""RESIDUAL NETWORKS CLASSIFY INPUTS BASED ON THEIR NEURAL TRANSIENT DYNAMICS""","['Residual Networks', 'Dynamical Systems', 'Classification']","""In this study, we analyze the input-output behavior of residual networks from a dynamical system point of view by disentangling the residual dynamics from the output activities before the classification stage. For a network with simple skip connections between every successive layer, and for logistic activation function, and shared weights between layers, we show analytically that there is a cooperation and competition dynamics between residuals corresponding to each input dimension. Interpreting these kind of networks as nonlinear filters, the steady state value of the residuals in the case of attractor networks are indicative of the common features between different input dimensions that the network has observed during training, and has encoded in those components. In cases where residuals do not converge to an attractor state, their internal dynamics are separable for each input class, and the network can reliably approximate the output. We bring analytical and empirical evidence that residual networks classify inputs based on the integration of the transient dynamics of the residuals, and will show how the network responds to input perturbations. We compare the network dynamics for a ResNet and a Multi-Layer Perceptron and show that the internal dynamics, and the noise evolution are fundamentally different in these networks, and ResNets are more robust to noisy inputs. Based on these findings, we also develop a new method to adjust the depth for residual networks during training. As it turns out, after pruning the depth of a ResNet using this algorithm,the network is still capable of classifying inputs with a high accuracy.""","""The paper uses dynamical systems theory to evaluate feed-forward neural networks. The theory is used to compute the optimal depth of resnets. An interesting approach, and a good initiative. At the same time, the approach seems not to be thought through well enough, and the work needs another level of maturation before publication. The application that is realised is too immature, and the corresponding contributions are not significant in their current form. All reviewers agree on rejection of the paper."""
1205,"""DHER: Hindsight Experience Replay for Dynamic Goals""","['Sparse rewards', 'Dynamic goals', 'Experience replay']","""Dealing with sparse rewards is one of the most important challenges in reinforcement learning (RL), especially when a goal is dynamic (e.g., to grasp a moving object). Hindsight experience replay (HER) has been shown an effective solution to handling sparse rewards with fixed goals. However, it does not account for dynamic goals in its vanilla form and, as a result, even degrades the performance of existing off-policy RL algorithms when the goal is changing over time. In this paper, we present Dynamic Hindsight Experience Replay (DHER), a novel approach for tasks with dynamic goals in the presence of sparse rewards. DHER automatically assembles successful experiences from two relevant failures and can be used to enhance an arbitrary off-policy RL algorithm when the tasks' goals are dynamic. We evaluate DHER on tasks of robotic manipulation and moving object tracking, and transfer the polices from simulation to physical robots. Extensive comparison and ablation studies demonstrate the superiority of our approach, showing that DHER is a crucial ingredient to enable RL to solve tasks with dynamic goals in manipulation and grid world domains.""","""This work proposes a method for extending hindsight experience replay to the setting where the goal is not fixed, but dynamic or moving. It proceeds by amending failed episodes by searching replay memory for a compatible trajectories from which to construct a trajectory that can be productively learned from. Reviewers were generally positive on the novelty and importance of the contribution. While noting its limitations, it was still felt that the key ideas could be useful and influential. The tasks considered are modifications of OpenAI robotics environments, adapted to the dynamic goal setting, as well as a 2D planar ""snake"" game. There were concerns about the strength of the baselines employed but reviewers seemed happy with the state of these post-revision. There were also concerns regarding clarity of presentation, particularly from AnonReviewer2, but significant progress was made on this front following discussions and revision. Despite remaining concerns over clarity I am convinced that this is an interesting problem setting worth studying and that the proposed method makes significant progress. The method has limitations with respect to the sorts of environments where we can reasonably expect it to work (where other aspects of the environment are relatively stable both within and across episodes), but there is lots of work in the literature, particularly where robotics is concerned, that focuses on exactly these kinds of environments. This submission is therefore highly relevant to current practice and by reviewers' accounts, generally well-executed in its post-revision form. I therefore recommend acceptance."""
1206,"""Poincare Glove: Hyperbolic Word Embeddings""","['word embeddings', 'hyperbolic spaces', 'poincare ball', 'hypernymy', 'analogy', 'similarity', 'gaussian embeddings']","""Words are not created equal. In fact, they form an aristocratic graph with a latent hierarchical structure that the next generation of unsupervised learned word embeddings should reveal. In this paper, justified by the notion of delta-hyperbolicity or tree-likeliness of a space, we propose to embed words in a Cartesian product of hyperbolic spaces which we theoretically connect to the Gaussian word embeddings and their Fisher geometry. This connection allows us to introduce a novel principled hypernymy score for word embeddings. Moreover, we adapt the well-known Glove algorithm to learn unsupervised word embeddings in this type of Riemannian manifolds. We further explain how to solve the analogy task using the Riemannian parallel transport that generalizes vector arithmetics to this new type of geometry. Empirically, based on extensive experiments, we prove that our embeddings, trained unsupervised, are the first to simultaneously outperform strong and popular baselines on the tasks of similarity, analogy and hypernymy detection. In particular, for word hypernymy, we obtain new state-of-the-art on fully unsupervised WBLESS classification accuracy.""","""Word vectors are well studied but this paper adds yet another interesting dimension to the field."""
1207,"""Learning to Screen for Fast Softmax Inference on Large Vocabulary Neural Networks""","['fast inference', 'softmax computation', 'natural language processing']","""Neural language models have been widely used in various NLP tasks, including machine translation, next word prediction and conversational agents. However, it is challenging to deploy these models on mobile devices due to their slow prediction speed, where the bottleneck is to compute top candidates in the softmax layer. In this paper, we introduce a novel softmax layer approximation algorithm by exploiting the clustering structure of context vectors. Our algorithm uses a light-weight screening model to predict a much smaller set of candidate words based on the given context, and then conducts an exact softmax only within that subset. Training such a procedure end-to-end is challenging as traditional clustering methods are discrete and non-differentiable, and thus unable to be used with back-propagation in the training process. Using the Gumbel softmax, we are able to train the screening model end-to-end on the training set to exploit data distribution. The algorithm achieves an order of magnitude faster inference than the original softmax layer for predicting top-k words in various tasks such as beam search in machine translation or next words prediction. For example, for machine translation task on German to English dataset with around 25K vocabulary, we can achieve 20.4 times speed up with 98.9% precision@1 and 99.3% precision@5 with the original softmax layer prediction, while state-of-the-art (Zhang et al., 2018) only achieves 6.7x speedup with 98.7% precision@1 and 98.1% precision@5 for the same task.""","""This paper introduces an approach for improving the scalability of neural network models with large output spaces, where naive soft-max inference scales linearly with the vocabulary size. The proposed approach is based on a clustering step combined with per-cluster, smaller soft-maxes. It retains differentiability with the Gumbel softmax trick. The experimental results are impressive. There are some minor flaws, however there's consensus among the reviewers the paper should be published. """
1208,"""LIT: Block-wise Intermediate Representation Training for Model Compression""",[],"""Knowledge distillation (KD) is a popular method for reducing the computational over- head of deep network inference, in which the output of a teacher model is used to train a smaller, faster student model. Hint training (i.e., FitNets) extends KD by regressing a student models intermediate representation to a teacher models intermediate representa- tion. In this work, we introduce bLock-wise Intermediate representation Training (LIT), a novel model compression technique that extends the use of intermediate represen- tations in deep network compression, outperforming KD and hint training. LIT has two key ideas: 1) LIT trains a student of the same width (but shallower depth) as the teacher by directly comparing the intermediate representations, and 2) LIT uses the intermediate representation from the previous block in the teacher model as an input to the current stu- dent block during training, avoiding unstable intermediate representations in the student network. We show that LIT provides substantial reductions in network depth without loss in accuracy for example, LIT can compress a ResNeXt-110 to a ResNeXt-20 (5.5) on CIFAR10 and a VDCNN-29 to a VDCNN-9 (3.2) on Amazon Reviews without loss in accuracy, outperforming KD and hint training in network size at a given accuracy. We also show that applying LIT to identical student/teacher architectures increases the accuracy of the student model above the teacher model, outperforming the recently-proposed Born Again Networks procedure on ResNet, ResNeXt, and VDCNN. Finally, we show that LIT can effectively compress GAN generators.""","""The authors propose a method for distilling a student network from a teacher network and while additionally constraining the intermediate representations from the student to match those of the teacher, where the student has the same width, but less depth than the teacher. The main novelty of the work is to use the intermediate representation from the teacher as an input to the student network, and the experimental comparison of the approach against previous work. The reviewers noted that the method is simple to implement, and the paper is clearly written and easy to follow. The reviewers raised some concerns, most notably that the authors were using validation accuracy to measure performance, and were thus potentially overfitting to the test data, and regarding the novelty of the work. Some of the criticisms were subsequently amended in the revised version where results were reported on a test set (the conclusions are as before). Overall, the scores for this paper were close to the threshold for acceptance, and while it was a tough decision, the AC ultimately felt that the overall novelty of the work was slightly below the acceptance bar."""
1209,"""Detecting Memorization in ReLU Networks""","['Memorization', 'Generalization', 'ReLU', 'Non-negative matrix factorization']","""We propose a new notion of 'non-linearity' of a network layer with respect to an input batch that is based on its proximity to a linear system, which is reflected in the non-negative rank of the activation matrix. We measure this non-linearity by applying non-negative factorization to the activation matrix. Considering batches of similar samples, we find that high non-linearity in deep layers is indicative of memorization. Furthermore, by applying our approach layer-by-layer, we find that the mechanism for memorization consists of distinct phases. We perform experiments on fully-connected and convolutional neural networks trained on several image and audio datasets. Our results demonstrate that as an indicator for memorization, our technique can be used to perform early stopping.""","""This paper proposes a new measure to detect memorization based on how well the activations of the network are approximated by a low-rank decomposition. They compare decompositions and find that non-negative matrix factorization provides the best results. They evaluate of several datasets and show that the measure is well correlated with generalization and can be used for early stopping. All reviewers found the work novel, but there were concerns about the usefulness of the method, the experimental setup and the assumptions made. Some of these concerns were addressed by the revisions but concerns about usefulness and insights remained. These issues need to be properly addressed before acceptance."""
1210,"""Don't Settle for Average, Go for the Max: Fuzzy Sets and Max-Pooled Word Vectors""","['word vectors', 'sentence representations', 'distributed representations', 'fuzzy sets', 'bag-of-words', 'unsupervised learning', 'word vector compositionality', 'max-pooling', 'Jaccard index']","""Recent literature suggests that averaged word vectors followed by simple post-processing outperform many deep learning methods on semantic textual similarity tasks. Furthermore, when averaged word vectors are trained supervised on large corpora of paraphrases, they achieve state-of-the-art results on standard STS benchmarks. Inspired by these insights, we push the limits of word embeddings even further. We propose a novel fuzzy bag-of-words (FBoW) representation for text that contains all the words in the vocabulary simultaneously but with different degrees of membership, which are derived from similarities between word vectors. We show that max-pooled word vectors are only a special case of fuzzy BoW and should be compared via fuzzy Jaccard index rather than cosine similarity. Finally, we propose DynaMax, a completely unsupervised and non-parametric similarity measure that dynamically extracts and max-pools good features depending on the sentence pair. This method is both efficient and easy to implement, yet outperforms current baselines on STS tasks by a large margin and is even competitive with supervised word vectors trained to directly optimise cosine similarity.""","""This paper presents new generalized methods for representing sentences and measuring their similarities based on word vectors. More specifically, the paper presents Fuzzy Bag-of-Words (FBoW), a generalized approach to composing sentence embeddings by combining word embeddings with different degrees of membership, which generalize more commonly used average or max-pooled vector representations. In addition, the paper presents DynaMax, an unsupervised and non-parametric similarity measure that can dynamically extract and max-pool features from a sentence pair. Pros: The proposed methods are natural generalization of exiting average and max-pooled vectors. The proposed methods are elegant, simple, easy to implement, and demonstrate strong performance on STS tasks. Cons: The paper is solid, no significant con other than that the proposed methods are not groundbreaking innovations per say. Verdict: The simplicity is what makes the proposed methods elegant. The empirical results are strong. The paper is worthy of acceptance."""
1211,"""Generative adversarial interpolative autoencoding: adversarial training on latent space interpolations encourages convex latent distributions""","['convex', 'GAN', 'autoencoder', 'interpolation', 'stimuli generation', 'adversarial', 'latent distribution']","""We present a neural network architecture based upon the Autoencoder (AE) and Generative Adversarial Network (GAN) that promotes a convex latent distribution by training adversarially on latent space interpolations. By using an AE as both the generator and discriminator of a GAN, we pass a pixel-wise error function across the discriminator, yielding an AE which produces sharp samples that match both high- and low-level features of the original images. Samples generated from interpolations between data in latent space remain within the distribution of real data as trained by the discriminator, and therefore preserve realistic resemblances to the network inputs.""","""The idea of the paper -- imposing a GAN type loss on the latent interpolations of an autoencoder -- is interesting. However there are strong concerns from R2 and R3 about limited experimental evaluation of the proposed method which falls short of demonstrating its advantages over latent spaces learned by existing GANs. Another point of concern was the use of only one real dataset (CelebA). Authors made substantial revisions to the paper in addressing many of the reviewers' points but these core concerns still persist with the current draft and it's not ready for publication at ICLR. Authors are encouraged to address these concerns and resubmit to another venue. """
1212,"""Out-of-Sample Extrapolation with Neuron Editing""","['generative adversarial networks', 'computational biology', 'generating', 'generation', 'extrapolation', 'out-of-sample', 'neural network inference']","""While neural networks can be trained to map from one specific dataset to another, they usually do not learn a generalized transformation that can extrapolate accurately outside the space of training. For instance, a generative adversarial network (GAN) exclusively trained to transform images of cars from light to dark might not have the same effect on images of horses. This is because neural networks are good at generation within the manifold of the data that they are trained on. However, generating new samples outside of the manifold or extrapolating ""out-of-sample"" is a much harder problem that has been less well studied. To address this, we introduce a technique called neuron editing that learns how neurons encode an edit for a particular transformation in a latent space. We use an autoencoder to decompose the variation within the dataset into activations of different neurons and generate transformed data by defining an editing transformation on those neurons. By performing the transformation in a latent trained space, we encode fairly complex and non-linear transformations to the data with much simpler distribution shifts to the neuron's activations. We showcase our technique on image domain/style transfer and two biological applications: removal of batch artifacts representing unwanted noise and modeling the effect of drug treatments to predict synergy between drugs.""","""This was a borderline paper, as reviewers generally agreed that the method was a new method that was appropriately explained and motivated and had reasonable experimental results. The main drawbacks were that the significance of the method was unclear. In particular, the method might be too inflexible due to being based on a hard-coded rule, and it is not clear why this is the right approach relative to e.g. GANs with a modified training objective). Reviewers also had difficulty assessing the significance of the results on biological datasets. While such results certainly add to the paper, the paper would be stronger if the argument for significance could be assessed from more standard datasets. A note on the review process: the reviewers initially scored the paper 6/6/6, but the review text for some of the reviews was more negative than a typical 6 score. To confirm this, I asked if any reviewers wanted to push for acceptance. None of the reviewers did (generally due to feeling the significance of the results was limited) and two of the reviewers decided to lower their scores to account for this."""
1213,"""Reward Constrained Policy Optimization""","['reinforcement learning', 'markov decision process', 'constrained markov decision process', 'deep learning']","""Solving tasks in Reinforcement Learning is no easy feat. As the goal of the agent is to maximize the accumulated reward, it often learns to exploit loopholes and misspecifications in the reward signal resulting in unwanted behavior. While constraints may solve this issue, there is no closed form solution for general constraints. In this work we present a novel multi-timescale approach for constrained policy optimization, called `Reward Constrained Policy Optimization' (RCPO), which uses an alternative penalty signal to guide the policy towards a constraint satisfying one. We prove the convergence of our approach and provide empirical evidence of its ability to train constraint satisfying policies.""","""This work is novel, reasonably clearly written with a thorough literature survey. The proposed approach also empirically seems promising. The paper could be improved with a bit more discussion about the sensitivity, particularly as a two-timescale approach can be more difficult to tune."""
1214,"""ProbGAN: Towards Probabilistic GAN with Theoretical Guarantees""","['Generative Adversarial Networks', 'Bayesian Deep Learning', 'Mode Collapse', 'Inception Score', 'Generator', 'Discriminator', 'CIFAR-10', 'STL-10', 'ImageNet']","""Probabilistic modelling is a principled framework to perform model aggregation, which has been a primary mechanism to combat mode collapse in the context of Generative Adversarial Networks (GAN). In this paper, we propose a novel probabilistic framework for GANs, ProbGAN, which iteratively learns a distribution over generators with a carefully crafted prior. Learning is efficiently triggered by a tailored stochastic gradient Hamiltonian Monte Carlo with a novel gradient approximation to perform Bayesian inference. Our theoretical analysis further reveals that our treatment is the first probabilistic framework that yields an equilibrium where generator distributions are faithful to the data distribution. Empirical evidence on synthetic high-dimensional multi-modal data and image databases (CIFAR-10, STL-10, and ImageNet) demonstrates the superiority of our method over both start-of-the-art multi-generator GANs and other probabilistic treatment for GANs.""","""The paper proposes a new method that builds on the Bayesian modelling framework for GANs and is supported by a theoretical analysis and an empirical evaluation that shows very promising results. All reviewers agree, that the method is interesting and the results are convincing, but that the model does not really fit in the standard Bayesian setting due to a data dependency of the priors. I would therefore encourage the authors to reflect this by adapting the title and making the differences more clear in the camera ready version."""
1215,"""Systematic Generalization: What Is Required and Can It Be Learned?""","['systematic generalization', 'language understanding', 'visual questions answering', 'neural module networks']","""Numerous models for grounded language understanding have been recently proposed, including (i) generic models that can be easily adapted to any given task and (ii) intuitively appealing modular models that require background knowledge to be instantiated. We compare both types of models in how much they lend themselves to a particular form of systematic generalization. Using a synthetic VQA test, we evaluate which models are capable of reasoning about all possible object pairs after training on only a small subset of them. Our findings show that the generalization of modular models is much more systematic and that it is highly sensitive to the module layout, i.e. to how exactly the modules are connected. We furthermore investigate if modular models that generalize well could be made more end-to-end by learning their layout and parametrization. We find that end-to-end methods from prior work often learn inappropriate layouts or parametrizations that do not facilitate systematic generalization. Our results suggest that, in addition to modularity, systematic generalization in language understanding may require explicit regularizers or priors. ""","""This paper generated a lot of discussion. Paper presents an empirical evaluation of generalization in models for visual reasoning. All reviewers generally agree that it presents a thorough evaluation with a good set of questions. The only remaining concerns of R3 (the sole negative vote) were lack of surprise in findings and lingering questions of whether these results generalize to realistic settings. The former suffers from hindsight bias and tends to be an unreliable indicator of the impact of a paper. The latter is an open question and should be worked on, but in the opinion of the AC, does not preclude publication of this manuscript. These experiments are well done and deserve to be published. If the findings don't generalize to more complex settings, we will let the noisy process of science correct our understanding in the future. """
1216,"""TherML: The Thermodynamics of Machine Learning""","['representation learning', 'information theory', 'information bottleneck', 'thermodynamics', 'predictive information']","""In this work we offer an information-theoretic framework for representation learning that connects with a wide class of existing objectives in machine learning. We develop a formal correspondence between this work and thermodynamics and discuss its implications.""","""Connecting different fields and bringing new insights to machine learning are always appreciated. But since it is challenging to do it needs to be done well. This paper falls short here. """
1217,"""Optimal Control Via Neural Networks: A Convex Approach""","['optimal control', 'input convex neural network', 'convex optimization']","""Control of complex systems involves both system identification and controller design. Deep neural networks have proven to be successful in many identification tasks, however, from model-based control perspective, these networks are difficult to work with because they are typically nonlinear and nonconvex. Therefore many systems are still identified and controlled based on simple linear models despite their poor representation capability. In this paper we bridge the gap between model accuracy and control tractability faced by neural networks, by explicitly constructing networks that are convex with respect to their inputs. We show that these input convex networks can be trained to obtain accurate models of complex physical systems. In particular, we design input convex recurrent neural networks to capture temporal behavior of dynamical systems. Then optimal controllers can be achieved via solving a convex model predictive control problem. Experiment results demonstrate the good potential of the proposed input convex neural network based approach in a variety of control applications. In particular we show that in the MuJoCo locomotion tasks, we could achieve over 10% higher performance using 5 times less time compared with state-of-the-art model-based reinforcement learning method; and in the building HVAC control example, our method achieved up to 20% energy reduction compared with classic linear models. ""","""The paper makes progress on a problem that is still largely unexplored, presents promising results, and builds bridges with prior work on optimal control. It designs input convex recurrent neural networks to capture temporal behavior of dynamical systems; this then allows optimal controllers to be computed by solving a convex model predictive control problem. There were initial critiques regarding some of the claims. These have now been clarified. Also, there is in the end a compromise between the (necessary) approximations of the input-convex model and the true dynamics, and being able to compute an optimal result. Overall, all reviewers and the AC are in agreement to see this paper accepted. There was extensive and productive interaction between the reviewers and authors. It makes contributions that will be of interest to many, and builds interesting bridges with known control methods."""
1218,"""On Regularization and Robustness of Deep Neural Networks""","['regularization', 'robustness', 'deep learning', 'convolutional networks', 'kernel methods']","""In this work, we study the connection between regularization and robustness of deep neural networks by viewing them as elements of a reproducing kernel Hilbert space (RKHS) of functions and by regularizing them using the RKHS norm. Even though this norm cannot be computed, we consider various approximations based on upper and lower bounds. These approximations lead to new strategies for regularization, but also to existing ones such as spectral norm penalties or constraints, gradient penalties, or adversarial training. Besides, the kernel framework allows us to obtain margin-based bounds on adversarial generalization. We show that our new algorithms lead to empirical benefits for learning on small datasets and learning adversarially robust models. We also discuss implications of our regularization framework for learning implicit generative models.""","""Reviewers generally found the RKHS perspective interesting, but did not feel that the results in the work (many of which were already known or follow easily from known theory) are sufficient to form a complete paper. Authors are encouraged to read the detailed reviewer comments which contain a number of critiques and suggestions for improvement."""
1219,"""Neural Variational Inference For Embedding Knowledge Graphs""","['Statistical Relational Learning', 'Knowledge Graphs', 'Knowledge Extraction', 'Latent Feature Models', 'Variational Inference.']","""Recent advances in Neural Variational Inference allowed for a renaissance in latent variable models in a variety of domains involving high-dimensional data. In this paper, we introduce two generic Variational Inference frameworks for generative models of Knowledge Graphs; Latent Fact Model and Latent Information Model. While traditional variational methods derive an analytical approximation for the intractable distribution over the latent variables, here we construct an inference network conditioned on the symbolic representation of entities and relation types in the Knowledge Graph, to provide the variational distributions. The new framework can create models able to discover underlying probabilistic semantics for the symbolic representation by utilising parameterisable distributions which permit training by back-propagation in the context of neural variational inference, resulting in a highly-scalable method. Under a Bernoulli sampling framework, we provide an alternative justification for commonly used techniques in large-scale stochastic variational inference, which drastically reduces training time at a cost of an additional approximation to the variational lower bound. The generative frameworks are flexible enough to allow training under any prior distribution that permits a re-parametrisation trick, as well as under any scoring function that permits maximum likelihood estimation of the parameters. Experiment results display the potential and efficiency of this framework by improving upon multiple benchmarks with Gaussian prior representations. Code publicly available on Github.""","""The paper proposes a novel variational inference framework for knowledge graphs which is evaluated on link prediction benchmark sets and is competitive to previous generative approaches. While the idea is interstnig and technically correct, the originality of the contribution is limited, and the paper would be clearly improved by providing a clearer motivation for using generative models instead of standard methods and a experimental demonstration of the benefits of using a generative instead of a discriminative model, especially since the standard method perform slightly better in the experiments. Overall, the work is slightly under the acceptance threshold. """
1220,"""Learning Heuristics for Automated Reasoning through Reinforcement Learning""","['reinforcement learning', 'deep learning', 'logics', 'formal methods', 'automated reasoning', 'backtracking search', 'satisfiability', 'quantified Boolean formulas']","""We demonstrate how to learn efficient heuristics for automated reasoning algorithms through deep reinforcement learning. We focus on backtracking search algorithms for quantified Boolean logics, which already can solve formulas of impressive size - up to 100s of thousands of variables. The main challenge is to find a representation of these formulas that lends itself to making predictions in a scalable way. For challenging problems, the heuristic learned through our approach reduces execution time by a factor of 10 compared to the existing handwritten heuristics.""","""The paper proposes the use of reinforcement learning to learn heuristics in backtracking search algorithm for quantified boolean formulas, using a neural network to learn a suitable representation of literals and clauses to predict actions. The writing and the description of the method and results are generally clear. The main novelty lies in finding a good architecture/representation of the input, and demonstrating the use of RL in a new domain. While there is no theoretical justification for why this heuristic should work better than existing ones, the experimental results look convincing, although they are somewhat limited and the improvements are dataset dependent. In practice, the overhead of the proposed method could be an issue. There was some disagreement among the reviewers as to whether the improvements and the results are significant enough for publication."""
1221,"""Unsupervised Conditional Generation using noise engineered mode matching GAN""","['Noise engineered GAN', 'Latent space engineering', 'Mode matching', 'Unsupervised learning']","""Conditional generation refers to the process of sampling from an unknown distribution conditioned on semantics of the data. This can be achieved by augmenting the generative model with the desired semantic labels, albeit it is not straightforward in an unsupervised setting where the semantic label of every data sample is unknown. In this paper, we address this issue by proposing a method that can generate samples conditioned on the properties of a latent distribution engineered in accordance with a certain data prior. In particular, a latent space inversion network is trained in tandem with a generative adversarial network such that the modal properties of the latent space distribution are induced in the data generating distribution. We demonstrate that our model, despite being fully unsupervised, is effective in learning meaningful representations through its mode matching property. We validate our method on multiple unsupervised tasks such as conditional generation, dataset attribute discovery and inference using three real world image datasets namely MNIST, CIFAR-10 and CELEB-A and show that the results are comparable to the state-of-the-art methods. ""","""The paper uses a multimodal prior in GANs and reconstructs the latents back from images in two stages to match the generated data modes to the latent space modes. It is empirically shown that this can prevent mode collapse to some extent (including intra-class collapse). However the paper lacks a comparison with state of the art GANs that have been shown to get better FID scores (~21 for SN-GAN [1] vs ~28 in the paper) so the benefit here is unclear, particularly in cases when the mode prior is unknown. Similarly for other applications used in the paper such as inference and attribute discovery, it falls short of demonstrating quantitative improvements with the approach. For example, there is a growing body of work on unsupervised disentanglement in generative models with several metrics to measure it, which could be used to evaluate the attribute discovery performance. R1 has brought up the point of lack of comparisons which the AC agrees with. Authors have made revisions in the paper including some comparisons but these feel insufficient to establish the benefits of the method over state of the art in preventing mode collapse. A borderline paper as reflected in the reviewer scores but can be made stronger with experiments showing convincing improvements over state of the art in at least one of the applications considered in the paper. [1] Miyato, T., Kataoka, T., Koyama, M., & Yoshida, Y. (2018). Spectral normalization for generative adversarial networks. ArXiv Preprint ArXiv:1802.05957."""
1222,"""Deep Imitative Models for Flexible Inference, Planning, and Control""","['imitation learning', 'forecasting', 'computer vision']","""Imitation learning provides an appealing framework for autonomous control: in many tasks, demonstrations of preferred behavior can be readily obtained from human experts, removing the need for costly and potentially dangerous online data collection in the real world. However, policies learned with imitation learning have limited flexibility to accommodate varied goals at test time. Model-based reinforcement learning (MBRL) offers considerably more flexibility, since a predictive model learned from data can be used to achieve various goals at test time. However, MBRL suffers from two shortcomings. First, the model does not help to choose desired or safe outcomes -- its dynamics estimate only what is possible, not what is preferred. Second, MBRL typically requires additional online data collection to ensure that the model is accurate in those situations that are actually encountered when attempting to achieve test time goals. Collecting this data with a partially trained model can be dangerous and time-consuming. In this paper, we aim to combine the benefits of imitation learning and MBRL, and propose imitative models: probabilistic predictive models able to plan expert-like trajectories to achieve arbitrary goals. We find this method substantially outperforms both direct imitation and MBRL in a simulated autonomous driving task, and can be learned efficiently from a fixed set of expert demonstrations without additional online data collection. We also show our model can flexibly incorporate user-supplied costs at test-time, can plan to sequences of goals, and can even perform well with imprecise goals, including goals on the wrong side of the road.""","""This paper proposes to combine RL and imitation learning, and the proposed approach seems convincing. As is typical in RL work, the evaluation of the method is not strong enough to convince the reviewers. Increasing community criticism on RL methods not scaling must be taken seriously here, despite the authors' disagreement. """
1223,"""Learning Partially Observed PDE Dynamics with Neural Networks""","['deep learning', 'spatio-temporal dynamics', 'physical processes', 'differential equations', 'dynamical systems']","""Spatio-Temporal processes bear a central importance in many applied scientific fields. Generally, differential equations are used to describe these processes. In this work, we address the problem of learning spatio-temporal dynamics with neural networks when only partial information on the system's state is available. Taking inspiration from the dynamical system approach, we outline a general framework in which complex dynamics generated by families of differential equations can be learned in a principled way. Two models are derived from this framework. We demonstrate how they can be applied in practice by considering the problem of forecasting fluid flows. We show how the underlying equations fit into our formalism and evaluate our method by comparing with standard baselines.""","""This paper introduces a few training methods to fit the dynamics of a PDE based on observations. Quality: Not great. The authors seem unaware of much related work both in the numerics and deep learning communities. The experiments aren't very illuminating, and the connections between the different methods are never clearly and explicitly laid out in one place. Clarity: Poor. The intro is long and rambly, and the main contributions aren't clearly motivated. A lot of time is spent mentioning things that could be done, without saying when this would be important or useful to do. An algorithm box or two would be a big improvement over the many long english explanations of the methods, and the diagrams with cycles in them. Originality: Not great. There has been a lot of work on fitting dynamics models using NNs, and also attempting to optimize PDE solvers, which is hardly engaged with. Significance: This work fails to make its own significance clear, by not exploring or explaining the scope and limitations of their proposed approach, or comparing against more baselines from the large set of related literature."""
1224,"""MisGAN: Learning from Incomplete Data with Generative Adversarial Networks""","['generative models', 'missing data']","""Generative adversarial networks (GANs) have been shown to provide an effective way to model complex distributions and have obtained impressive results on various challenging tasks. However, typical GANs require fully-observed data during training. In this paper, we present a GAN-based framework for learning from complex, high-dimensional incomplete data. The proposed framework learns a complete data generator along with a mask generator that models the missing data distribution. We further demonstrate how to impute missing data by equipping our framework with an adversarially trained imputer. We evaluate the proposed framework using a series of experiments with several types of missing data processes under the missing completely at random assumption.""","""The paper proposes an adversarial framework that learns a generative model along with a mask generator to model missing data and by this enables a GAN to learn from incomplete data. The method builds on AmbientGAN but it is a novel and clever adjustment to the specific problem setting of learning from incomplete data, that is of high practical interest."""
1225,"""Human Action Recognition Based on Spatial-Temporal Attention""",[],"""Many state-of-the-art methods of recognizing human action are based on attention mechanism, which shows the importance of attention mechanism in action recognition. With the rapid development of neural networks, human action recognition has been achieved great improvement by using convolutional neural networks (CNN) or recurrent neural networks (RNN). In this paper, we propose a model based on spatial-temporal attention weighted LSTM. This model pays attention to the key part in each video frame, and also focuses on the important frames in each video sequence, thus the most important theme for our model is how to find out the key point spatially and the key frames temporally. We show a feasible architecture which can solve those two problems effectively and achieve a satisfactory result. Our model is trained and tested on three datasets including UCF-11, UCF-101, and HMDB51. Those results demonstrate a high performance of our model in human action recognition.""","""Average score of 3.33, highest score of 4. The AC recommends rejection. """
1226,"""Learning a Neural-network-based Representation for Open Set Recognition""",['open set recognition'],"""In this paper, we present a neural network based representation for addressing the open set recognition problem. In this representation instances from the same class are close to each other while instances from different classes are further apart, resulting in statistically significant improvement when compared to other approaches on three datasets from two different domains. ""","""The paper presents an approach to address the open-set recognition task based on inter and intra class distances. All reviewers are concerned with novelty and more experimental comparisons. Authors have added some results, but reviewers did not think these were enough to make the paper convincing enough. Overall I agree with reviewers and recommend to reject the paper."""
1227,"""Generative Adversarial Self-Imitation Learning""",[],"""This paper explores a simple regularizer for reinforcement learning by proposing Generative Adversarial Self-Imitation Learning (GASIL), which encourages the agent to imitate past good trajectories via generative adversarial imitation learning framework. Instead of directly maximizing rewards, GASIL focuses on reproducing past good trajectories, which can potentially make long-term credit assignment easier when rewards are sparse and delayed. GASIL can be easily combined with any policy gradient objective by using GASIL as a learned reward shaping function. Our experimental results show that GASIL improves the performance of proximal policy optimization on 2D Point Mass and MuJoCo environments with delayed reward and stochastic dynamics.""","""The paper proposes an extension to reinforcement learning with self-imitation (SIL)[Oh et al. 2018]. It is based on the idea of leveraging previously encountered high-reward trajectories for reward shaping. This shaping is learned automatically using an adversarial setup, similar to GAIL [Ho & Ermon, 2016]. The paper clearly presents the proposed approach and relation to previous work. Empirical evaluation shows strong performance on a 2D point mass problem designed to examine the algorithms behavior. Of particular note are the insightful visualizations in Figure 2 and 3 which shed light on the algorithm's learning behavior. Empirical results on the Mujoco domain show that the proposed approach is particularly strong under delayed-reward (20 steps) and noisy-observation settings. The reviewers and AC note the following potential weaknesses: The paper presents an empirical validation showing improvements over PPO, in particular in Mujoco tasks with delayed rewards and with noisy observations. However, given the close relation to SIL, a direct comparison with the latter algorithm seems more appropriate. Reviewers 2 and 3 pointed out that the empirical validation of SIL was more extensive, including results on a wide range of Atari games. The authors provided results on several hard-exploration Atari games in the rebuttal period, but the results of the comparison to SIL were inconclusive. Given that the main contribution of the paper is empirical, the reviewers and the AC consider the contribution incremental. The reviewers noted that the proposed method was presented with little theoretical justification, which limits the contribution of the paper. During the rebuttal phase, the authors sketched a theoretical argument in their rebuttal, but noted that they are not able to provide a guarantee that trajectories in the replay buffer constitute an unbiased sample from the optimal policy, and that policy gradient methods in general are not guaranteed to converge to a globally optimal policy. The AC notes that conceptual insights can also be provided by motivating algorithmic or modeling choices, or through detailed analysis of the obtained results with the goal to further understanding of the observed behavior. Any such form of developing further insights would strengthen the contribution of the submission."""
1228,"""An Active Learning Framework for Efficient Robust Policy Search""",['Deep Reinforcement Learning'],"""Robust Policy Search is the problem of learning policies that do not degrade in performance when subject to unseen environment model parameters. It is particularly relevant for transferring policies learned in a simulation environment to the real world. Several existing approaches involve sampling large batches of trajectories which reflect the differences in various possible environments, and then selecting some subset of these to learn robust policies, such as the ones that result in the worst performance. We propose an active learning based framework, EffAcTS, to selectively choose model parameters for this purpose so as to collect only as much data as necessary to select such a subset. We apply this framework to an existing method, namely EPOpt, and experimentally validate the gains in sample efficiency and the performance of our approach on standard continuous control tasks. We also present a Multi-Task Learning perspective to the problem of Robust Policy Search, and draw connections from our proposed framework to existing work on Multi-Task Learning.""","""The paper addresses sample-efficient robust policy search borrowing ideas from active learning. The reviews raised important concerns regarding (1) the complexity of the proposed technique, which combines many separate pieces and (2) the significance of experimental results. The empirical setup adopted is not standard in RL, and a clear comparison against EPOpt is lacking. I appreciate the changes made to address the comment, and I encourage the authors to continue improving the paper by simplifying the model and including a few baseline comparisons in the experiments."""
1229,"""Stop memorizing: A data-dependent regularization framework for intrinsic pattern learning""","['deep neural networks', 'memorizing', 'data-dependent regularization']","""Deep neural networks (DNNs) typically have enough capacity to fit random data by brute force even when conventional data-dependent regularizations focusing on the geometry of the features are imposed. We find out that the reason for this is the inconsistency between the enforced geometry and the standard softmax cross entropy loss. To resolve this, we propose a new framework for data-dependent DNN regularization, the Geometrically-Regularized-Self-Validating neural Networks (GRSVNet). During training, the geometry enforced on one batch of features is simultaneously validated on a separate batch using a validation loss consistent with the geometry. We study a particular case of GRSVNet, the Orthogonal-Low-rank Embedding (OLE)-GRSVNet, which is capable of producing highly discriminative features residing in orthogonal low-rank subspaces. Numerical experiments show that OLE-GRSVNet outperforms DNNs with conventional regularization when trained on real data. More importantly, unlike conventional DNNs, OLE-GRSVNet refuses to memorize random data or random labels, suggesting it only learns intrinsic patterns by reducing the memorizing capacity of the baseline DNN.""","""The paper proposes an interesting data-dependent regularization method for orthogonal-low-rank embedding (OLE). Despite the novelty of the method, the reviewers and AC note that it's unclear whether the approach can extend other settings with multi-class or continuous labels or other loss functions. """
1230,"""Learning to Separate Domains in Generalized Zero-Shot and Open Set Learning: a probabilistic perspective""","['Generalized zero-shot learning', 'domain division', 'bootstrapping', 'Kolmogorov-Smirnov']","""This paper studies the problem of domain division which aims to segment instances drawn from different probabilistic distributions. This problem exists in many previous recognition tasks, such as Open Set Learning (OSL) and Generalized Zero-Shot Learning (G-ZSL), where the testing instances come from either seen or unseen/novel classes with different probabilistic distributions. Previous works only calibrate the condent prediction of classiers of seen classes (WSVM Scheirer et al. (2014)) or taking unseen classes as outliers Socher et al. (2013). In contrast, this paper proposes a probabilistic way of directly estimating and ne-tuning the decision boundary between seen and unseen classes. In particular, we propose a domain division algorithm to split the testing instances into known, unknown and uncertain domains, and then conduct recognition tasks in each domain. Two statistical tools, namely, bootstrapping and KolmogorovSmirnov (K-S) Test, for the rst time, are introduced to uncover and ne-tune the decision boundary of each domain. Critically, the uncertain domain is newly introduced in our framework to adopt those instances whose domain labels cannot be predicted condently. Extensive experiments demonstrate that our approach achieved the state-of-the-art performance on OSL and G-ZSL benchmarks.""","""AR1 finds the paper overly lengthy and ill-focused on contributions of this work. Moreover, AR1 would like to see more results for G-ZSL. AR2 finds the paper is lacking in clarity, e.g. Eq. 9, and complete definition of the end-to-end decision pipeline is missing. AR2 points that the manuscript relies on GZSL and comparisons to it but other more recent methods could be also cited: - Generalized Zero-Shot Learning via Synthesized Examples by Verma et al. - Zero-Shot Kernel Learning by Zhang et al. - Model Selection for Generalized Zero-shot Learning by Zhang et al. - Generalized Zero-Shot Learning with Deep Calibration Network by Liu et al. - Multi-modal Cycle-consistent Generalized Zero-Shot Learning by Felix et al. - Open Set Learning with Counterfactual Images - Feature Generating Networks for Zero-Shot Learning Though, the authors are welcome to find even more relevant papers in google scholar. Overall, AC finds the paper interesting and finds the idea has some merits. Nonetheless, two reviewers maintained their scores below borderline due to numerous worries highlighted above. The authors are encouraged to work on presentation of this method and comparisons to more recent papers where possible. AC encourages the authors to re-submit their improved manuscript as, at this time, it feels this paper is not ready and cannot be accepted to ICLR. """
1231,"""Importance Resampling for Off-policy Policy Evaluation""","['Reinforcement Learning', 'Off-policy policy evaluation', 'importance resampling', 'importance sampling']","""Importance sampling is a common approach to off-policy learning in reinforcement learning. While it is consistent and unbiased, it can result in high variance updates to the parameters for the value function. Weighted importance sampling (WIS) has been explored to reduce variance for off-policy policy evaluation, but only for linear value function approximation. In this work, we explore a resampling strategy to reduce variance, rather than a reweighting strategy. We propose Importance Resampling (IR) for off-policy learning, that resamples experience from the replay buffer and applies a standard on-policy update. The approach avoids using importance sampling ratios directly in the update, instead correcting the distribution over transitions before the update. We characterize the bias and consistency of the our estimator, particularly compared to WIS. We then demonstrate in several toy domains that IR has improved sample efficiency and parameter sensitivity, as compared to several baseline WIS estimators and to IS. We conclude with a demonstration showing IR improves over IS for learning a value function from images in a racing car simulator.""","""The paper proposes to use importance resampling (IR) as an alternative to the more popular importance sampling (IS) approach to off-policy RL. The hope is to reduce variance, as shown in experiments. However, there is no analysis why/when IR will be better than IS for variance reduction, and a few baselines were suggested by reviewers. While the authors rebuttal was helpful in clarifying several issues, the overall contribution does not seem strong enough for ICLR, on both theoretical and empirical sides. The high variance of IS is known, and the following work may be referenced for better 1st order updates when IS weights are used: Karampatziakis & Langford (UAI'11). In section 3, the paper says that most off-policy work uses d_mu, instead of d_pi, to weigh states. This is true, but in the current context (infinite-horizon RL), there are more recent works that should probably be referenced: pseudo-url pseudo-url"""
1232,"""Volumetric Convolution: Automatic Representation Learning in Unit Ball""","['convolution', 'unit sphere', '3D object recognition']","""Convolution is an efficient technique to obtain abstract feature representations using hierarchical layers in deep networks. Although performing convolution in Euclidean geometries is fairly straightforward, its extension to other topological spaces---such as a sphere S^2 or a unit ball B^3---entails unique challenges. In this work, we propose a novel `""volumetric convolution"" operation that can effectively convolve arbitrary functions in B^3. We develop a theoretical framework for ""volumetric convolution"" based on Zernike polynomials and efficiently implement it as a differentiable and an easily pluggable layer for deep networks. Furthermore, our formulation leads to derivation of a novel formula to measure the symmetry of a function in B^3 around an arbitrary axis, that is useful in 3D shape analysis tasks. We demonstrate the efficacy of proposed volumetric convolution operation on a possible use-case i.e., 3D object recognition task.""","""Using volumetric convolutions, this paper focuses on learning in (rather than on) the unit sphere. The novelty of the approach is debatable, and the mathematical analysis not strong enough to merit that. In combination with good but not outstanding results, interest of the research community is doubted. An extended experimental analysis of the method would greatly improve the paper."""
1233,"""Feature quantization for parsimonious and interpretable predictive models""","['discretization', 'grouping', 'interpretability', 'shallow neural networks']","""For regulatory and interpretability reasons, the logistic regression is still widely used by financial institutions to learn the refunding probability of a loan from applicant's historical data. To improve prediction accuracy and interpretability, a preprocessing step quantizing both continuous and categorical data is usually performed: continuous features are discretized by assigning factor levels to intervals and, if numerous, levels of categorical features are grouped. However, a better predictive accuracy can be reached by embedding this quantization estimation step directly into the predictive estimation step itself. By doing so, the predictive loss has to be optimized on a huge and untractable discontinuous quantization set. To overcome this difficulty, we introduce a specific two-step optimization strategy: first, the optimization problem is relaxed by approximating discontinuous quantization functions by smooth functions; second, the resulting relaxed optimization problem is solved via a particular neural network and stochastic gradient descent. The strategy gives then access to good candidates for the original optimization problem after a straightforward maximum a posteriori procedure to obtain cutpoints. The good performances of this approach, which we call glmdisc, are illustrated on simulated and real data from the UCI library and Crdit Agricole Consumer Finance (a major European historic player in the consumer credit market). The results show that practitioners finally have an automatic all-in-one tool that answers their recurring needs of quantization for predictive tasks.""","""All reviewers agree to reject. While there were many positive points to this work, reviewers believed that it was not yet ready for acceptance."""
1234,"""SupportNet: solving catastrophic forgetting in class incremental learning with support data""",[],"""A plain well-trained deep learning model often does not have the ability to learn new knowledge without forgetting the previously learned knowledge, which is known as catastrophic forgetting. Here we propose a novel method, SupportNet, to efficiently and effectively solve the catastrophic forgetting problem in the class incremental learning scenario. SupportNet combines the strength of deep learning and support vector machine (SVM), where SVM is used to identify the support data from the old data, which are fed to the deep learning model together with the new data for further training so that the model can review the essential information of the old data when learning the new information. Two powerful consolidation regularizers are applied to stabilize the learned representation and ensure the robustness of the learned model. We validate our method with comprehensive experiments on various tasks, which show that SupportNet drastically outperforms the state-of-the-art incremental learning methods and even reaches similar performance as the deep learning model trained from scratch on both old and new data.""","""The authors propose using a SVM, trained as a last layer of a neural network, to identify exemplars (support vectors) to save and use to prevent forgetting as the model is trained on further tasks. The method is effective on several supervised benchmarks and is compared to several other methods, including VCL, iCARL, and GEM. The reviewers had various objections to the initial paper that centered around comparisons to other methods and reporting of detailed performance numbers, which the authors resolved convincingly in their revised paper. However, the AC and 2 of the reviewers were unconvinced of the contribution of the approach. Although no one has used this particular strategy, of using support vectors to prevent forgetting, the approach is a simplistic composition of the NN and the SVM which is heuristic, at least in how the authors present it. Most importantly, the approach is limited to supervised classification problems, yet catastrophic forgetting is not commonly considered to be a problem for the supervised classifier setting; rather it is a problem for inherently sequential learning environments such as RL (MNIST and CIFAR are just commonly used in the literature for ease of evaluation)."""
1235,"""Revisiting Reweighted Wake-Sleep""","['variational inference', 'approximate inference', 'generative models', 'gradient estimators']",""" Discrete latent-variable models, while applicable in a variety of settings, can often be difficult to learn. Sampling discrete latent variables can result in high-variance gradient estimators for two primary reasons: 1) branching on the samples within the model, and 2) the lack of a pathwise derivative for the samples. While current state-of-the-art methods employ control-variate schemes for the former and continuous-relaxation methods for the latter, their utility is limited by the complexities of implementing and training effective control-variate schemes and the necessity of evaluating (potentially exponentially) many branch paths in the model. Here, we revisit the Reweighted Wake Sleep (RWS; Bornschein and Bengio, 2015) algorithm, and through extensive evaluations, show that it circumvents both these issues, outperforming current state-of-the-art methods in learning discrete latent-variable models. Moreover, we observe that, unlike the Importance-weighted Autoencoder, RWS learns better models and inference networks with increasing numbers of particles, and that its benefits extend to continuous latent-variable models as well. Our results suggest that RWS is a competitive, often preferable, alternative for learning deep generative models.""","""The paper presents a well conducted empirical study of the Reweighted Wake Sleep (RWS) algorithm (Bornschein and Bengio, 2015). It shows that it performs consistently better than alternatives such as Importance Weighted Autoencoder (IWAE) for the hard problem of learning deep generative models with discrete latent variables acting as a stochastic control flow. The work is well-written and extracts valuable insights supported by empirical observations: in particular the fact that increasing the number of particles improves learning in RWS but hurts in IWAE, and the fact that RWS can also be successfully applied to continuous variables. The reviewers and AC note the following weaknesses of the work as it currently stands: a) it is almost exclusively empirical and while reasonable explanations are argued, it does not provide a formal theoretical analysis justifying the observed behaviour b) experiments are limited to MNIST and synthetic data, confirmation of the findings on larger-scale real-world data and model would provide a more complete and convincing evidence. The paper should be made stronger on at least one (and ideally both) of these accounts. """
1236,"""Deep Ensemble Bayesian Active Learning : Adressing the Mode Collapse issue in Monte Carlo dropout via Ensembles""","['Active Learning', 'Deep Learning', 'Bayesian Neural Networks', 'Bayesian Deep Learning', 'Ensembles']","""In image classification tasks, the ability of deep convolutional neural networks (CNNs) to deal with complex image data has proved to be unrivalled. Deep CNNs, however, require large amounts of labeled training data to reach their full potential. In specialised domains such as healthcare, labeled data can be difficult and expensive to obtain. One way to alleviate this problem is to rely on active learning, a learning technique that aims to reduce the amount of labelled data needed for a specific task while still delivering satisfactory performance. We propose a new active learning strategy designed for deep neural networks. This method improves upon the current state-of-the-art deep Bayesian active learning method, which suffers from the mode collapse problem. We correct for this deficiency by making use of the expressive power and statistical properties of model ensembles. Our proposed method manages to capture superior data uncertainty, which translates into improved classification performance. We demonstrate empirically that our ensemble method yields faster convergence of CNNs trained on the MNIST and CIFAR-10 datasets.""","""The reviewers in general found the paper approachable, well written and clear. They noted that the empirical observation of mode collapse in active learning was an interesting insight. However, all the reviewers had concerns with novelty, particularly in light of Lakshminarayanan et al. who also train ensembles to get a measure of uncertainty. An interesting addition to the paper might be some theoretical insight about what the model corresponds to when one ensembles multiple models from MC Dropout. One reviewer noted that it's not clear that the ensemble is capturing the desired posterior. As a note, I don't believe there is agreement in the community that MC dropout is state-of-the-art in terms of capturing uncertainty for deep neural networks, as argued in the author response (and the abstract). To the contrary, I believe a variety of papers have improved over the results from that work (e.g. see experiments in Multiplicative Normalizing Flows from over a year ago)."""
1237,"""Stochastic Quantized Activation: To prevent Overfitting in Fast Adversarial Training""","['adversarial examples', 'deep learning']","""Existing neural networks are vulnerable to ""adversarial examples""---created by adding maliciously designed small perturbations in inputs to induce a misclassification by the networks. The most investigated defense strategy is adversarial training which augments training data with adversarial examples. However, applying single-step adversaries in adversarial training does not support the robustness of the networks, instead, they will even make the networks to be overfitted. In contrast to the single-step, multi-step training results in the state-of-the-art performance on MNIST and CIFAR10, yet it needs a massive amount of time. Therefore, we propose a method, Stochastic Quantized Activation (SQA) that solves overfitting problems in single-step adversarial training and fastly achieves the robustness comparable to the multi-step. SQA attenuates the adversarial effects by providing random selectivity to activation functions and allows the network to learn robustness with only single-step training. Throughout the experiment, our method demonstrates the state-of-the-art robustness against one of the strongest white-box attacks as PGD training, but with much less computational cost. Finally, we visualize the learning process of the network with SQA to handle strong adversaries, which is different from existing methods.""","""While the paper contains interesting ideas, the reviewers suggest improving the clarity and experimental study of the paper. The work holds promises but is not ready for publication at ICLR."""
1238,"""ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness""","['deep learning', 'psychophysics', 'representation learning', 'object recognition', 'robustness', 'neural networks', 'data augmentation']","""Convolutional Neural Networks (CNNs) are commonly thought to recognise objects by learning increasingly complex representations of object shapes. Some recent studies suggest a more important role of image textures. We here put these conflicting hypotheses to a quantitative test by evaluating CNNs and human observers on images with a texture-shape cue conflict. We show that ImageNet-trained CNNs are strongly biased towards recognising textures rather than shapes, which is in stark contrast to human behavioural evidence and reveals fundamentally different classification strategies. We then demonstrate that the same standard architecture (ResNet-50) that learns a texture-based representation on ImageNet is able to learn a shape-based representation instead when trained on 'Stylized-ImageNet', a stylized version of ImageNet. This provides a much better fit for human behavioural performance in our well-controlled psychophysical lab setting (nine experiments totalling 48,560 psychophysical trials across 97 observers) and comes with a number of unexpected emergent benefits such as improved object detection performance and previously unseen robustness towards a wide range of image distortions, highlighting advantages of a shape-based representation.""","""This paper proposes a hypothesis about the kinds of visual information for which popular neural networks are most selective. It then proposes a series of empirical experiments on synthetically modified training sets to test this and related hypotheses. The main conclusions of the paper are contained in the title, and the presentation was consistently rated as very clear. As such, it is both interesting to a relatively wide audience and accessible. Although the paper is comparatively limited in theoretical or algorithmic contribution, the empirical results and experimental design are of sufficient quality to inform design choices of future neural networks, and to better understand the reasons for their current behavior. The reviewers were unanimous in their appreciation of the contributions, and all recommended that the paper be accepted. """
1239,"""Where Off-Policy Deep Reinforcement Learning Fails""","['reinforcement learning', 'off-policy', 'imitation', 'batch reinforcement learning']","""This work examines batch reinforcement learning--the task of maximally exploiting a given batch of off-policy data, without further data collection. We demonstrate that due to errors introduced by extrapolation, standard off-policy deep reinforcement learning algorithms, such as DQN and DDPG, are only capable of learning with data correlated to their current policy, making them ineffective for most off-policy applications. We introduce a novel class of off-policy algorithms, batch-constrained reinforcement learning, which restricts the action space to force the agent towards behaving on-policy with respect to a subset of the given data. We extend this notion to deep reinforcement learning, and to the best of our knowledge, present the first continuous control deep reinforcement learning algorithm which can learn effectively from uncorrelated off-policy data.""","""The paper proposes batch-constrained approach to batch RL, where the policy is optimized under the constrain that at a state only actions appearing in the training data are allowed. An extension to continuous cases is given. While the paper has some interesting idea and the problem of dealing with extrapolation in RL is important, the approach appears somewhat ad hoc and the contributions limited. For example, the constraint is based on whether (s,a) is in B, but this condition can be quite delicate in a stochastic problem (seeing a in s *once* may still allow large extrapolation error if that only observed transition is not representative). Section 4.1 gives some nice insights for the special finite MDP case, but those results are a little weak (requiring strong assumption that may not hold in practice) --- an example being the requirement that s' be included in data if (s,a) is in data and P(s'|s,a)>0 [beginning of section 4.1]. In contrast, there are other more robust and principled ways, such as counterfactual risk minimization (CRM) for contextual bandits (pseudo-url). For MDPs, the Bayesian version of DQN (the cited Azizzadenesheli et al., as well as Lipton et al. at AAAI'18) can be used to constrain the learned policy as well, with a simple modification of using the CRM idea for bandits. Would these algorithms be reasonable baselines?"""
1240,"""Deterministic PAC-Bayesian generalization bounds for deep networks via generalizing noise-resilience""","['generalization', 'PAC-Bayes', 'SGD', 'learning theory', 'implicit regularization']","""The ability of overparameterized deep networks to generalize well has been linked to the fact that stochastic gradient descent (SGD) finds solutions that lie in flat, wide minima in the training loss -- minima where the output of the network is resilient to small random noise added to its parameters. So far this observation has been used to provide generalization guarantees only for neural networks whose parameters are either \textit{stochastic} or \textit{compressed}. In this work, we present a general PAC-Bayesian framework that leverages this observation to provide a bound on the original network learned -- a network that is deterministic and uncompressed. What enables us to do this is a key novelty in our approach: our framework allows us to show that if on training data, the interactions between the weight matrices satisfy certain conditions that imply a wide training loss minimum, these conditions themselves {\em generalize} to the interactions between the matrices on test data, thereby implying a wide test loss minimum. We then apply our general framework in a setup where we assume that the pre-activation values of the network are not too small (although we assume this only on the training data). In this setup, we provide a generalization guarantee for the original (deterministic, uncompressed) network, that does not scale with product of the spectral norms of the weight matrices -- a guarantee that would not have been possible with prior approaches.""","""Existing PAC Bayes analysis gives generalization bounds for stochastic networks/classifiers. This paper develops a new approach to obtain generalization bounds for the original network, by generalizing noise resilience property from training data to test data. All reviewers agree that the techniques developed in the paper (namely Theorem 3.1) are novel and interesting. There was disagreement between reviewers on the usefulness of the new generalization bound (Theorem 4.1) shown in this paper using the above techniques. I believe authors have sufficiently addressed these concerns in their response and updated draft. Hence, despite the concerns of R3 on limitations of this bound and its dependence on pre-activation values, I agree with R2 and R4 that the techniques developed in the paper are of interest to the community and deserve publication. I suggest authors to keep comments of R3 in mind while preparing the final version. """
1241,"""Whitening and Coloring Batch Transform for GANs""","['Generative Adversarial Networks', 'conditional GANs', 'Batch Normalization']","""Batch Normalization (BN) is a common technique used to speed-up and stabilize training. On the other hand, the learnable parameters of BN are commonly used in conditional Generative Adversarial Networks (cGANs) for representing class-specific information using conditional Batch Normalization (cBN). In this paper we propose to generalize both BN and cBN using a Whitening and Coloring based batch normalization. We show that our conditional Coloring can represent categorical conditioning information which largely helps the cGAN qualitative results. Moreover, we show that full-feature whitening is important in a general GAN scenario in which the training process is known to be highly unstable. We test our approach on different datasets and using different GAN networks and training protocols, showing a consistent improvement in all the tested frameworks. Our CIFAR-10 conditioned results are higher than all previous works on this dataset.""","""The paper addresses normalisation and conditioning of GANs. The authors propose to replace class-conditional batch norm with whitening and class-conditional coloring. Evaluation demonstrates that the method performs very well, and the ablation studies confirm the design choices. After extensive discussion, all reviewers agreed that this is a solid contribution, and the paper should be accepted. """
1242,"""PCNN: Environment Adaptive Model Without Finetuning""","['Class skew', 'Runtime adaption']","""Convolutional Neural Networks (CNNs) have achieved tremendous success for many computer vision tasks, which shows a promising perspective of deploying CNNs on mobile platforms. An obstacle to this promising perspective is the tension between intensive resource consumption of CNNs and limited resource budget on mobile platforms. Existing works generally utilize a simpler architecture with lower accuracy for a higher energy-efficiency, \textit{i.e.}, trading accuracy for resource consumption. An emerging opportunity to both increasing accuracy and decreasing resource consumption is \textbf{class skew}, \textit{i.e.}, the strong temporal and spatial locality of the appearance of classes. However, it is challenging to efficiently utilize the class skew due to both the frequent switches and the huge number of class skews. Existing works use transfer learning to adapt the model towards the class skew during runtime, which consumes resource intensively. In this paper, we propose \textbf{probability layer}, an \textit{easily-implemented and highly flexible add-on module} to adapt the model efficiently during runtime \textit{without any fine-tuning} and achieving an \textit{equivalent or better} performance than transfer learning. Further, both \textit{increasing accuracy} and \textit{decreasing resource consumption} can be achieved during runtime through the combination of probability layer and pruning methods.""","""All reviewers rate the paper as below threshold. While the authors responded to an earlier request for clarification, there is no rebuttal to the actual reviews. Thus, there is no basis by which the paper can be accepted."""
1243,"""Learning to Represent Edits""","['Representation Learning', 'Source Code', 'Natural Language', 'edit']","""We introduce the problem of learning distributed representations of edits. By combining a ""neural editor"" with an ""edit encoder"", our models learn to represent the salient information of an edit and can be used to apply edits to new inputs. We experiment on natural language and source code edit data. Our evaluation yields promising results that suggest that our neural network models learn to capture the structure and semantics of edits. We hope that this interesting task and data source will inspire other researchers to work further on this problem.""","""This paper investigates learning to represent edit operations for two domains: text and source code. The primary contributions of the paper are in the specific task formulation and the new dataset (for source code edits). The technical novelty is relatively weak. Pros: The paper introduces a new dataset for source code edits. Cons: Reviewers raised various concerns about human evaluation and many other experimental details, most of which the rebuttal have successfully addressed. As a result, R3 updated their score from 4 to 6. Verdict: Possible weak accept. None of the remaining issues after the rebuttal is a serious deal breaker (e.g., task simplification by assuming the knowledge of when and where the edit must be applied, simplifying the real-world application of the automatic edits). However, the overall impact and novelty of the paper is relatively weak."""
1244,"""Countering Language Drift via Grounding""","['grounding', 'policy gradient', 'language drift', 'reinforcement learning']","""While reinforcement learning (RL) shows a lot of promise for natural language processinge.g. when fine-tuning natural language systems for optimizing a certain objectivethere has been little investigation into potential language drift: when an external reward is used to train a system, the agents communication protocol may easily and radically diverge from natural language. By re-casting translation as a communication game, we show that language drift indeed happens when pre-trained agents are fine-tuned with policy gradient methods. We contend that simply adding a ""naturalness"" constraint to the reward, e.g. by using language model log likelihood, does not fully address the issue, and argue that (perceptual) grounding is required. That is, while language model constraints impose syntactic conformity, they do not lead to semantic correspondence. Our experiments show that grounded models give the best communication performance, while retaining English syntax along with the ability to convey the intended semantics.""","""This paper proposes a method to resolve ""language drift,"" where a pre-trained X->language model trained in an X->language->Y pipeline drifts away from being natural language. In particular, it proposes to add an auxiliary training objective that performs grounding with multimodal input to fix this problem. Results are good on a task where translation is done between two languages. The main concern that was raised with this paper by most of the reviewers is the validity of the proposed task itself. Even after extensive discussion with the authors, it is not clear that there is a very convincing scenario where we both have a pre-trained X->language, care about the intermediate results, and have some sort of grounded input to fix this drift. While I do understand the MT task is supposed to be a testbed for the true objective, it feel it is necessary to additionally have one convincing use case where this is a real problem and not just the artificially contrived. This use case could either be of practical use (e.g. potentially useful in an application), or of interest from the point of view of cognitive plausibility (e.g. similar to how children actually learn, and inspired by cognitive science literature). A concern that offshoots from this is that because the underlying idea is compelling (some sort of grounding to inform language learning), a paper at a high-profile conference such as ICLR may help re-popularize this line of research, which has been a niche for a while. Normally I would say this is definitely a good thing; I think considering grounding in language learning is definitely an important research direction, and have been a fan of this line of work since reading Roy's seminal work on it from 15 years ago. However, if the task used in this paper, which is of questionable value and reality, becomes the benchmark for this line of work I think this might lead other follow-up work in the wrong direction. I feel that this is a critical issue, and the paper will be much stronger after a more realistic task setting is added. Thus, I am not recommending acceptance at this time, but would definitely like the authors to think hard and carefully about a good and realistic benchmark for the task, and follow up with a revised version of the paper in the future."""
1245,"""ChainGAN: A sequential approach to GANs""","['Machine Learning', 'Sequential Models', 'GANs']","""We propose a new architecture and training methodology for generative adversarial networks. Current approaches attempt to learn the transformation from a noise sample to a generated data sample in one shot. Our proposed generator architecture, called ChainGAN, uses a two-step process. It first attempts to transform a noise vector into a crude sample, similar to a traditional generator. Next, a chain of networks, called editors, attempt to sequentially enhance this sample. We train each of these units independently, instead of with end-to-end backpropagation on the entire chain. Our model is robust, efficient, and flexible as we can apply it to various network architectures. We provide rationale for our choices and experimentally evaluate our model, achieving competitive results on several datasets.""","""The paper presents a GAN-based generative model, where the generator consists of the base generator followed by several editors, each trained separately with its own discriminator. The reviewers found the idea interesting, but the evaluation insufficient. No rebuttal was provided."""
1246,"""A unified theory of adaptive stochastic gradient descent as Bayesian filtering""","['SGD', 'Bayesian', 'RMSprop', 'Adam']","""We formulate stochastic gradient descent (SGD) as a novel factorised Bayesian filtering problem, in which each parameter is inferred separately, conditioned on the corresopnding backpropagated gradient. Inference in this setting naturally gives rise to BRMSprop and BAdam: Bayesian variants of RMSprop and Adam. Remarkably, the Bayesian approach recovers many features of state-of-the-art adaptive SGD methods, including amongst others root-mean-square normalization, Nesterov acceleration and AdamW. As such, the Bayesian approach provides one explanation for the empirical effectiveness of state-of-the-art adaptive SGD algorithms. Empirically comparing BRMSprop and BAdam with naive RMSprop and Adam on MNIST, we find that Bayesian methods have the potential to considerably reduce test loss and classification error.""","""The aim of this paper is to interpret various optimizers such as RMSprop, Adam, and NAG, as approximate Kalman filtering of the optimal parameters. These algorithms are derived as inference procedures in various dynamical systems. The main empirical result is the algorithms achieve slightly better test accuracy on MNIST compared to an unregularized network trained with Adam or RMSprop. This was a controversial paper, and each of the reviewers had a significant back-and-forth with the authors. The controversy reflects that this is a pretty interesting and relevant topic: a proper Bayesian framework could provide significant guidance for developing better optimizers and regularizers. Unfortunately, I don't think this paper delivers on its promise of a unifying Bayesian framework for these various methods, and I don't think it's quite ready for publication at ICLR. There was some controversy about relationships to various recently published papers giving Bayesian interpretations of optimizers. The authors believe the added value of this submission is that it recovers features such as momentum and root-mean-square normalization. This would be a very interesting contribution beyond those works. But R2 and R3 feel like these particular features were derived using fairly ad-hoc assumptions or approximations almost designed to obtain existing algorithms, and from reading the paper I have to say I agree with the reviewers. There was a lot of back-and-forth about the correctness of various theoretical claims. But overall, my impression is that the theoretical arguments in this paper exceed the bar for a primarily practical/empirical paper, but aren't rigorous enough for the paper to stand purely based on the theoretical contributions. Unfortunately, the empirical part of the paper is rather lacking. The only experiment reported is on MNIST, and the only result is improved test error. The baseline gets below 99% test accuracy, below the level achieved by the original LeNet, suggesting the baseline may be somehow broken. Simply measuring test error doesn't really get at the benefits of Bayesian approaches, as it doesn't distinguish it from the many other regularizers that have been proposed. Since the proposed method is nearly identical to things like Adam or NAG, I don't see any reason it can't be evaluated on more challenging problems (as reviewers have asked for). Overall, while I find the ideas promising, I think the paper needs considerable work before it is ready for publication at ICLR. """
1247,"""Hierarchical Reinforcement Learning via Advantage-Weighted Information Maximization""","['Hierarchical reinforcement learning', 'Representation learning', 'Continuous control']","""Real-world tasks are often highly structured. Hierarchical reinforcement learning (HRL) has attracted research interest as an approach for leveraging the hierarchical structure of a given task in reinforcement learning (RL). However, identifying the hierarchical policy structure that enhances the performance of RL is not a trivial task. In this paper, we propose an HRL method that learns a latent variable of a hierarchical policy using mutual information maximization. Our approach can be interpreted as a way to learn a discrete and latent representation of the state-action space. To learn option policies that correspond to modes of the advantage function, we introduce advantage-weighted importance sampling. In our HRL method, the gating policy learns to select option policies based on an option-value function, and these option policies are optimized based on the deterministic policy gradient method. This framework is derived by leveraging the analogy between a monolithic policy in standard RL and a hierarchical policy in HRL by using a deterministic option policy. Experimental results indicate that our HRL approach can learn a diversity of options and that it can enhance the performance of RL in continuous control tasks.""","""This paper proposes a method for hierarchical reinforcement learning that aims to maximize mutual information between options and state-action pairs. The approach and empirical analysis is interesting. The initial submission had many issues with clarity. However, the new revisions of the paper have significantly improved the clarity, better describing the idea and improving the terminology. The main remaining weakness is the scope of the experimental results. However, the reviewers agree that the paper exceeds the bar for publication at ICLR with the existing experiments."""
1248,"""Language Modeling Teaches You More Syntax than Translation Does: Lessons Learned Through Auxiliary Task Analysis""","['representation learning', 'recurrent neural networks', 'syntax', 'part-of-speech tagging']","""Recent work using auxiliary prediction task classifiers to investigate the properties of LSTM representations has begun to shed light on why pretrained representations, like ELMo (Peters et al., 2018) and CoVe (McCann et al., 2017), are so beneficial for neural language understanding models. We still, though, do not yet have a clear understanding of how the choice of pretraining objective affects the type of linguistic information that models learn. With this in mind, we compare four objectives - language modeling, translation, skip-thought, and autoencoding - on their ability to induce syntactic and part-of-speech information. We make a fair comparison between the tasks by holding constant the quantity and genre of the training data, as well as the LSTM architecture. We find that representations from language models consistently perform best on our syntactic auxiliary prediction tasks, even when trained on relatively small amounts of data. These results suggest that language modeling may be the best data-rich pretraining task for transfer learning applications requiring syntactic information. We also find that the representations from randomly-initialized, frozen LSTMs perform strikingly well on our syntactic auxiliary tasks, but this effect disappears when the amount of training data for the auxiliary tasks is reduced.""","""Strengths: -- Solid experiments -- The paper is well written Weaknesses: -- The findings are not entirely novel and not so surprising, previous papers (e.g., Brevlins et al (ACL 2018)) have already suggested that LM objectives are preferable and also using LM objective for pretraining is already the standard practice (see details in R1 and R3). There is a consensus between the two reviewers who provided detailed comments and engaged in discussion with the authors. """
1249,"""Manifold Mixup: Learning Better Representations by Interpolating Hidden States""","['Regularizer', 'Supervised Learning', 'Semi-supervised Learning', 'Better representation learning', 'Deep Neural Networks.']","""Deep networks often perform well on the data distribution on which they are trained, yet give incorrect (and often very confident) answers when evaluated on points from off of the training distribution. This is exemplified by the adversarial examples phenomenon but can also be seen in terms of model generalization and domain shift. Ideally, a model would assign lower confidence to points unlike those from the training distribution. We propose a regularizer which addresses this issue by training with interpolated hidden states and encouraging the classifier to be less confident at these points. Because the hidden states are learned, this has an important effect of encouraging the hidden states for a class to be concentrated in such a way so that interpolations within the same class or between two different classes do not intersect with the real data points from other classes. This has a major advantage in that it avoids the underfitting which can result from interpolating in the input space. We prove that the exact condition for this problem of underfitting to be avoided by Manifold Mixup is that the dimensionality of the hidden states exceeds the number of classes, which is often the case in practice. Additionally, this concentration can be seen as making the features in earlier layers more discriminative. We show that despite requiring no significant additional computation, Manifold Mixup achieves large improvements over strong baselines in supervised learning, robustness to single-step adversarial attacks, semi-supervised learning, and Negative Log-Likelihood on held out samples.""","""The paper contains useful information and shows relative improvements compared to mixup. However, some of the main claims are not substantiated enough to be fully convincing. For example, the claims that manifold mixup can prevent can manifold collision issue where the interpolation between two samples collides with a sample from other class is incorrect. The authors are encouraged to incorporate remarks of the reviewers."""
1250,"""Decoupling Gating from Linearity""","['Artificial Neural Networks', 'Neural Networks', 'ReLU', 'GaLU', 'Deep Learning']","""The gap between the empirical success of deep learning and the lack of strong theoretical guarantees calls for studying simpler models. By observing that a ReLU neuron is a product of a linear function with a gate (the latter determines whether the neuron is active or not), where both share a jointly trained weight vector, we propose to decouple the two. We introduce GaLU networks networks in which each neuron is a product of a Linear Unit, defined by a weight vector which is being trained, with a Gate, defined by a different weight vector which is not being trained. Generally speaking, given a base model and a simpler version of it, the two parameters that determine the quality of the simpler version are whether its practical performance is close enough to the base model and whether it is easier to analyze it theoretically. We show that GaLU networks perform similarly to ReLU networks on standard datasets and we initiate a study of their theoretical properties, demonstrating that they are indeed easier to analyze. We believe that further research of GaLU networks may be fruitful for the development of a theory of deep learning.""","""The reviewers reached a consensus that the paper is not ready for publication in ICLR. (see more details in the reviews below. )"""
1251,"""Looking for ELMo's friends: Sentence-Level Pretraining Beyond Language Modeling""","['natural language processing', 'transfer learning', 'multitask learning']","""Work on the problem of contextualized word representationthe development of reusable neural network components for sentence understandinghas recently seen a surge of progress centered on the unsupervised pretraining task of language modeling with methods like ELMo (Peters et al., 2018). This paper contributes the first large-scale systematic study comparing different pretraining tasks in this context, both as complements to language modeling and as potential alternatives. The primary results of the study support the use of language modeling as a pretraining task and set a new state of the art among comparable models using multitask learning with language models. However, a closer look at these results reveals worryingly strong baselines and strikingly varied results across target tasks, suggesting that the widely-used paradigm of pretraining and freezing sentence encoders may not be an ideal platform for further work. ""","""This paper presents an extensive empirical study to sentence-level pre-training. The paper compares pre-trained language models to other potential alternative pre-training options, and concludes that while pre-trained language models are generally stronger than other alternatives, the robustness and generality of the currently available method is less than ideal, at least with respect to ELMO-based pretraining. Pros: The paper presents an extensive empirical study that offers new insights on pre-trained language models with respect to a variety of sentence-level tasks. Cons: The primarily contributions of this paper is empirical and technical novelty is relatively weak. Also, the insights are based just on ELMO, which may have a relatively weak empirical impact. The reviews were generally positive but marginally positive, which reflect that insights are interesting but not overwhelmingly interesting. None of these is a deal-breaker per say, but the paper does not provide sufficiently strong novelty, whether based on insights or otherwise, relative to other papers being considered for acceptance. Verdict: Leaning toward reject due to relatively weak novelty and empirical impact. Additional note on the final decision: The insights provided by the paper are valuable, thus the paper was originally recommended for an accept. However, during the calibration process across all areas, it became evident that we cannot accept all valuable papers, each presenting different types of hard work and novel contributions. Consequently, some papers with mostly positive (but marginally positive) reviews could not be included in the final cut, despite their unique values, hard work, and novel contributions. """
1252,"""Inducing Cooperation via Learning to reshape rewards in semi-cooperative multi-agent reinforcement learning""","['multi-agent reinforcement learning', 'deep reinforcement learning', 'multi-agent systems']","""We propose a deep reinforcement learning algorithm for semi-cooperative multi-agent tasks, where agents are equipped with their separate reward functions, yet with willingness to cooperate. Under these semi-cooperative scenarios, popular methods of centralized training with decentralized execution for inducing cooperation and removing the non-stationarity problem do not work well due to lack of a common shared reward as well as inscalability in centralized training. Our algorithm, called Peer-Evaluation based Dual DQN (PED-DQN), proposes to give peer evaluation signals to observed agents, which quantifies how they feel about a certain transition. This exchange of peer evaluation over time turns out to render agents to gradually reshape their reward functions so that their action choices from the myopic best-response tend to result in the good joint action with high cooperation. This evaluation-based method also allows flexible and scalable training by not assuming knowledge of the number of other agents and their observation and action spaces. We provide the performance evaluation of PED-DQN for the scenarios ranging from a simple two-person prisoner's dilemma to more complex semi-cooperative multi-agent tasks. In special cases where agents share a common reward function as in the centralized training methods, we show that inter-agent evaluation leads to better performance ""","""This work introduces a reward-shaping scheme for multi-agent settings based on the TD-error of other agents. Overall, reviewers were positive about the direction and the presentation but had a variety of concerns and questions and felt more experiments were necessary to validate the claims of flexibility and scalability, with results more comparable to the scale of the contemporary multi-agent literature. One note in particular: a feed-forward Q network is used in a partially observable environment, which the authors seemed to dismiss in their rebuttal. I agree with the reviewer that this is an important consideration when comparing to baselines which were developed with recurrent networks in mind. A revised manuscript addressed concerns with the presentation but did not introduce new results or plots, and reviewers were not convinced to alter their evaluation. There is agreement that this is an interesting paper, so I recommend that the authors conduct a more thorough empirical evaluation and submit to another venue."""
1253,"""Formal Limitations on the Measurement of Mutual Information""","['mutual information', 'predictive coding', 'unsupervised learning', 'predictive learning', 'generalization bounds', 'MINE', 'DIM', 'contrastive predictive coding']","""Motivated by applications to unsupervised learning, we consider the problem of measuring mutual information. Recent analysis has shown that naive kNN estimators of mutual information have serious statistical limitations motivating more refined methods. In this paper we prove that serious statistical limitations are inherent to any measurement method. More specifically, we show that any distribution-free high-confidence lower bound on mutual information cannot be larger than N) where pseudo-formula is the size of the data sample. We also analyze the Donsker-Varadhan lower bound on KL divergence in particular and show that, when simple statistical considerations are taken into account, this bound can never produce a high-confidence value larger than N While large high-confidence lower bounds are impossible, in practice one can use estimators without formal guarantees. We suggest expressing mutual information as a difference of entropies and using cross entropy as an entropy estimator. We observe that, although cross entropy is only an upper bound on entropy, cross-entropy estimates converge to the true cross entropy at the rate of pseudo-formula .""","""The paper proves that the Donsker-Varadhan lower bound on KL divergence cannot be used to estimate KL divergences of more than tens of bits, and that more generally any distribution-free high-confidence lower bound on mutual information cannot be larger than O(ln N) where N is the size of the data sample. As an alternative for applications such as maximum mutual information predictive coding, a form of representation learning, the paper proposes using the cross-entropy upper bound on entropy and estimating mutual information as a difference of two cross-entropies. These cross-entropy bounds converge to the true entropy as 1/\sqrt(N), but at the cost of providing neither an upper nor a lower bound on mutual information. There was a divergence of opinion between the reviewers on this paper. The most negative reviewer (R3) thought there should be experiments confirming that the DV bound fails when mutual information is high, was concerned that the theory applied only in the case of discrete distributions, and was concerned that the proposed optimization problem in Section 6 would be challenging due to its adversarial (max-inf) structure. The authors responded that they felt the theory could stand on its own without empirical tests (a point with which R1 agreed); that although their exposition was for discrete variables, the analysis applies to the continuous case as well; and that they agreed with the point about the difficulty of the optimization, but that GANs face similar difficulties. Because R3 did not participate in the discussion and the AC believes that the authors adequately addressed most of R3's issues in their response and revision, this review has been discounted. The next most negative reviewer (R2) wanted a discussion relating the ideas in this paper to kNN and kernel-based estimators of mutual information, wanted an empirical evaluation (like R3), and was concerned about whether the difference of cross-entropies provides an upper or lower bound on mutual information. In their response and revision the authors added some discussion of kNN methods (but not enough to make R2 happy) and clarified that the difference of cross-entropies provides neither an upper nor a lower bound. The most positive reviewer (R1) thinks the theoretical contribution of the paper is significant enough to justify publication in ICLR. The AC likes the theoretical work and feels that it raises important concerns about MINE, but concurs with R2 and R3 that some empirical validation of the theory is needed for the paper to appear in ICLR. The authors are strongly encouraged to perform an empirical validation of the theory and to submit this work to another machine learning venue."""
1254,"""A2BCD: Asynchronous Acceleration with Optimal Complexity""","['asynchronous', 'optimization', 'parallel', 'accelerated', 'complexity']",""" In this paper, we propose the Asynchronous Accelerated Nonuniform Randomized Block Coordinate Descent algorithm (A2BCD). We prove A2BCD converges linearly to a solution of the convex minimization problem at the same rate as NU_ACDM, so long as the maximum delay is not too large. This is the first asynchronous Nesterov-accelerated algorithm that attains any provable speedup. Moreover, we then prove that these algorithms both have optimal complexity. Asynchronous algorithms complete much faster iterations, and A2BCD has optimal complexity. Hence we observe in experiments that A2BCD is the top-performing coordinate descent algorithm, converging up to 4-5x faster than NU_ACDM on some data sets in terms of wall-clock time. To motivate our theory and proof techniques, we also derive and analyze a continuous-time analog of our algorithm and prove it converges at the same rate.""","""The reviewers all agreed that this paper makes a strong contribution to ICLR by providing the first asynchronous analysis of a Nesterov-accelerated coordinate descent method."""
1255,"""AntisymmetricRNN: A Dynamical System View on Recurrent Neural Networks""",[],"""Recurrent neural networks have gained widespread use in modeling sequential data. Learning long-term dependencies using these models remains difficult though, due to exploding or vanishing gradients. In this paper, we draw connections between recurrent networks and ordinary differential equations. A special form of recurrent networks called the AntisymmetricRNN is proposed under this theoretical framework, which is able to capture long-term dependencies thanks to the stability property of its underlying differential equation. Existing approaches to improving RNN trainability often incur significant computation overhead. In comparison, AntisymmetricRNN achieves the same goal by design. We showcase the advantage of this new architecture through extensive simulations and experiments. AntisymmetricRNN exhibits much more predictable dynamics. It outperforms regular LSTM models on tasks requiring long-term memory and matches the performance on tasks where short-term dependencies dominate despite being much simpler.""","""The paper presents a novel idea with a compelling experimental study. Good paper, accept."""
1256,"""Learning Kolmogorov Models for Binary Random Variables""","['Kolmogorov model', 'interpretable models', 'causal relations mining', 'non-convex optimization', 'relaxations']","""We propose a framework for learning a Kolmogorov model, for a collection of binary random variables. More specifically, we derive conditions that link (in the sense of implications in mathematical logic) outcomes of specific random variables and extract valuable relations from the data. We also propose an efficient algorithm for computing the model and show its first-order optimality, despite the combinatorial nature of the learning problem. We exemplify our general framework to recommendation systems and gene expression data. We believe that the work is a significant step toward interpretable machine learning. ""","""This work propose a method for learning a Kolmogorov model, which is a binary random variable model that is very similar (or identical) to a matrix factorization model. The work proposes an alternative optimization approach that is again similar to matrix factorization approaches. Unfortunately, no discussion or experiments are made to compare the proposed problem and method with standard matrix factorization; without such comparison, it is unclear if the proposed is substantially new, or a reformation of a standard problem. The authors are encouraged to improve the draft to clarify the connection matrix factorization and standard factor models. """
1257,"""Representation-Constrained Autoencoders and an Application to Wireless Positioning""","['Autoencoder', 'dimensionality reduction', 'wireless positioning', 'channel charting', 'localization']","""In a number of practical applications that rely on dimensionality reduction, the dataset or measurement process provides valuable side information that can be incorporated when learning low-dimensional embeddings. We propose the inclusion of pairwise representation constraints into autoencoders (AEs) with the goal of promoting application-specific structure. We use synthetic results to show that only a small amount of AE representation constraints are required to substantially improve the local and global neighborhood preserving properties of the learned embeddings. To demonstrate the efficacy of our approach and to illustrate a practical application that naturally provides such representation constraints, we focus on wireless positioning using a recently proposed channel charting framework. We show that representation-constrained AEs recover the global geometry of the learned low-dimensional representations, which enables channel charting to perform approximate positioning without access to global navigation satellite systems or supervised learning methods that rely on extensive measurement campaigns. ""","""The reviewers found the work interesting and sensible. The application of latent space constrained autoencoders to wireless positioning certainly seems novel. Applications can certainly be exciting additions to the conference program. However, the reviewers weren't convinced that the technical content of the paper was sufficiently novel to be interesting to the ICLR community. In particular, the reviewers seem concerned that there are no comparisons to more recent methods for dimensionality reduction and learning latent embeddings, such as variational auto-encoders. Certainly a comparison to more recent work constraining latent representations seems warranted to justify this particular approach."""
1258,"""Bayesian Policy Optimization for Model Uncertainty""","['Bayes-Adaptive Markov Decision Process', 'Model Uncertainty', 'Bayes Policy Optimization']","""Addressing uncertainty is critical for autonomous systems to robustly adapt to the real world. We formulate the problem of model uncertainty as a continuous Bayes-Adaptive Markov Decision Process (BAMDP), where an agent maintains a posterior distribution over latent model parameters given a history of observations and maximizes its expected long-term reward with respect to this belief distribution. Our algorithm, Bayesian Policy Optimization, builds on recent policy optimization algorithms to learn a universal policy that navigates the exploration-exploitation trade-off to maximize the Bayesian value function. To address challenges from discretizing the continuous latent parameter space, we propose a new policy network architecture that encodes the belief distribution independently from the observable state. Our method significantly outperforms algorithms that address model uncertainty without explicitly reasoning about belief distributions and is competitive with state-of-the-art Partially Observable Markov Decision Process solvers.""","""The paper proposed a deep, Bayesian optimization approach to RL with model uncertainty (BAMDP). The algorithm is a variant of policy gradient, which in each iteration uses a Bayes filter on sampled MDPs to update the posterior belief distribution of the parameters. An extension is also made to POMDPs. The work is a combination of existing techniques, and the algorithmic novelty is a bit low. Initial reviews suggested the empirical study could be improved with better baselines, and the main idea of the proposed method could be expended. The revised version moves towards this direction, and the author responses were helpful. Overall, the paper is a useful contribution."""
1259,"""GENERALIZED ADAPTIVE MOMENT ESTIMATION""","['adaptive moment estimation', 'SGD', 'AMSGrad']","""Adaptive gradient methods have experienced great success in training deep neural networks (DNNs). The basic idea of the methods is to track and properly make use of the first and/or second moments of the gradient for model-parameter updates over iterations for the purpose of removing the need for manual interference. In this work, we propose a new adaptive gradient method, referred to as generalized adaptive moment estimation (Game). From a high level perspective, the new method introduces two more parameters w.r.t. AMSGrad (S. J. Reddi & Kumar (2018)) and one more parameter w.r.t. PAdam (Chen & Gu (2018)) to enlarge the parameter- selection space for performance enhancement while reducing the memory cost per iteration compared to AMSGrad and PAdam. The saved memory space amounts to the number of model parameters, which is significant for large-scale DNNs. Our motivation for introducing additional parameters in Game is to provide algorithmic flexibility to facilitate a reduction of the performance gap between training and validation datasets when training a DNN. Convergence analysis is provided for applying Game to solve both convex optimization and smooth nonconvex optmization. Empirical studies for training four convolutional neural networks over MNIST and CIFAR10 show that under proper parameter selection, Game produces promising validation performance as compared to AMSGrad and PAdam.""","""The reviewers find the per difficult to read. Reviewers also had concerns regarding the correctness of various claims in the paper. The paper was also found lacking in experimental analysis, as it only tested on relatively small datasets, and only no a CNN architecture. Overall, the paper appears to be lacking in quality and clarity, and questionable in correctness and originality."""
1260,"""Characterizing the Accuracy/Complexity Landscape of Explanations of Deep Networks through Knowledge Extraction""","['Deep Networks', 'Explainability', 'Knowledge Extraction']","""Knowledge extraction techniques are used to convert neural networks into symbolic descriptions with the objective of producing more comprehensible learning models. The central challenge is to find an explanation which is more comprehensible than the original model while still representing that model faithfully. The distributed nature of deep networks has led many to believe that the hidden features of a neural network cannot be explained by logical descriptions simple enough to be understood by humans, and that decompositional knowledge extraction should be abandoned in favour of other methods. In this paper we examine this question systematically by proposing a knowledge extraction method using \textit{M-of-N} rules which allows us to map the complexity/accuracy landscape of rules describing hidden features in a Convolutional Neural Network (CNN). Experiments reported in this paper show that the shape of this landscape reveals an optimal trade off between comprehensibility and accuracy, showing that each latent variable has an optimal \textit{M-of-N} rule to describe its behaviour. We find that the rules with optimal tradeoff in the first and final layer have a high degree of explainability whereas the rules with the optimal tradeoff in the second and third layer are less explainable. The results shed light on the feasibility of rule extraction from deep networks, and point to the value of decompositional knowledge extraction as a method of explainability.""","""The presented paper introduces a method to represent neural networks as logical rules of varying complexity, and demonstrate a tradeoff between complexity and error. Reviews yield unanimous reject, with insufficient responses by authors. Pros: + Paper well written Cons: - R1 states inadequacy of baselines, which authors do not address. - R3&4 raise issues about the novelty of the idea. - R2&4 raise issues on limited scope of evaluation, and asked for additional experiments on at least 2 datasets which authors did not provide. Area chair notes the similarity of this work to other works on network compression, i.e. compression of bits to represent weights and activations. By converting neurons to logical clauses, this is essentially a similar method. Authors should familiarize themselves with this field and use it as a baseline comparison. i.e.: pseudo-url """
1261,"""Hint-based Training for Non-Autoregressive Translation""","['Natural Language Processing', 'Machine Translation', 'Non-Autoregressive Model']","""Machine translation is an important real-world application, and neural network-based AutoRegressive Translation (ART) models have achieved very promising accuracy. Due to the unparallelizable nature of the autoregressive factorization, ART models have to generate tokens one by one during decoding and thus suffer from high inference latency. Recently, Non-AutoRegressive Translation (NART) models were proposed to reduce the inference time. However, they could only achieve inferior accuracy compared with ART models. To improve the accuracy of NART models, in this paper, we propose to leverage the hints from a well-trained ART model to train the NART model. We define two hints for the machine translation task: hints from hidden states and hints from word alignments, and use such hints to regularize the optimization of NART models. Experimental results show that the NART model trained with hints could achieve significantly better translation performance than previous NART models on several tasks. In particular, for the WMT14 En-De and De-En task, we obtain BLEU scores of 25.20 and 29.52 respectively, which largely outperforms the previous non-autoregressive baselines. It is even comparable to a strong LSTM-based ART model (24.60 on WMT14 En-De), but one order of magnitude faster in inference.""",""" + sufficiently strong results + a fast / parallelizable model - Novelty with respect to previous work is not as great (see AnonReviewer1 and AnonReviewer2's comments) - The same reviewers raised concerns about the discussion of related work (e.g., positioning with respect to work on knowledge distillation). I agree that the very related work of Roy et al should be mentioned, even though it has not been published it has been on arxiv since May. - Ablation studies are only on smaller IWSLT datasets, confirming that the hints from an auto-regressive model are beneficial (whereas the main results are on WMT) - I agree with R1 that the important modeling details (e.g., describing how the latent structure is generated) should not be described only in the appendix, esp given non-standard modeling choices. R1 is concerned that a model which does not have any autoregressive components (i.e. not even for the latent state) may have trouble representing multiple modes. I do find it surprising that the model with non-autoregressive latent state works well however I do not find this a sufficient ground for rejection on its own. However, emphasizing this point and discussing the implication in the paper makes a lot of sense, and should have been done. As of now, it is downplayed. R1 is concerned that such model may be gaming BLEU: as BLEU is less sensitive to long-distance dependencies, they may get damaged for the model which does not have any autoregressive components. Again, given the standards in the field, I do not think it is fair to require human evaluation, but I agree that including it would strengthen the paper and the arguments. Overall, I do believe that the paper is sufficiently interesting and should get published but I also believe that it needs further revisions / further experiments. """
1262,"""TENSOR RING NETS ADAPTED DEEP MULTI-TASK LEARNING""","['deep learning', 'deep multi-task learning', 'tensor factorization', 'tensor ring nets']","""Recent deep multi-task learning (MTL) has been witnessed its success in alleviating data scarcity of some task by utilizing domain-specific knowledge from related tasks. Nonetheless, several major issues of deep MTL, including the effectiveness of sharing mechanisms, the efficiency of model complexity and the flexibility of network architectures, still remain largely unaddressed. To this end, we propose a novel generalized latent-subspace based knowledge sharing mechanism for linking task-specific models, namely tensor ring multi-task learning (TRMTL). TRMTL has a highly compact representation, and it is very effective in transferring task-invariant knowledge while being super flexible in learning task-specific features, successfully mitigating the dilemma of both negative-transfer in lower layers and under-transfer in higher layers. Under our TRMTL, it is feasible for each task to have heterogenous input data dimensionality or distinct feature sizes at different hidden layers. Experiments on a variety of datasets demonstrate our model is capable of significantly improving each single tasks performance, particularly favourable in scenarios where some of the tasks have insufficient data.""","""AR1 is concerned about the poor organisation of this paper. AR2 is concerned about the similarity between TRL and TR. The authors show some empirical results to support their intuition, however, no theoretical guarantees are provided regarding TRL superiority. Moreover, experiments for the Taskonomy dataset as well as on RNN have not been demonstrated, thus AR2 did not increase his/her score. AR3 is the most critical and finds the clarity and explanations not ready for publication. AC agrees with the reviewers in that the proposed idea has some merits, e.g. the reduction in the number of parameters seem a good point of this idea. However, reviewer urges the authors to seek non-trivial theoretical analysis for this method. Otherwise, it indeed is just an intelligent application paper and, as such, it cannot be accepted to ICLR."""
1263,"""Opportunistic Learning: Budgeted Cost-Sensitive Learning from Data Streams""","['Cost-Aware Learning', 'Feature Acquisition', 'Reinforcement Learning', 'Stream Learning', 'Deep Q-Learning']","""In many real-world learning scenarios, features are only acquirable at a cost constrained under a budget. In this paper, we propose a novel approach for cost-sensitive feature acquisition at the prediction-time. The suggested method acquires features incrementally based on a context-aware feature-value function. We formulate the problem in the reinforcement learning paradigm, and introduce a reward function based on the utility of each feature. Specifically, MC dropout sampling is used to measure expected variations of the model uncertainty which is used as a feature-value function. Furthermore, we suggest sharing representations between the class predictor and value function estimator networks. The suggested approach is completely online and is readily applicable to stream learning setups. The solution is evaluated on three different datasets including the well-known MNIST dataset as a benchmark as well as two cost-sensitive datasets: Yahoo Learning to Rank and a dataset in the medical domain for diabetes classification. According to the results, the proposed method is able to efficiently acquire features and make accurate predictions. ""","""This paper presents a reinforcement learning approach for online cost-aware feature acquisition. The utility of each feature is measured in terms of expected variations of the model uncertainty (using MC dropout sampling as an estimate of certainty) which is subsequently used as a reward function in the reinforcement learning formulation. The empirical evaluations show improvements over prior approaches in terms of accuracy-cost trade-off on three datasets. AC can confirm that all three reviewers have read the author responses and have significantly contributed to the revision of the manuscript. Initially, R1 and R2 raised important concerns regarding low technical novelty. R1 requested an ablation study to understand which of the following components gives the most improvement: 1) using proper certainty estimation; 2) using immediate reward; 3) new policy architecture. Pleased to report that the authors addressed the ablation study in their rebuttal and confirmed that MC-dropout certainty plays a crucial rule in the performance of the proposed method. R1 subsequently increased the assigned score to 6. R2 raised concerns about related prior work Contardo et al 2016, which similarly evaluates the most informative features given budget constraints with a recurrent neural network approach. After a long discussion and a detailed rebuttal, R2 upgraded the rating from below the threshold to 7, albeit acknowledging an incremental technical contribution. R3 raised important concerns regarding presentation clarity that were subsequently addressed by the authors. In conclusion, all three reviewers were convinced by the authors rebuttal and have upgraded their initial rating, and AC recommends acceptance of this paper congratulations to the authors! """
1264,"""End-to-End Learning of Video Compression Using Spatio-Temporal Autoencoders""",[],"""Deep learning (DL) is having a revolutionary impact in image processing, with DL-based approaches now holding the state of the art in many tasks, including image compression. However, video compression has so far resisted the DL revolution, with the very few proposed approaches being based on complex and impractical architectures with multiple networks. This paper proposes what we believe is the first approach to end-to-end learning of a single network for video compression. We tackle the problem in a novel way, avoiding explicit motion estimation/prediction, by formalizing it as the rate-distortion optimization of a single spatio-temporal autoencoder; i.e., we jointly learn a latent-space projection transform and a synthesis transform for low bitrate video compression. The quantizer uses a rounding scheme, which is relaxed during training, and an entropy estimation technique to enforce an information bottleneck, inspired by recent advances in image compression. We compare the obtained video compression networks with standard widely-used codecs, showing better performance than the MPEG-4 standard, being competitive with H.264/AVC for low bitrates. ""","""The paper proposes a neural network architecture for video compression. The reviewers point out lack of novelty with respect to recent neural compression works on static images, which the present paper extends by adding a temporal consistency loss. More importantly, reviewers point our severe problems with the metrics used to measure compression quality, which the authors promise to take into account in a future manuscript. """
1265,"""Evaluating Robustness of Neural Networks with Mixed Integer Programming""","['verification', 'adversarial robustness', 'adversarial examples', 'deep learning']","""Neural networks trained only to optimize for training accuracy can often be fooled by adversarial examples --- slightly perturbed inputs misclassified with high confidence. Verification of networks enables us to gauge their vulnerability to such adversarial examples. We formulate verification of piecewise-linear neural networks as a mixed integer program. On a representative task of finding minimum adversarial distortions, our verifier is two to three orders of magnitude quicker than the state-of-the-art. We achieve this computational speedup via tight formulations for non-linearities, as well as a novel presolve algorithm that makes full use of all information available. The computational speedup allows us to verify properties on convolutional and residual networks with over 100,000 ReLUs --- several orders of magnitude more than networks previously verified by any complete verifier. In particular, we determine for the first time the exact adversarial accuracy of an MNIST classifier to perturbations with bounded l- norm =0.1: for this classifier, we find an adversarial example for 4.38% of samples, and a certificate of robustness to norm-bounded perturbations for the remainder. Across all robust training procedures and network architectures considered, and for both the MNIST and CIFAR-10 datasets, we are able to certify more samples than the state-of-the-art and find more adversarial examples than a strong first-order attack.""",""" The paper investigates mixed-integer linear programming methods for neural net robustness verification in presence of adversarial attckas. The paper addresses and important problem, is well-written, presents a novel approach and demonstrates empirical improvements; all reviewers agree that this is a solid contribution to the field."""
1266,"""The Universal Approximation Power of Finite-Width Deep ReLU Networks""","['rate-distortion optimality', 'ReLU', 'deep learning', 'approximation theory', 'Weierstrass function']","""We show that finite-width deep ReLU neural networks yield rate-distortion optimal approximation (Blcskei et al., 2018) of a wide class of functions, including polynomials, windowed sinusoidal functions, one-dimensional oscillatory textures, and the Weierstrass function, a fractal function which is continuous but nowhere differentiable. Together with the recently established universal approximation result for affine function systems (Blcskei et al., 2018), this demonstrates that deep neural networks approximate vastly different signal structures generated by the affine group, the Weyl-Heisenberg group, or through warping, and even certain fractals, all with approximation error decaying exponentially in the number of neurons. We also prove that in the approximation of sufficiently smooth functions finite-width deep networks require strictly fewer neurons than finite-depth wide networks.""","""The paper contributes to the theoretical understanding of finite width ReLU networks. It contributes new ideas and constructions to investigate the representational power of such networks. In particular, the analysis works without skip connections. Referees found the paper refreshingly well-written and pleasant to read. There is a concern that the paper may be overstating the novelty and innovation of the results, as some of them are easy implications, and there are other previous works that have obtained results on finite width networks (see AnonReviewer4's comments). On the other hand, the authors were careful to cite when they reuse proof techniques from these and other works (AnonReviewer2). Another concern is that the considered target function space might be too narrow (see AnonReviewer2's comments). The authors clarify that the choice was because the considered classes are known to be hard to approximate and there are no known classical methods that would yield exponential approximation accuracy. Another concern is that the results might not be suitable to ICLR, having an emphasis on approximation theory and less on learning (see AnonReviewer3's comments). The reviewers consistently rate the paper as not very strong, with one marginally above acceptance threshold and two marginally below acceptance threshold ratings. While this appears to be a well written paper with valuable new ideas in regard to the approximation properties of networks, the contributions were not convincing enough. I would suggest that developing a clearer connecting to learning and broader classes of target functions could increase the appeal of the paper. """
1267,"""DPSNet: End-to-end Deep Plane Sweep Stereo""","['Deep Learning', 'Stereo', 'Depth', 'Geometry']","""Multiview stereo aims to reconstruct scene depth from images acquired by a camera under arbitrary motion. Recent methods address this problem through deep learning, which can utilize semantic cues to deal with challenges such as textureless and reflective regions. In this paper, we present a convolutional neural network called DPSNet (Deep Plane Sweep Network) whose design is inspired by best practices of traditional geometry-based approaches. Rather than directly estimating depth and/or optical flow correspondence from image pairs as done in many previous deep learning methods, DPSNet takes a plane sweep approach that involves building a cost volume from deep features using the plane sweep algorithm, regularizing the cost volume via a context-aware cost aggregation, and regressing the depth map from the cost volume. The cost volume is constructed using a differentiable warping process that allows for end-to-end training of the network. Through the effective incorporation of conventional multiview stereo concepts within a deep learning framework, DPSNet achieves state-of-the-art reconstruction results on a variety of challenging datasets.""","""A deep neural network pipeline for multiview stereo is presented. After rebuttal and discussion, all reviewers learn toward accepting the paper. Reviewer3 points to good results, but is concerned that the technical aspects are somewhat straightforward, and thus the contribution in this area is limited. The AC concurs with the reviewers."""
1268,"""Transferrable End-to-End Learning for Protein Interface Prediction""","['transfer learning', 'protein interface prediction', 'deep learning', 'structural biology']","""While there has been an explosion in the number of experimentally determined, atomically detailed structures of proteins, how to represent these structures in a machine learning context remains an open research question. In this work we demonstrate that representations learned from raw atomic coordinates can outperform hand-engineered structural features while displaying a much higher degree of transferrability. To do so, we focus on a central problem in biology: predicting how proteins interact with one anotherthat is, which surfaces of one protein bind to which surfaces of another protein. We present Siamese Atomic Surfacelet Network (SASNet), the first end-to-end learning method for protein interface prediction. Despite using only spatial coordinates and identities of atoms as inputs, SASNet outperforms state-of-the-art methods that rely on hand-engineered, high-level features. These results are particularly striking because we train the method entirely on a significantly biased data set that does not account for the fact that proteins deform when binding to one another. Demonstrating the first successful application of transfer learning to atomic-level data, our network maintains high performance, without retraining, when tested on real cases in which proteins do deform.""","""Two out of three reviews for this paper were provided in detail, but all three reviewers agreed unanimously that this paper is below the acceptance bar for ICLR. The reviewers admired the clarity of writing, and appreciated the importance of the application, but none recommended the paper for acceptance due largely to concerns on the experimental setup."""
1269,"""Imposing Category Trees Onto Word-Embeddings Using A Geometric Construction""","['category tree', 'word-embeddings', 'geometry']","""We present a novel method to precisely impose tree-structured category information onto word-embeddings, resulting in ball embeddings in higher dimensional spaces (N-balls for short). Inclusion relations among N-balls implicitly encode subordinate relations among categories. The similarity measurement in terms of the cosine function is enriched by category information. Using a geometric construction method instead of back-propagation, we create large N-ball embeddings that satisfy two conditions: (1) category trees are precisely imposed onto word embeddings at zero energy cost; (2) pre-trained word embeddings are well preserved. A new benchmark data set is created for validating the category of unknown words. Experiments show that N-ball embeddings, carrying category information, significantly outperform word embeddings in the test of nearest neighborhoods, and demonstrate surprisingly good performance in validating categories of unknown words. Source codes and data-sets are free for public access \url{pseudo-url} and \url{pseudo-url}. ""","""The authors provide an interesting method to infuse hierarchical information into existing word vectors. This could help with a variety of tasks that require both knowledge base information and textual co-occurrence counts. Despite some of the shortcomings that the reviewers point out, I believe this could be one missing puzzle piece of connecting symbolic information/sets/logic/KBs with neural nets and hence I recommend acceptance of this paper."""
1270,"""Doubly Sparse: Sparse Mixture of Sparse Experts for Efficient Softmax Inference""","['hierarchical softmax', 'model compression']","""Computations for the softmax function in neural network models are expensive when the number of output classes is large. This can become a significant issue in both training and inference for such models. In this paper, we present Doubly Sparse Softmax (DS-Softmax), Sparse Mixture of Sparse of Sparse Experts, to improve the efficiency for softmax inference. During training, our method learns a two-level class hierarchy by dividing entire output class space into several partially overlapping experts. Each expert is responsible for a learned subset of the output class space and each output class only belongs to a small number of those experts. During inference, our method quickly locates the most probable expert to compute small-scale softmax. Our method is learning-based and requires no knowledge of the output class partition space a priori. We empirically evaluate our method on several real-world tasks and demonstrate that we can achieve significant computation reductions without loss of performance.""","""This work proposes a new approximation method for softmax layers with large number of classes. The idea is to use a sparse two-layer mixture of experts. This approach successfully reduces the computation requires on the PTB and Wiki-2 datasets which have up to 32k classes. However, the reviewers argue that the work lacks relevant baselines such as D-softmax and adaptive-softmax. The authors argue that they focus on training and not inference and should do worse, but this should be substantiated in the paper by actual experimental results."""
1271,"""Variance Reduction for Reinforcement Learning in Input-Driven Environments""","['reinforcement learning', 'policy gradient', 'input-driven environments', 'variance reduction', 'baseline']","""We consider reinforcement learning in input-driven environments, where an exogenous, stochastic input process affects the dynamics of the system. Input processes arise in many applications, including queuing systems, robotics control with disturbances, and object tracking. Since the state dynamics and rewards depend on the input process, the state alone provides limited information for the expected future returns. Therefore, policy gradient methods with standard state-dependent baselines suffer high variance during training. We derive a bias-free, input-dependent baseline to reduce this variance, and analytically show its benefits over state-dependent baselines. We then propose a meta-learning approach to overcome the complexity of learning a baseline that depends on a long sequence of inputs. Our experimental results show that across environments from queuing systems, computer networks, and MuJoCo robotic locomotion, input-dependent baselines consistently improve training stability and result in better eventual policies.""","""This paper proposes an input-dependent baseline function to reduce variance in policy gradient estimation without adding bias. The approach is novel and theoretically validated, and the experimental results are convincing. The authors addressed nearly all of the reviewer's concerns. I recommend acceptance."""
1272,"""Defensive Quantization: When Efficiency Meets Robustness""","['defensive quantization', 'model quantization', 'adversarial attack', 'efficiency', 'robustness']","""Neural network quantization is becoming an industry standard to efficiently deploy deep learning models on hardware platforms, such as CPU, GPU, TPU, and FPGAs. However, we observe that the conventional quantization approaches are vulnerable to adversarial attacks. This paper aims to raise people's awareness about the security of the quantized models, and we designed a novel quantization methodology to jointly optimize the efficiency and robustness of deep learning models. We first conduct an empirical study to show that vanilla quantization suffers more from adversarial attacks. We observe that the inferior robustness comes from the error amplification effect, where the quantization operation further enlarges the distance caused by amplified noise. Then we propose a novel Defensive Quantization (DQ) method by controlling the Lipschitz constant of the network during quantization, such that the magnitude of the adversarial noise remains non-expansive during inference. Extensive experiments on CIFAR-10 and SVHN datasets demonstrate that our new quantization method can defend neural networks against adversarial examples, and even achieves superior robustness than their full-precision counterparts, while maintaining the same hardware efficiency as vanilla quantization approaches. As a by-product, DQ can also improve the accuracy of quantized models without adversarial attack. ""","""The reviewers agree the paper brings a novel perspective by controlling the conditioning of the model when performing quantization. The experiments are convincing experiments. We encourage the authors to incorporate additional references suggested in the reviews. We recommend acceptance. """
1273,"""AD-VAT: An Asymmetric Dueling mechanism for learning Visual Active Tracking""","['Active tracking', 'reinforcement learning', 'adversarial learning', 'multi agent']","""Visual Active Tracking (VAT) aims at following a target object by autonomously controlling the motion system of a tracker given visual observations. Previous work has shown that the tracker can be trained in a simulator via reinforcement learning and deployed in real-world scenarios. However, during training, such a method requires manually specifying the moving path of the target object to be tracked, which cannot ensure the trackers generalization on the unseen object moving patterns. To learn a robust tracker for VAT, in this paper, we propose a novel adversarial RL method which adopts an Asymmetric Dueling mechanism, referred to as AD-VAT. In AD-VAT, both the tracker and the target are approximated by end-to-end neural networks, and are trained via RL in a dueling/competitive manner: i.e., the tracker intends to lockup the target, while the target tries to escape from the tracker. They are asymmetric in that the target is aware of the tracker, but not vice versa. Specifically, besides its own observation, the target is fed with the trackers observation and action, and learns to predict the trackers reward as an auxiliary task. We show that such an asymmetric dueling mechanism produces a stronger target, which in turn induces a more robust tracker. To stabilize the training, we also propose a novel partial zero-sum reward for the tracker/target. The experimental results, in both 2D and 3D environments, demonstrate that the proposed method leads to a faster convergence in training and yields more robust tracking behaviors in different testing scenarios. For supplementary videos, see: pseudo-url The code is available at pseudo-url""","""The paper presents an adversarial learning framework for active visual tracking, a tracking setup where the tracker has camera control in order to follow a target object. The paper builds upon Luo et al. 2018 and proposes jointly learning tracker and target policies (as opposed to tracker policy alone). This automatically creates a curriculum of target trajectory difficulty, as opposed to the engineer designing the target trajectories. The paper further proposes a method for preventing the target to fast outperform the tracker and thus cause his policy to plateau. Experiments presented justify the problem formulation and design choices, and outperform Luo et al. . The task considered is very important, active surveillance with drones is just one sue case. A downside of the paper is that certain sentences have English mistakes, such as this one: ""The authors learn a policy that maps raw-pixel observation to control signal straightly with a Conv-LSTM network. Not only can it save the effort in tuning an extra camera controller, but also does it outperform the..."" However, overall the manuscript is well written, well structured, and easy to follow. The authors are encouraged to correct any remaining English mistakes in the manuscript. """
1274,"""Q-map: a Convolutional Approach for Goal-Oriented Reinforcement Learning""","['reinforcement learning', 'goal-oriented', 'convolutions', 'off-policy']","""Goal-oriented learning has become a core concept in reinforcement learning (RL), extending the reward signal as a sole way to define tasks. However, as parameterizing value functions with goals increases the learning complexity, efficiently reusing past experience to update estimates towards several goals at once becomes desirable but usually requires independent updates per goal. Considering that a significant number of RL environments can support spatial coordinates as goals, such as on-screen location of the character in ATARI or SNES games, we propose a novel goal-oriented agent called Q-map that utilizes an autoencoder-like neural network to predict the minimum number of steps towards each coordinate in a single forward pass. This architecture is similar to Horde with parameter sharing and allows the agent to discover correlations between visual patterns and navigation. For example learning how to use a ladder in a game could be transferred to other ladders later. We show how this network can be efficiently trained with a 3D variant of Q-learning to update the estimates towards all goals at once. While the Q-map agent could be used for a wide range of applications, we propose a novel exploration mechanism in place of epsilon-greedy that relies on goal selection at a desired distance followed by several steps taken towards it, allowing long and coherent exploratory steps in the environment. We demonstrate the accuracy and generalization qualities of the Q-map agent on a grid-world environment and then demonstrate the efficiency of the proposed exploration mechanism on the notoriously difficult Montezuma's Revenge and Super Mario All-Stars games.""","""The paper proposes to use a convolutional/de-convolutional Q function over on-screen goal locations, and applied to the problem of structured exploration. Reviewers pointed out the similarity to the UNREAL architecture, the difference being that the auxiliary Q functions learned are actually used to act in this case. Reviewers raised concerns regarding novelty, the formality of the writing, a lack of comparisons to other exploration methods, and the need for ground truth about the sprite location at training time. A minor revision to the text was made, but the reviewers did not feel their main criticisms were addressed. While the method shows promise, given that the authors acknowledge that the method is somewhat incremental, a more thorough quantitative and ablative study would be necessary in order to recommend acceptance."""
1275,"""Visual Reasoning by Progressive Module Networks""",[],"""Humans learn to solve tasks of increasing complexity by building on top of previously acquired knowledge. Typically, there exists a natural progression in the tasks that we learn most do not require completely independent solutions, but can be broken down into simpler subtasks. We propose to represent a solver for each task as a neural module that calls existing modules (solvers for simpler tasks) in a functional program-like manner. Lower modules are a black box to the calling module, and communicate only via a query and an output. Thus, a module for a new task learns to query existing modules and composes their outputs in order to produce its own output. Our model effectively combines previous skill-sets, does not suffer from forgetting, and is fully differentiable. We test our model in learning a set of visual reasoning tasks, and demonstrate improved performances in all tasks by learning progressively. By evaluating the reasoning process using human judges, we show that our model is more interpretable than an attention-based baseline. ""","""Important problem (modular & interpretable approaches for VQA and visual reasoning); well-written manuscript, sensible approach. Paper was reviewed by three experts. Initially there were some concerns but after the author response and reviewer discussion, all three unanimously recommend acceptance. """
1276,"""Wasserstein proximal of GANs""","['Optimal transport', 'Wasserstein gradient', 'Generative adversarial network', 'Unsupervised learning']","""We introduce a new method for training GANs by applying the Wasserstein-2 metric proximal on the generators. The approach is based on the gradient operator induced by optimal transport, which connects the geometry of sample space and parameter space in implicit deep generative models. From this theory, we obtain an easy-to-implement regularizer for the parameter updates. Our experiments demonstrate that this method improves the speed and stability in training GANs in terms of wall-clock time and Fr\'echet Inception Distance (FID) learning curves. ""","""Both R3 and R1 argue for rejection, while R2 argues for a weak accept. Given that we have to reject borderline paper, the AC concludes with ""revise and resubmit""."""
1277,"""On the Turing Completeness of Modern Neural Network Architectures""","['Transformer', 'NeuralGPU', 'Turing completeness']","""Alternatives to recurrent neural networks, in particular, architectures based on attention or convolutions, have been gaining momentum for processing input sequences. In spite of their relevance, the computational properties of these alternatives have not yet been fully explored. We study the computational power of two of the most paradigmatic architectures exemplifying these mechanisms: the Transformer (Vaswani et al., 2017) and the Neural GPU (Kaiser & Sutskever, 2016). We show both models to be Turing complete exclusively based on their capacity to compute and access internal dense representations of the data. In particular, neither the Transformer nor the Neural GPU requires access to an external memory to become Turing complete. Our study also reveals some minimal sets of elements needed to obtain these completeness results.""","""This paper provides a theoretical analysis of the Turing completeness of popular neural network architectures, specifically Neural Transformers and the Neural GPU. The reviewers agreed that this paper provides a meaningful theoretical contribution and should be accepted to the conference. Work of a theoretical nature is, amongst other types of work, called for by the ICLR CFP, but is not a very popular category for submissions, nor is it an easy one. As such, I am happy to follow the reviewers' recommendation and support this paper."""
1278,"""Adversarial Exploration Strategy for Self-Supervised Imitation Learning""","['adversarial exploration', 'self-supervised', 'imitation learning']","""We present an adversarial exploration strategy, a simple yet effective imitation learning scheme that incentivizes exploration of an environment without any extrinsic reward or human demonstration. Our framework consists of a deep reinforcement learning (DRL) agent and an inverse dynamics model contesting with each other. The former collects training samples for the latter, and its objective is to maximize the error of the latter. The latter is trained with samples collected by the former, and generates rewards for the former when it fails to predict the actual action taken by the former. In such a competitive setting, the DRL agent learns to generate samples that the inverse dynamics model fails to predict correctly, and the inverse dynamics model learns to adapt to the challenging samples. We further propose a reward structure that ensures the DRL agent collects only moderately hard samples and not overly hard ones that prevent the inverse model from imitating effectively. We evaluate the effectiveness of our method on several OpenAI gym robotic arm and hand manipulation tasks against a number of baseline models. Experimental results show that our method is comparable to that directly trained with expert demonstrations, and superior to the other baselines even without any human priors.""","""This paper proposes a method for incentivizing exploration in self-supervised learning using an inverse model, and then uses the learned inverse model for imitating an expert demonstration. The approach of incentivizing the agent to visit transitions where a learned model performs poorly. This relates to prior work (e.g. [1]), but using an inverse model instead of a forward model. The results are promising on challenging problem domains, and the method is simple. The authors have addressed several of the reviewer concerns throughout the discussion period. However, three primary concerns remain: (A) First and foremost: There has been confusion about the problem setting and the comparisons. I think these confusions have stemmed from the writing in the paper not being sufficiently clear. First, it should be made clear in the plots that the ""Demos"" comparison is akin to an oracle. Second, the difference between self-supervised imitation learning (IL) and traditional IL needs to be spelled out more clearly in the paper. Given that self-supervised imitation learning is not a previously established term, the problem statement needs to be clearly and formally described (and without relying heavily on prior papers). Further, the term self-supervised imitation learning does not seem to be an appropriate term, since imitation learning from an expert is, by definition, not self-supervised, as it involves supervisory information from an expert. Changing this term and clearly defining the problem would likely lead to less confusion about the method and the relevant comparisons. (B) The ""Demos"" comparison is meant as an upper bound on the performance of this particular approach. However, it is also important to understand what the upper bound is on these problems in general, irrespective of whether or not an inverse model is used. Training a policy with behavior cloning on demonstrations with many (s,a) pairs would be able to better provide such a comparison. (C) Inverse models inherently model the part of the environment that is directly controllable (e.g. the robot arm), and often do not effectively model other aspects of the environment that are only indirectly controllable (e.g. the objects). If the method overcomes this issue, then that should be discussed in the paper. Otherwise, the limitation should be outlined and discussed in more detail, including text that outlines which forms of problems and environments this approach is expected to be able to handle, and which of those it cannot handle. Generally, this paper is quite borderline, as indicated by the reviewer's scores. After going through the reviews and parts of the paper in detail, I am inclined to recommend reject as I think the above concerns do not outweigh the pros. One more minor comment is that the paper should consider mentioning the related work by Torabi et al. [2], which considers a similar approach in a slightly different problem setting. [1] Stadie et al. pseudo-url [2] Torabi et al. IJCAI '18 (pseudo-url)"""
1279,"""AntMan: Sparse Low-Rank Compression To Accelerate RNN Inference""","['model compression', 'RNN', 'perforamnce optimization', 'langugage model', 'machine reading comprehension', 'knowledge distillation', 'teacher-student']","""Wide adoption of complex RNN based models is hindered by their inference performance, cost and memory requirements. To address this issue, we develop AntMan, combining structured sparsity with low-rank decomposition synergistically, to reduce model computation, size and execution time of RNNs while attaining desired accuracy. AntMan extends knowledge distillation based training to learn the compressed models efficiently. Our evaluation shows that AntMan offers up to 100x computation reduction with less than 1pt accuracy drop for language and machine reading comprehension models. Our evaluation also shows that for a given accuracy target, AntMan produces 5x smaller models than the state-of-art. Lastly, we show that AntMan offers super-linear speed gains compared to theoretical speedup, demonstrating its practical value on commodity hardware.""","""The submission proposes a method that combines sparsification and low rank projections to compress a neural network. This is in line with nearly all state-of-the-art methods. The specific combination proposed in this instance are SVD for low-rank and localized group projections (LGP) for sparsity. The main concern about the paper is the lack of stronger comparison to sota compression techniques. The authors justify their choice in the rebuttal, but ultimately only compare to relatively straightforward baselines. The additional comparison with e.g. Table 6 of the appendix does not give sufficient information to replicate or to know how the reduction in parameters was achieved. The scores for this paper were borderline, and the reviewers were largely in consensus with their scores and the points raised in the reviews. Given the highly selective nature of ICLR, the overall evaluations and remaining questions about the paper and comparison to baselines indicates that it does not pass the threshold for acceptance."""
1280,"""DppNet: Approximating Determinantal Point Processes with Deep Networks""","['dpp', 'submodularity', 'determinant']","""Determinantal Point Processes (DPPs) provide an elegant and versatile way to sample sets of items that balance the point-wise quality with the set-wise diversity of selected items. For this reason, they have gained prominence in many machine learning applications that rely on subset selection. However, sampling from a DPP over a ground set of size N is a costly operation, requiring in general an O(N^3) preprocessing cost and an O(Nk^3) sampling cost for subsets of size k. We approach this problem by introducing DppNets: generative deep models that produce DPP-like samples for arbitrary ground sets. We develop an inhibitive attention mechanism based on transformer networks that captures a notion of dissimilarity between feature vectors. We show theoretically that such an approximation is sensible as it maintains the guarantees of inhibition or dissimilarity that makes DPP so powerful and unique. Empirically, we demonstrate that samples from our model receive high likelihood under the more expensive DPP alternative.""","""The paper addresses the complexity issue of Determinantal Point Processes via generative deep models. The reviewers and AC note the critical limitation of applicability of this paper to variable ground set sizes, whether authors' rebuttal is not convincing enough. AC thinks the proposed method has potential and is interesting, but decided that the authors need more works to publish."""
1281,"""Fortified Networks: Improving the Robustness of Deep Networks by Modeling the Manifold of Hidden Representations""","['adversarial examples', 'adversarial training', 'autoencoders', 'hidden state']","""Deep networks have achieved impressive results across a variety of important tasks. However, a known weakness is a failure to perform well when evaluated on data which differ from the training distribution, even if these differences are very small, as is the case with adversarial examples. We propose \emph{Fortified Networks}, a simple transformation of existing networks, which fortifies the hidden layers in a deep network by identifying when the hidden states are off of the data manifold, and maps these hidden states back to parts of the data manifold where the network performs well. Our principal contribution is to show that fortifying these hidden states improves the robustness of deep networks and our experiments (i) demonstrate improved robustness to standard adversarial attacks in both black-box and white-box threat models; (ii) suggest that our improvements are not primarily due to the problem of deceptively good results due to degraded quality in the gradient signal (the gradient masking problem) and (iii) show the advantage of doing this fortification in the hidden layers instead of the input space. We demonstrate improvements in adversarial robustness on three datasets (MNIST, Fashion MNIST, CIFAR10), across several attack parameters, both white-box and black-box settings, and the most widely studied attacks (FGSM, PGD, Carlini-Wagner). We show that these improvements are achieved across a wide variety of hyperparameters. ""","""This paper suggests a method for defending against adversarial examples and out-of-distribution samples via projection onto the data manifold. The paper suggests a new method for detecting when hidden layers are off of the manifold, and uses auto encoders to map them back onto the manifold. The paper is well-written and the method is novel and interesting. However, most of the reviewers agree that the original robustness evaluations were not sufficient due to restricting the evaluation to using FGSM baseline and comparison with thermometer encoding (which both are known to not be fully effective baselines). After rebuttal, Reviewer 4 points out that the method offers very little robustness over adversarial training alone, even though it is combined with adversarial training, which suggests that the method itself provides very little robustness. """
1282,"""Radial Basis Feature Transformation to Arm CNNs Against Adversarial Attacks""","['Radial basis feature transformation', 'convolutional neural networks', 'adversarial defense']","""The linear and non-flexible nature of deep convolutional models makes them vulnerable to carefully crafted adversarial perturbations. To tackle this problem, in this paper, we propose a nonlinear radial basis convolutional feature transformation by learning the Mahalanobis distance function that maps the input convolutional features from the same class into tight clusters. In such a space, the clusters become compact and well-separated, which prevent small adversarial perturbations from forcing a sample to cross the decision boundary. We test the proposed method on three publicly available image classification and segmentation data-sets namely, MNIST, ISBI ISIC skin lesion, and NIH ChestX-ray14. We evaluate the robustness of our method to different gradient (targeted and untargeted) and non-gradient based attacks and compare it to several non-gradient masking defense strategies. Our results demonstrate that the proposed method can boost the performance of deep convolutional neural networks against adversarial perturbations without accuracy drop on clean data.""","""Strengths of the paper: Based on previous work suggesting that radial basis features can help defend against adversarial attacks, the paper proposes a concrete method for incorporating them in deep networks. The paper evaluates the method on multiple datasets, including MNIST and ISBI International Skin Imaging Collaboration (ISIC) Challenge. Weaknesses: Reviewers 2 and 3 felt that the paper was not clearly written, and cited several concrete questions about the method that could not be understood from the paper. There were additional concerns of lacking comparison to existing methods, and Reviewer 1 pointed out that a competing method gave higher performance, although this was not reported in the present submission. Points of contention: The authors did not provide a response to the reviewer concerns. Consensus: All reviewers recommended that the paper be rejected, and the authors did not provide a rebuttal."""
1283,"""Likelihood-based Permutation Invariant Loss Function for Probability Distributions""","['Set reconstruction', 'maximum likelihood', 'permutation invariance']","""We propose a permutation-invariant loss function designed for the neural networks reconstructing a set of elements without considering the order within its vector representation. Unlike popular approaches for encoding and decoding a set, our work does not rely on a carefully engineered network topology nor by any additional sequential algorithm. The proposed method, Set Cross Entropy, has a natural information-theoretic interpretation and is related to the metrics defined for sets. We evaluate the proposed approach in two object reconstruction tasks and a rule learning task.""","""This paper proposes a new permutation invariant loss (where the order doesn't matter), motivated by set autoencoding settings. This is an important problem, and the authors' solution is interesting. The reviewers, however, found the exposition to be unclear, in particular the explanation on how the loss function is derived was confusing for two of the reviewers. Reviewers also found the experimental results to be not convincing, even after the revision. This is a borderline paper: the idea is valuable and I'd encourage the authors to develop it further, improving exposition and including additional experiments as suggested by the reviewers. """
1284,"""Laplacian Networks: Bounding Indicator Function Smoothness for Neural Networks Robustness""","['GSP', 'robustness', 'noise', 'deep learning', 'neural networks']",""" For the past few years, Deep Neural Network (DNN) robustness has become a question of paramount importance. As a matter of fact, in sensitive settings misclassification can lead to dramatic consequences. Such misclassifications are likely to occur when facing adversarial attacks, hardware failures or limitations, and imperfect signal acquisition. To address this question, authors have proposed different approaches aiming at increasing the robustness of DNNs, such as adding regularizers or training using noisy examples. In this paper we propose a new regularizer built upon the Laplacian of similarity graphs obtained from the representation of training data at each layer of the DNN architecture. This regularizer penalizes large changes (across consecutive layers in the architecture) in the distance between examples of different classes, and as such enforces smooth variations of the class boundaries. Since it is agnostic to the type of deformations that are expected when predicting with the DNN, the proposed regularizer can be combined with existing ad-hoc methods. We provide theoretical justification for this regularizer and demonstrate its effectiveness to improve robustness of DNNs on classical supervised learning vision datasets. ""","""The paper proposes a new graph-based regularizer to improve the robustness of deep nets. The idea is to encourage smoothness on a graph built on the features at different layers. Experiments on CIFAR-10 show that the method provides robustness over very different types of perturbations such as adversarial examples or quantization. The reviewers raised concerns around the significance of the results, the reliance on a single dataset and the unexplained link between adversarial examples and the regularization. Despite the revision, the reviewers maintain their concerns. For this reason this work is not ready for publication."""
1285,"""Better Accuracy with Quantified Privacy: Representations Learned via Reconstructive Adversarial Network""","['end-user privacy', 'utility', 'feature learning', 'adversarial training']","""The remarkable success of machine learning, especially deep learning, has produced a variety of cloud-based services for mobile users. Such services require an end user to send data to the service provider, which presents a serious challenge to end-user privacy. To address this concern, prior works either add noise to the data or send features extracted from the raw data. They struggle to balance between the utility and privacy because added noise reduces utility and raw data can be reconstructed from extracted features. This work represents a methodical departure from prior works: we balance between a measure of privacy and another of utility by leveraging adversarial learning to find a sweeter tradeoff. We design an encoder that optimizes against the reconstruction error (a measure of privacy), adversarially by a Decoder, and the inference accuracy (a measure of utility) by a Classifier. The result is RAN, a novel deep model with a new training algorithm that automatically extracts features for classification that are both private and useful. It turns out that adversarially forcing the extracted features to only conveys the intended information required by classification leads to an implicit regularization leading to better classification accuracy than the original model which completely ignores privacy. Thus, we achieve better privacy with better utility, a surprising possibility in machine learning! We conducted extensive experiments on five popular datasets over four training schemes, and demonstrate the superiority of RAN compared with existing alternatives.""","""Following the unanimous vote of the reviewers, this paper is not ready for publication at ICLR. A significant concern is that the definition of privacy used here is not adequately justified. This opens up issues of: 1) possible attacks, 2) privacy-guarantees that are not worst-case, among others. """
1286,"""Improved robustness to adversarial examples using Lipschitz regularization of the loss""","['Adversarial training', 'adversarial examples', 'deep neural networks', 'regularization', 'Lipschitz constant']","""We augment adversarial training (AT) with worst case adversarial training (WCAT) which improves adversarial robustness by 11% over the current state- of-the-art result in the `2-norm on CIFAR-10. We interpret adversarial training as Total Variation Regularization, which is a fundamental tool in mathematical im- age processing, and WCAT as Lipschitz regularization, which appears in Image Inpainting. We obtain verifiable worst and average case robustness guarantees, based on the expected and maximum values of the norm of the gradient of the loss.""","""This paper suggests augmenting adversarial training with a Lipschitz regularization of the loss, and suggests that this improves the adversarial robustness of deep neural networks. The idea of using such regularization seems novel. However, several reviewers were seriously concerned with the quality of the writing. In particular, the paper contains claims that not only are not needed but also are incorrect. Also, the Reviewer 2 in particular was also concerned with the presentation of prior work on Lipschitz regularization. Such poor quality of the presentation makes it impossible to properly evaluate the actual paper contribution. """
1287,"""Locally Linear Unsupervised Feature Selection""","['Unsupervised Learning', 'Feature Selection', 'Dimension Reduction']","""The paper, interested in unsupervised feature selection, aims to retain the features best accounting for the local patterns in the data. The proposed approach, called Locally Linear Unsupervised Feature Selection, relies on a dimensionality reduction method to characterize such patterns; each feature is thereafter assessed according to its compliance w.r.t. the local patterns, taking inspiration from Locally Linear Embedding (Roweis and Saul, 2000). The experimental validation of the approach on the scikit-feature benchmark suite demonstrates its effectiveness compared to the state of the art.""","""This paper presents an LLE-based unsupervised feature selection approach. While one of the reviewers has acknowledged that the paper is well-written with clear mathematical explanations of the key ideas, it also lacks a sufficiently strong theoretical foundation as the authors have acknowledged in their responses; as well as novelty in its tight connection to LLE. When theoretical backbone is weak, the role of empirical results is paramount, but the paper is not convincing in that regard."""
1288,"""Bayesian Deep Learning via Stochastic Gradient MCMC with a Stochastic Approximation Adaptation""","['generalized stochastic approximation', 'stochastic gradient Markov chain Monte Carlo', 'adaptive algorithm', 'EM algorithm', 'convolutional neural networks', 'Bayesian inference', 'sparse prior', 'spike and slab prior', 'local trap']","""We propose a robust Bayesian deep learning algorithm to infer complex posteriors with latent variables. Inspired by dropout, a popular tool for regularization and model ensemble, we assign sparse priors to the weights in deep neural networks (DNN) in order to achieve automatic dropout and avoid over-fitting. By alternatively sampling from posterior distribution through stochastic gradient Markov Chain Monte Carlo (SG-MCMC) and optimizing latent variables via stochastic approximation (SA), the trajectory of the target weights is proved to converge to the true posterior distribution conditioned on optimal latent variables. This ensures a stronger regularization on the over-fitted parameter space and more accurate uncertainty quantification on the decisive variables. Simulations from large-p-small-n regressions showcase the robustness of this method when applied to models with latent variables. Additionally, its application on the convolutional neural networks (CNN) leads to state-of-the-art performance on MNIST and Fashion MNIST datasets and improved resistance to adversarial attacks. ""","""This paper proposes a Bayesian alternative to dropout for deep networks by extending the EM-based variable selection method with SG-MCMC for sampling weights and stochastic approximation for tuning hyper-parameters. The method is well presented with a clear motivation. The combination of SMVS, SG-MCMC, and SA as a mixed optimization-sampling approach is technically sound. The main concern raised by the readers is the limited originality. SG-MCMC has been studied extensively for Bayesian deep networks and applying the spike-and-slab prior as an alternative to dropout is a straightforward idea. The main contribution of the paper appears to be extending EMVS to deep net with commonly used sampling techniques for Bayesian networks. Another concern is the lack of experimental justification for the advantage of the proposed method. While the authors promise to include more experiment results in the camera-ready version, it requires a considerable amount of effort and the decision unfortunately has to be made based on the current revision."""
1289,"""Compound Density Networks""","['uncertainty in neural networks', 'ensemble', 'mixture model']","""Despite the huge success of deep neural networks (NNs), finding good mechanisms for quantifying their prediction uncertainty is still an open problem. It was recently shown, that using an ensemble of NNs trained with a proper scoring rule leads to results competitive to those of Bayesian NNs. This ensemble method can be understood as finite mixture model with uniform mixing weights. We build on this mixture model approach and increase its flexibility by replacing the fixed mixing weights by an adaptive, input-dependent distribution (specifying the probability of each component) represented by an NN, and by considering uncountably many mixture components. The resulting model can be seen as the continuous counterpart to mixture density networks and is therefore referred to as compound density network. We empirically show that the proposed model results in better uncertainty estimates and is more robust to adversarial examples than previous approaches.""","""Reviewers are in a consensus and weakly recommended to reject after engaging with the authors, with the reviewers updating their scores on Dec 11 after engagement. The authors answered most of the reviewers' concerns, however from further discussions with the reviewers there are still some points which lead them to rank the paper lower than others. I thus lean to reject. Please take reviewers' comments into consideration to improve submission should you choose to resubmit."""
1290,"""On the Trajectory of Stochastic Gradient Descent in the Information Plane""","['Stochastic gradient descent', 'Deep neural networks', 'Entropy', 'Information theory', 'Markov chains', 'Hidden Markov process.']","""Studying the evolution of information theoretic quantities during Stochastic Gradient Descent (SGD) learning of Artificial Neural Networks (ANNs) has gained popularity in recent years. Nevertheless, these type of experiments require estimating mutual information and entropy which becomes intractable for moderately large problems. In this work we propose a framework for understanding SGD learning in the information plane which consists of observing entropy and conditional entropy of the output labels of ANN. Through experimental results and theoretical justifications it is shown that, under some assumptions, the SGD learning trajectories appear to be similar for different ANN architectures. First, the SGD learning is modeled as a Hidden Markov Process (HMP) whose entropy tends to increase to the maximum. Then, it is shown that the SGD learning trajectory appears to move close to the shortest path between the initial and final joint distributions in the space of probability measures equipped with the total variation metric. Furthermore, it is shown that the trajectory of learning in the information plane can provide an alternative for observing the learning process, with potentially richer information about the learning than the trajectories in training and test error. ""","""The paper proposes a quantity to monitor learning on an information plane which is related to the information curves considered in the bottleneck analysis but is more reliable and easier to compute. The main concern with the paper is the lack of interpretation and elaboration of potential uses. A concern is raised that the proposed method abstracts away way too much detail, so that the shapes of the curves are to be expected and contain little useful information (see AnonReviewer2 comments). The authors agree to some of the main issues, as they pointed out in the discussion, although they maintain that the method could still contain useful information. The reviewers are not very convinced by this paper, with ratings either marginally above the acceptance threshold, marginally below the acceptance threshold, or strong reject. """
1291,"""Modulating transfer between tasks in gradient-based meta-learning""","['meta-learning', 'clustering', 'learning-to-learn', 'mixture', 'hierarchical Bayes', 'hierarchical model', 'gradient-based meta-learning']","""Learning-to-learn or meta-learning leverages data-driven inductive bias to increase the efficiency of learning on a novel task. This approach encounters difficulty when transfer is not mutually beneficial, for instance, when tasks are sufficiently dissimilar or change over time. Here, we use the connection between gradient-based meta-learning and hierarchical Bayes to propose a mixture of hierarchical Bayesian models over the parameters of an arbitrary function approximator such as a neural network. Generalizing the model-agnostic meta-learning (MAML) algorithm, we present a stochastic expectation maximization procedure to jointly estimate parameter initializations for gradient descent as well as a latent assignment of tasks to initializations. This approach better captures the diversity of training tasks as opposed to consolidating inductive biases into a single set of hyperparameters. Our experiments demonstrate better generalization on the standard miniImageNet benchmark for 1-shot classification. We further derive a novel and scalable non-parametric variant of our method that captures the evolution of a task distribution over time as demonstrated on a set of few-shot regression tasks.""","""This paper is extending the meta-learning MAML method to the mixture case. Specifically, the global parameters of the method are now modeled as a mixture. The authors also derive the elaborate associated inference for this approach. The paper is well written although Rev2 raises some presentation issues that can surely improve the quality of the paper, if addressed in depth. The results do not convince any of the three reviewers. Rev3 asks for a clearer exposition of the results to increase convincingness. Rev2 and Rev1 also make similar comments. Rev1 also questions the motivation of the approach, although the other two reviewers seem to find the approach well motivated. Although it certainly helps to prove the motivation within a very tailored to the method application, the AC weighted the opinion of all reviewers and did not consider the paper to lack in the motivation aspect. The reviewers were overall not very impressed with this paper and that does not seem to stem from lack of novelty or technical correctness. Instead, it seems that this work is rather inconclusive (or at least it is presented in an inconclusive manner): Rev1 says that the important questions (like trade-offs and other practical issues) are not answered, Rev2 suggests that maybe this paper is trying to address too much, and all three reviewers are not convinced by the experiments and derived insights. Finally, Rev2 points out some inherent caveats of the method; although they do not seem to be severe enough to undermine the overall quality of the approach, it would be instructive to have them investigated more thoroughly (even if not completely solving them)."""
1292,"""Unsupervised Discovery of Parts, Structure, and Dynamics""","['Self-Supervised Learning', 'Visual Prediction', 'Hierarchical Models']","""Humans easily recognize object parts and their hierarchical structure by watching how they move; they can then predict how each part moves in the future. In this paper, we propose a novel formulation that simultaneously learns a hierarchical, disentangled object representation and a dynamics model for object parts from unlabeled videos. Our Parts, Structure, and Dynamics (PSD) model learns to, first, recognize the object parts via a layered image representation; second, predict hierarchy via a structural descriptor that composes low-level concepts into a hierarchical structure; and third, model the system dynamics by predicting the future. Experiments on multiple real and synthetic datasets demonstrate that our PSD model works well on all three tasks: segmenting object parts, building their hierarchical structure, and capturing their motion distributions.""","""The paper proposes a novel method that learns decompositions of an image over parts, their hierarchical structure and their motion dynamics given temporal image pairs. The problem tackled is of great importance for unsupervised learning from videos. One downside of the paper is the simple datasets used to demonstrate the effectiveness of the method. All reviewers though agree on it being a valuable contribution for ICLR. In the related work section the paper mentions ""...Some systems emphasize learning from pixels but without an explicitly object-based representation (Fragkiadaki et al., 2016 ..."". The paper you cite in fact emphasized the importance of having object-centric predictive models and the generalization that comes from this design choice, thus, it may be potentially not the right citation. """
1293,"""ANYTIME MINIBATCH: EXPLOITING STRAGGLERS IN ONLINE DISTRIBUTED OPTIMIZATION""","['distributed optimization', 'gradient descent', 'minibatch', 'stragglers']","""Distributed optimization is vital in solving large-scale machine learning problems. A widely-shared feature of distributed optimization techniques is the requirement that all nodes complete their assigned tasks in each computational epoch before the system can proceed to the next epoch. In such settings, slow nodes, called stragglers, can greatly slow progress. To mitigate the impact of stragglers, we propose an online distributed optimization method called Anytime Minibatch. In this approach, all nodes are given a fixed time to compute the gradients of as many data samples as possible. The result is a variable per-node minibatch size. Workers then get a fixed communication time to average their minibatch gradients via several rounds of consensus, which are then used to update primal variables via dual averaging. Anytime Minibatch prevents stragglers from holding up the system without wasting the work that stragglers can complete. We present a convergence analysis and analyze the wall time performance. Our numerical results show that our approach is up to 1.5 times faster in Amazon EC2 and it is up to five times faster when there is greater variability in compute node performance.""","""The reviewers that provided extensive reviews agree that the paper is well-written and contains solid technical material. The paper however should be edited to address specific concerns regarding theoretical and empirical aspects of this work. """
1294,"""Reduced-Gate Convolutional LSTM Design Using Predictive Coding for Next-Frame Video Prediction""","['rgcLSTM', 'convolutional LSTM', 'unsupervised learning', 'predictive coding', 'video prediction', 'moving MNIST', 'KITTI datasets', 'deep learning']","""Spatiotemporal sequence prediction is an important problem in deep learning. We study next-frame video prediction using a deep-learning-based predictive coding framework that uses convolutional, long short-term memory (convLSTM) modules. We introduce a novel reduced-gate convolutional LSTM architecture. Our reduced-gate model achieves better next-frame prediction accuracy than the original convolutional LSTM while using a smaller parameter budget, thereby reducing training time. We tested our reduced gate modules within a predictive coding architecture on the moving MNIST and KITTI datasets. We found that our reduced-gate model has a significant reduction of approximately 40 percent of the total number of training parameters and training time in comparison with the standard LSTM model which makes it attractive for hardware implementation especially on small devices.""","""The submission suggests reducing the parameters in a conv-lSTM by replacing the 3 gates in the standard LSTM with one gate. The idea is to get a more efficient convolutional LSTM and use it for video prediction. Two of the reviewers found the manuscript and description of the work difficult to follow and the justification for the proposed method lacking. Additionally, the contribution of this submission feels rather thin, and the experimental results are not very convincing: the absolute training time is too coarse of a measurement (and convergence may depend on many factors), and the improvements over PredNet seem somewhat marginal. Finally, I agree with the reviewer that mentioned that a proper comparison with baselines should be done in such a way that the number of parameters is comparable (if #params is a main claim of the paper!). It is entirely plausible that if you reduce the number of parameters in PredNet by 40% (in some other way), its performance would also benefit. With all this in mind, I do not recommend this paper be accepted at this time."""
1295,"""Variation Network: Learning High-level Attributes for Controlled Input Manipulation""","['Generative models', 'Input manipulation', 'Unsupervised feature learning', 'Variations']","""This paper presents the Variation Network (VarNet), a generative model providing means to manipulate the high-level attributes of a given input. The originality of our approach is that VarNet is not only capable of handling pre-defined attributes but can also learn the relevant attributes of the dataset by itself. These two settings can be easily combined which makes VarNet applicable for a wide variety of tasks. Further, VarNet has a sound probabilistic interpretation which grants us with a novel way to navigate in the latent spaces as well as means to control how the attributes are learned. We demonstrate experimentally that this model is capable of performing interesting input manipulation and that the learned attributes are relevant and interpretable.""","""The authors propose a generative model based on variational autoencoders that provides means to manipulate the high-level attributes of a given input. The attributes can be either pre-defined ground truth attributes or unknown attributes automatically discovered from the data. While the reviewers acknowledged the potential usefulness of the proposed approach, they raised important concerns that were viewed by AC as a critical issue: (1) very limited experimental evaluation (e.g. no baseline or ablation results, no quantitative results); comparisons on other more complex datasets and more in-depth analysis would substantially strengthen the evaluation and would allow to assess the scope of the contribution of this work see, for example, R3s suggestion to use other dataset like dSprites or CelebA, where the ground truth attributes are known; (2) lack of presentation clarity see R2s latest comment how to improve. A general consensus among reviewers and AC suggests, in its current state the manuscript is not ready for a publication. It needs clarification, more empirical studies and polish to achieve the desired goal. """
1296,"""CDeepEx: Contrastive Deep Explanations""","['Deep learning', 'Explanation', 'Network interpretation', 'Contrastive explanation']","""We propose a method which can visually explain the classification decision of deep neural networks (DNNs). There are many proposed methods in machine learning and computer vision seeking to clarify the decision of machine learning black boxes, specifically DNNs. All of these methods try to gain insight into why the network ""chose class A"" as an answer. Humans, when searching for explanations, ask two types of questions. The first question is, ""Why did you choose this answer?"" The second question asks, ""Why did you not choose answer B over A?"" The previously proposed methods are either not able to provide the latter directly or efficiently. We introduce a method capable of answering the second question both directly and efficiently. In this work, we limit the inputs to be images. In general, the proposed method generates explanations in the input space of any model capable of efficient evaluation and gradient evaluation. We provide results, showing the superiority of this approach for gaining insight into the inner representation of machine learning models.""","""Paper studies an important problem -- producing contrastive explanations (why did the network predict class B not A?). Two major concerns raised by reviewers -- the use of one learned ""black-box"" method to explain another and lack of human-studies to quantify results -- make it very difficult to accept this manuscript in its current state. We encourage the authors to incorporate reviewer feedback to make this manuscript stronger for a future submission; this is an important research topic. """
1297,"""Neural Networks with Structural Resistance to Adversarial Attacks""","['machine learning', 'adversarial attacks']","""In adversarial attacks to machine-learning classifiers, small perturbations are added to input that is correctly classified. The perturbations yield adversarial examples, which are virtually indistinguishable from the unperturbed input, and yet are misclassified. In standard neural networks used for deep learning, attackers can craft adversarial examples from most input to cause a misclassification of their choice. We introduce a new type of network units, called RBFI units, whose non-linear structure makes them inherently resistant to adversarial attacks. On permutation-invariant MNIST, in absence of adversarial attacks, networks using RBFI units match the performance of networks using sigmoid units, and are slightly below the accuracy of networks with ReLU units. When subjected to adversarial attacks based on projected gradient descent or fast gradient-sign methods, networks with RBFI units retain accuracies above 75%, while ReLU or Sigmoid see their accuracies reduced to below 1%. Further, RBFI networks trained on regular input either exceed or closely match the accuracy of sigmoid and ReLU network trained with the help of adversarial examples. The non-linear structure of RBFI units makes them difficult to train using standard gradient descent. We show that RBFI networks of RBFI units can be efficiently trained to high accuracies using pseudogradients, computed using functions especially crafted to facilitate learning instead of their true derivatives. ""","""The paper presents a novel unit making the networks intrinsically more robust to gradient-based adversarial attacks. The authors have addressed some concerns of the reviewers (e.g. regarding pseudo-gradient attacks) but experimental section could benefit from a larger scale evaluation (e.g. Imagenet)."""
1298,"""Knows When it Doesnt Know: Deep Abstaining Classifiers""","['deep learning', 'robust learning', 'abstention', 'representation learning', 'abstaining classifier', 'open-set detection']","""We introduce the deep abstaining classifier -- a deep neural network trained with a novel loss function that provides an abstention option during training. This allows the DNN to abstain on confusing or difficult-to-learn examples while improving performance on the non-abstained samples. We show that such deep abstaining classifiers can: (i) learn representations for structured noise -- where noisy training labels or confusing examples are correlated with underlying features -- and then learn to abstain based on such features; (ii) enable robust learning in the presence of arbitrary or unstructured noise by identifying noisy samples; and (iii) be used as an effective out-of-category detector that learns to reliably abstain when presented with samples from unknown classes. We provide analytical results on loss function behavior that enable automatic tuning of accuracy and coverage, and demonstrate the utility of the deep abstaining classifier using multiple image benchmarks, Results indicate significant improvement in learning in the presence of label noise.""","""The reviewers felt that the method was natural and the writing was mostly clear (although could be improved by providing better signposting and fixing typos). However, there was also general agreement that comparison to other methods was weak; one reviewer also points out that the way that the reported numbers compare the methods on different sets of data, which might be an inaccurate measure of performance (this is more minor than the overall issue of lack of comparisons). While the authors provided more comparison experiments during the author response, it is in general the responsibility of authors to have a close-to-final work at the time of submission."""
1299,"""Learning Representations in Model-Free Hierarchical Reinforcement Learning""","['Reinforcement Learning', 'Model-Free Hierarchical Reinforcement Learning', 'Subgoal Discovery', 'Unsupervised Learning', 'Temporal Difference', 'Temporal Abstraction', 'Intrinsic Motivation', 'Markov Decision Processes', 'Deep Reinforcement Learning', 'Optimization']","""Common approaches to Reinforcement Learning (RL) are seriously challenged by large-scale applications involving huge state spaces and sparse delayed reward feedback. Hierarchical Reinforcement Learning (HRL) methods attempt to address this scalability issue by learning action selection policies at multiple levels of temporal abstraction. Abstraction can be had by identifying a relatively small set of states that are likely to be useful as subgoals, in concert with the learning of corresponding skill policies to achieve those subgoals. Many approaches to subgoal discovery in HRL depend on the analysis of a model of the environment, but the need to learn such a model introduces its own problems of scale. Once subgoals are identified, skills may be learned through intrinsic motivation, introducing an internal reward signal marking subgoal attainment. In this paper, we present a novel model-free method for subgoal discovery using incremental unsupervised learning over a small memory of the most recent experiences of the agent. When combined with an intrinsic motivation learning mechanism, this method learns subgoals and skills together, based on experiences in the environment. Thus, we offer an original approach to HRL that does not require the acquisition of a model of the environment, suitable for large-scale applications. We demonstrate the efficiency of our method on two RL problems with sparse delayed feedback: a variant of the rooms environment and the ATARI 2600 game called Montezuma's Revenge. ""","""Pros: - good results on Montezuma Cons: - moderate novelty - questionable generalization - lack of ablations and analysis - lack of stronger baselines - no rebuttal The reviewers agree that the paper should be rejected in its current form, and the authors have not bothered revising it to take into account the detailed reviews."""
1300,"""ON THE USE OF CONVOLUTIONAL AUTO-ENCODER FOR INCREMENTAL CLASSIFIER LEARNING IN CONTEXT AWARE ADVERTISEMENT""","['Incremental learning', 'deep learning', 'autoencoder', 'privacy', 'convolutional neural network']","""Context Aware Advertisement (CAA) is a type of advertisement appearing on websites or mobile apps. The advertisement is targeted on specific group of users and/or the content displayed on the websites or apps. This paper focuses on classifying images displayed on the websites by incremental learning classifier with Deep Convolutional Neural Network (DCNN) especially for Context Aware Advertisement (CAA) framework. Incrementally learning new knowledge with DCNN leads to catastrophic forgetting as previously stored information is replaced with new information. To prevent catastrophic forgetting, part of previously learned knowledge should be stored for the life time of incremental classifier. Storing information for life time involves privacy and legal concerns especially in context aware advertising framework. Here, we propose an incremental classifier learning method which addresses privacy and legal concerns while taking care of catastrophic forgetting problem. We conduct experiments on different datasets including CIFAR-100. Experimental results show that proposed system achieves relatively high performance compared to the state-of-the-art incremental learning methods.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. The paper tackles an interesting and relevant problem for ICLR: incremental classifier learning applied to image data streams. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The proposed method is not clearly explained and not reproducible. In particular the contribution on top of the baseline iCaRL method is unclear. It seems to be mainly the use of CAE which is a minor change. - The experimental comparisons are incomplete. For example, in Table 4 the authors don't discuss the storage requirements of GAN and FearNet baselines. - The authors state that one of their main contributions is fullfilling privacy and legal requirements. They claim this is done by using CAEs which generate image embeddings that they store rather than the original images. However it's quite well known that a lot of data about the original images can be recovered from such embeddings (e.g. Dosovitskiy & Brox. ""Inverting visual representations with convolutional networks."" CVPR 2016.). These concerns all impacted the final decision. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. There were no major points of contention and no author feedback. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be rejected. """
1301,"""A rotation-equivariant convolutional neural network model of primary visual cortex""","['rotation equivariance', 'equivariance', 'primary visual cortex', 'V1', 'neuroscience', 'system identification']","""Classical models describe primary visual cortex (V1) as a filter bank of orientation-selective linear-nonlinear (LN) or energy models, but these models fail to predict neural responses to natural stimuli accurately. Recent work shows that convolutional neural networks (CNNs) can be trained to predict V1 activity more accurately, but it remains unclear which features are extracted by V1 neurons beyond orientation selectivity and phase invariance. Here we work towards systematically studying V1 computations by categorizing neurons into groups that perform similar computations. We present a framework for identifying common features independent of individual neurons' orientation selectivity by using a rotation-equivariant convolutional neural network, which automatically extracts every feature at multiple different orientations. We fit this rotation-equivariant CNN to responses of a population of 6000 neurons to natural images recorded in mouse primary visual cortex using two-photon imaging. We show that our rotation-equivariant network outperforms a regular CNN with the same number of feature maps and reveals a number of common features, which are shared by many V1 neurons and are pooled sparsely to predict neural activity. Our findings are a first step towards a powerful new tool to study the nonlinear functional organization of visual cortex.""","""The overall consensus after an extended discussion of the paper is that this work should be accepted to ICLR. The back-and-forth between reviewers and authors was very productive, and resulted in substantial clarification of the work, and modification (trending positive) of the reviewer scores."""
1302,"""Dataset Distillation""","['knowledge distillation', 'deep learning', 'few-shot learning', 'adversarial attack']","""Model distillation aims to distill the knowledge of a complex model into a simpler one. In this paper, we consider an alternative formulation called {\em dataset distillation}: we keep the model fixed and instead attempt to distill the knowledge from a large training dataset into a small one. The idea is to {\em synthesize} a small number of data points that do not need to come from the correct data distribution, but will, when given to the learning algorithm as training data, approximate the model trained on the original data. For example, we show that it is possible to compress pseudo-formula MNIST training images into just pseudo-formula synthetic {\em distilled images} (one per class) and achieve close to original performance with only a few steps of gradient descent, given a particular fixed network initialization. We evaluate our method in a wide range of initialization settings and with different learning objectives. Experiments on multiple datasets show the advantage of our approach compared to alternative methods in most settings. ""","""The reviewers agree that the idea for dataset distillation is novel, however it is unclear how practical it can be. The paper has been significantly improved through the addition of new baselines, however ultimately the performance is not quite good enough for the reviewers to advocate strongly on its behalf. Perhaps the paper would be better motivated by finding a realistic scenario in which it would make sense for someone to use this approach over reasonable alternatives."""
1303,"""Direct Optimization through \max$ for Discrete Variational Auto-Encoder""","['discrete variational auto encoders', 'generative models', 'perturbation models']","""Reparameterization of variational auto-encoders is an effective method for reducing the variance of their gradient estimates. However, when the latent variables are discrete, a reparameterization is problematic due to discontinuities in the discrete space. In this work, we extend the direct loss minimization technique to discrete variational auto-encoders. We first reparameterize a discrete random variable using the \max function of the Gumbel-Max perturbation model. We then use direct optimization to propagate gradients through the non-differentiable \max using two perturbed \max$ operations. ""","""The paper presents a novel gradient estimator for optimizing VAEs with discrete latents, that is based on using a Direct Loss Minimization approach (as initially developed for structured prediction) on top of the Gumble-max trick. This is an interesting and original alternative to the use of REINFORCE or Gumble Softmax. The approach is mathematically well detailed, but exposition could be easier to follow if it used a more standard notation. After clarifications by the authors, reviewers agreed that the main theorerm is correct. The proposed method is shown empirically to converge faster than Gumbel-softmax, REBAR, and RELAX baselines in number of epochs. However, as questioned by one reviewer, the proposed method appears to require many more forward passes (evaluations) of the decoder for each example.Authors replied by highlighting that an argmax can be more computationally efficient than softmax (in cases when the discrete latent space is structured), and also clarified in the paper their use of an essential computational approximation they make for discrete product spaces. These are important aspects that affect computational complexity. But they do not address the question raised about using significantly more decoder evaluations for each example. A fair comparison for sampling based gradient estimation methods should rest on actual number of decoder evaluations and on resulting timing. The paper currently does not sufficiently discuss the computational complexity of the proposed estimator against alternatives, nor take this essential aspect into account in the empirical comparisons it reports. We encourage the authors to refocus the paper and fully develop and showcase a use case where the approach could yield a clear a computational advantage, like the structured encoder setting they mentioned in the rebuttal. """
1304,"""Optimal margin Distribution Network""","['Optimal margin distribution', 'Deep neural network', 'Generalization bound']","""Recent research about margin theory has proved that maximizing the minimum margin like support vector machines does not necessarily lead to better performance, and instead, it is crucial to optimize the margin distribution. In the meantime, margin theory has been used to explain the empirical success of deep network in recent studies. In this paper, we present ODN (the Optimal margin Distribution Network), a network which embeds a loss function in regard to the optimal margin distribution. We give a theoretical analysis for our method using the PAC-Bayesian framework, which confirms the significance of the margin distribution for classification within the framework of deep networks. In addition, empirical results show that the ODN model always outperforms the baseline cross-entropy loss model consistently across different regularization situations. And our ODN model also outperforms the cross-entropy loss (Xent), hinge loss and soft hinge loss model in generalization task through limited training data.""","""The paper proposed an optimal margin distribution loss and applied PAC-Bayesian bounds that are from Sanov large deviation inequalities to give generalization error bounds for such a loss. Some interesting empirical results are shown to support the proposed method. The majority of reviewers think the papers empirical results are encouraging, although still in premature stage. The theoretical analysis is a kind of being standard. After reading the authors response and revision, the reviewers do not change much of their opinions and think the paper better undergoes systematic further study on their proposal for big improvement. Based on current ratings, the paper is therefore proposed to borderline lean rejection. """
1305,"""Conditional Inference in Pre-trained Variational Autoencoders via Cross-coding""",[],"""Variational Autoencoders (VAEs) are a popular generative model, but one in which conditional inference can be challenging. If the decomposition into query and evidence variables is fixed, conditional VAEs provide an attractive solution. To support arbitrary queries, one is generally reduced to Markov Chain Monte Carlo sampling methods that can suffer from long mixing times. In this paper, we propose an idea we term cross-coding to approximate the distribution over the latent variables after conditioning on an evidence assignment to some subset of the variables. This allows generating query samples without retraining the full VAE. We experimentally evaluate three variations of cross-coding showing that (i) can be quickly optimized for different decompositions of evidence and query and (ii) they quantitatively and qualitatively outperform Hamiltonian Monte Carlo.""","""This paper proposes to approximate arbitrary conditional distribution of a pertained VAE using variational inferences. The paper is technically sound and clearly written. A few variants of the inference network are also compared and evaluated in experiments. The main problems of the paper are as follows: 1. The motivation of training an inference network for a fixed decoder is not well explained. 2. The application of VI is standard, and offers limited novelty or significance of the proposed method. 3. The introduction of the new term cross-coding is not necessary and does not bring new insights than a standard VI method. The authors argued in the feedback that the central contribution is using augmented VI to do conditioning inference, similar to Rezende at al, but didn't address reviewers' main concerns. I encourage the authors to incorporate the reviewers' comments in a future revision, and explain why this proposed method bring significant contribution to either address a real problem or improve VI methodology."""
1306,"""Exemplar Guided Unsupervised Image-to-Image Translation with Semantic Consistency""","['image-to-image translation', 'image generation', 'domain adaptation']","""Image-to-image translation has recently received significant attention due to advances in deep learning. Most works focus on learning either a one-to-one mapping in an unsupervised way or a many-to-many mapping in a supervised way. However, a more practical setting is many-to-many mapping in an unsupervised way, which is harder due to the lack of supervision and the complex inner- and cross-domain variations. To alleviate these issues, we propose the Exemplar Guided & Semantically Consistent Image-to-image Translation (EGSC-IT) network which conditions the translation process on an exemplar image in the target domain. We assume that an image comprises of a content component which is shared across domains, and a style component specific to each domain. Under the guidance of an exemplar from the target domain we apply Adaptive Instance Normalization to the shared content component, which allows us to transfer the style information of the target domain to the source domain. To avoid semantic inconsistencies during translation that naturally appear due to the large inner- and cross-domain variations, we introduce the concept of feature masks that provide coarse semantic guidance without requiring the use of any semantic labels. Experimental results on various datasets show that EGSC-IT does not only translate the source image to diverse instances in the target domain, but also preserves the semantic consistency during the process. ""","""This paper proposes an image to image translation technique which decomposes into style and content transfer using a semantic consistency loss to encourage corresponding semantics (using feature masks) before and after translation. Performance is evaluated on a set of MNIST variants as well as from simulated to real world driving imagery. All reviewers found this paper well written with clear contribution compared to related work by focusing on the problem when one-to-one mappings are not available across two domains which also have multimodal content or sub-style. The main weakness as discussed by the reviewers relates to the experiments and whether or not the set provided does effectively validate the proposed approach. The authors argue their use of MNIST as a toy problem but with full control to clearly validate their approach. Their semantic segmentation experiment shows modest performance improvement. Based on the experiments as is and the relative novelty of the proposed approach, the AC recommends poster and encourages the authors to extend their analysis of the current results in a final version."""
1307,"""Neural Graph Evolution: Towards Efficient Automatic Robot Design""","['Reinforcement learning', 'graph neural networks', 'robotics', 'deep learning', 'transfer learning']","""Despite the recent successes in robotic locomotion control, the design of robot relies heavily on human engineering. Automatic robot design has been a long studied subject, but the recent progress has been slowed due to the large combinatorial search space and the difficulty in evaluating the found candidates. To address the two challenges, we formulate automatic robot design as a graph search problem and perform evolution search in graph space. We propose Neural Graph Evolution (NGE), which performs selection on current candidates and evolves new ones iteratively. Different from previous approaches, NGE uses graph neural networks to parameterize the control policies, which reduces evaluation cost on new candidates with the help of skill transfer from previously evaluated designs. In addition, NGE applies Graph Mutation with Uncertainty (GM-UC) by incorporating model uncertainty, which reduces the search space by balancing exploration and exploitation. We show that NGE significantly outperforms previous methods by an order of magnitude. As shown in experiments, NGE is the first algorithm that can automatically discover kinematically preferred robotic graph structures, such as a fish with two symmetrical flat side-fins and a tail, or a cheetah with athletic front and back legs. Instead of using thousands of cores for weeks, NGE efficiently solves searching problem within a day on a single 64 CPU-core Amazon EC2 machine. ""","""Lean in favor Strengths: The paper tackles the difficult problem of automatic robot design. The approach uses graph neural networks to parameterize the control policies, which allows for weight sharing / transfer to new policies even as the topology changes. Understanding how to efficiently explore through non-differentiable changes to the body is an important problem (AC). The authors will release the code and environments, which will be useful in an area where there are currently no good baselines (AC). Weaknesses: There are concerns (particularly R2, R1) over the lack of a strong baseline, and with the results being demonstrated on a limited number of environments (R1) (fish, 2D walker). In response, the authors clarified the nomenclature and description of a number of the baselines, and added others. AC: there is no submitted video (searches for ""video"" on the PDF text produces no hits); this is seen by the AC as being a real limitation from the perspective of evaluation. AC agrees with some of the reviewer remarks that some of the original stated claims are too strong. AC: the simplified fluid model of Mujoco (pseudo-url) is unable to model the fluid state, in particular the induced fluid vortices that are responsible for a good portion of fish locomotion, i.e., ""Passive and active flow control by swimming fishes and mammals"" and other papers. Acknowledging this kind of limitation will make the paper stronger, not weaker; the ML community can learn from much existing work at the interface of biology and fluid mechancis. There remain points of contention, i.e., the sufficiency of the baselines. However, the reviewers R2 and R3 have not responded to the detailed replies from the authors, including additional baselines (totaling 5 at present) and pointing out that baselines such as CMA-ES (R2) in a continuous space and therefore do not translate in any obvious way to the given problem at hand. On balance, with the additional baselines and related clarifications, the AC feels that this paper makes a useful and valid contribution to the field, and will help establish a benchmark in an important area. The authors are strongly encouraged to further state caveats and limitations, and to emphasize why some candidate baseline methods are not readily applicable. """
1308,"""MILE: A Multi-Level Framework for Scalable Graph Embedding""","['Network Embedding', 'Graph Convolutional Networks', 'Deep Learning']","""Recently there has been a surge of interest in designing graph embedding methods. Few, if any, can scale to a large-sized graph with millions of nodes due to both computational complexity and memory requirements. In this paper, we relax this limitation by introducing the MultI-Level Embedding (MILE) framework a generic methodology allowing contemporary graph embedding methods to scale to large graphs. MILE repeatedly coarsens the graph into smaller ones using a hybrid matching technique to maintain the backbone structure of the graph. It then applies existing embedding methods on the coarsest graph and refines the embeddings to the original graph through a novel graph convolution neural network that it learns. The proposed MILE framework is agnostic to the underlying graph embedding techniques and can be applied to many existing graph embedding methods without modifying them. We employ our framework on several popular graph embedding techniques and conduct embedding for real-world graphs. Experimental results on five large-scale datasets demonstrate that MILE significantly boosts the speed (order of magnitude) of graph embedding while also often generating embeddings of better quality for the task of node classification. MILE can comfortably scale to a graph with 9 million nodes and 40 million edges, on which existing methods run out of memory or take too long to compute on a modern workstation.""","""Significant spread of scores across the reviewers and unfortunately not much discussion despite prompts from the area chair and the authors. The most positive reviewer is the least confident one. Very close to the decision boundary but after careful consideration by the senior PCs just below the acceptance threshold. There is significant literature already on this topic. The ""thought delta"" created by this paper and the empirical results are also not sufficient for acceptance."""
1309,"""Information asymmetry in KL-regularized RL""","['Deep Reinforcement Learning', 'Continuous Control', 'RL as Inference']","""Many real world tasks exhibit rich structure that is repeated across different parts of the state space or in time. In this work we study the possibility of leveraging such repeated structure to speed up and regularize learning. We start from the KL regularized expected reward objective which introduces an additional component, a default policy. Instead of relying on a fixed default policy, we learn it from data. But crucially, we restrict the amount of information the default policy receives, forcing it to learn reusable behaviors that help the policy learn faster. We formalize this strategy and discuss connections to information bottleneck approaches and to the variational EM algorithm. We present empirical results in both discrete and continuous action domains and demonstrate that, for certain tasks, learning a default policy alongside the policy can significantly speed up and improve learning. Please watch the video demonstrating learned experts and default policies on several continuous control tasks ( pseudo-url ).""","""Strengths The paper introduces a promising and novel idea, i.e., regularizing RL via an informationally asymmetric default policy The paper is well written. It has solid and extensive experimental results. Weaknesses There is a lack of benefit on dense-reward problems as a limitation, which the authors further acknowledge as a limitation. There also some similarities to HRL approaches. A lack of theoretical results is also suggested. To be fair, the paper makes a number of connections with various bits of theory, although it perhaps does not directly result in any new theoretical analysis. A concern of one reviewer is the need for extensive compute, and making comparisons to stronger (maxent) baselines. The authors provide a convincing reply on these issues. Points of Contention While the scores are non-uniform (7,7,5), the most critical review, R1(5), is in fact quite positive on many aspects of the paper, i.e., ""this paper would have good impact in coming up with new learning algorithms which are inspired from cognitive science literature as well as mathematically grounded."" The specific critiques of R1 were covered in detail by the authors. Overall The paper presents a novel and fairly intuitive idea, with very solid experimental results. While the methods has theoretical results, the results themselves are more experimental than theoretic. The reviewers are largely enthused about the paper. The AC recommends acceptance as a poster. """
1310,"""Meta-Learning with Latent Embedding Optimization""","['meta-learning', 'few-shot', 'miniImageNet', 'tieredImageNet', 'hypernetworks', 'generative', 'latent embedding', 'optimization']","""Gradient-based meta-learning techniques are both widely applicable and proficient at solving challenging few-shot learning and fast adaptation problems. However, they have practical difficulties when operating on high-dimensional parameter spaces in extreme low-data regimes. We show that it is possible to bypass these limitations by learning a data-dependent latent generative representation of model parameters, and performing gradient-based meta-learning in this low-dimensional latent space. The resulting approach, latent embedding optimization (LEO), decouples the gradient-based adaptation procedure from the underlying high-dimensional space of model parameters. Our evaluation shows that LEO can achieve state-of-the-art performance on the competitive miniImageNet and tieredImageNet few-shot classification tasks. Further analysis indicates LEO is able to capture uncertainty in the data, and can perform adaptation more effectively by optimizing in latent space.""","""This work builds on MAML by (1) switching from a single underlying set of parameters to a distribution in a latent lower-dimensional space, and (2) conditioning the initial parameter of each subproblem on the input data. All reviewers agree that the solid experimental results are impressive, with careful ablation studies to show how conditional parameter generation and optimization in the lower-dimensional space both contribute to the performance. While there were some initial concerns on clarity and experimental details, we feel the revised version has addressed those in a satisfying way."""
1311,"""Generating Liquid Simulations with Deformation-aware Neural Networks""","['deformation learning', 'spatial transformer networks', 'fluid simulation']","""We propose a novel approach for deformation-aware neural networks that learn the weighting and synthesis of dense volumetric deformation fields. Our method specifically targets the space-time representation of physical surfaces from liquid simulations. Liquids exhibit highly complex, non-linear behavior under changing simulation conditions such as different initial conditions. Our algorithm captures these complex phenomena in two stages: a first neural network computes a weighting function for a set of pre-computed deformations, while a second network directly generates a deformation field for refining the surface. Key for successful training runs in this setting is a suitable loss function that encodes the effect of the deformations, and a robust calculation of the corresponding gradients. To demonstrate the effectiveness of our approach, we showcase our method with several complex examples of flowing liquids with topology changes. Our representation makes it possible to rapidly generate the desired implicit surfaces. We have implemented a mobile application to demonstrate that real-time interactions with complex liquid effects are possible with our approach.""","""This paper presents a novel method for synthesizing fluid simulations, constrained to a set of parameterized variations, such as the size and position of a water ball that is dropped. The results are solid; there is little related work to compare to, in terms of methods that can ""compute""/recall simulations at that speed. The method is 2000x faster than the orginal simulations. This comes with the caveats that: (a) the results are specific to the given set of parameterized environments; the method is learning a compressed version of the original animations; (b) there is a loss of accuracy, and therefore also a loss of visual plausibility. The AC notes that the paper should use the ICLR format for citations, i.e., ""(foo et al.)"" rather than ""(19)"". The AC also suggests that limitations should also be clearly documented, i.e., as seen from the perspective of those working in the fluid simulation domain. The principle (and only?) contentious issue relates to the suitability of the paper for the ICLR audience, given its focus on the specific domain of fluid simulations. The AC is of two minds on this: (i) the fluid simulation domain has different characteristics to other domains, and thu understanding the ICLR audience can benefit from the specific nature of the predictive problems that come the fluid simulation domain; new problems can drive new methods. There is a loose connection between the given work and residual nets, and of course res-nets have also been recently reconceptualized as PDEs. (ii) it's not clear how much the ICLR audience will get out of the specific solutions being described; it requires understanding spatial transformer networks and a number of other domain-specific issues. A problem with this type of paper in terms of graphics/SIGGRAPH is that it can also be seen as ""falling short"" there, simply because it is not yet competitive in terms of visual quality or the generality of fluid simulators; it really fulfills a different niche than classical fluid simulators. The AC leans slightly in favor of acceptance, but is otherwise on the fence. """
1312,"""BabyAI: A Platform to Study the Sample Efficiency of Grounded Language Learning""","['language', 'learning', 'efficiency', 'imitation learning', 'reinforcement learning']","""Allowing humans to interactively train artificial agents to understand language instructions is desirable for both practical and scientific reasons. Though, given the lack of sample efficiency in current learning methods, reaching this goal may require substantial research efforts. We introduce the BabyAI research platform, with the goal of supporting investigations towards including humans in the loop for grounded language learning. The BabyAI platform comprises an extensible suite of 19 levels of increasing difficulty. Each level gradually leads the agent towards acquiring a combinatorially rich synthetic language, which is a proper subset of English. The platform also provides a hand-crafted bot agent, which simulates a human teacher. We report estimated amount of supervision required for training neural reinforcement and behavioral-cloning agents on some BabyAI levels. We put forward strong evidence that current deep learning methods are not yet sufficiently sample-efficient in the context of learning a language with compositional properties.""","""This paper presents ""BabyAI"", a research platform to support grounded language learning. The platform supports a suite of 19 levels, based on *synthetic* natural language of increasing difficulties. The platform uniquely supports simulated ""human-in-the-loop"" learning, where a human teacher is simulated as a heuristic expert agent speaking in synthetic language. Pros: A new platform to support grounded natural language learning with 19 levels of increasing difficulties. The platform also supports a heuristic expert agent to simulate a human teacher, which aims to mimic ""human-in-the-loop"" learning. The platform seems to be the result of a substantial amount of engineering, thus nontrivial to develop. While not representing the real communication or true natural language, the platform is likely to be useful for DL/RL researchers to perform prototype research on interactive and grounded language learning. Cons: Everything in the presented platform is based on synthetic natural language. While the use of synthetic language is not entirely satisfactory, such limit is relatively common among the simulation environments available today, and lifting that limitation is not straightforward. The primary contribution of the paper is a new platform (resource). There are no insights or methods. Verdict: Potential weak accept. The potential impact of this work is that the platform will likely be useful for DL/RL research on interactive and grounded language learning."""
1313,"""Text Infilling""","['text generation', 'text infilling', 'self attention', 'sequence to sequence']","""Recent years have seen remarkable progress of text generation in different contexts, including the most common setting of generating text from scratch, the increasingly popular paradigm of retrieval and editing, and others. Text infilling, which fills missing text portions of a sentence or paragraph, is also of numerous use in real life. Previous work has focused on restricted settings, by either assuming single word per missing portion, or limiting to single missing portion to the end of text. This paper studies the general task of text infilling, where the input text can have an arbitrary number of portions to be filled, each of which may require an arbitrary unknown number of tokens. We develop a self-attention model with segment-aware position encoding for precise global context modeling. We further create a variety of supervised data by masking out text in different domains with varying missing ratios and mask strategies. Extensive experiments show the proposed model performs significantly better than other methods, and generates meaningful text patches.""","""although the problem of text infilling itself is interesting, all the reviewers were not certain about the extent of experiments and how they shed light on whether, how and why the proposed approach is better than existing approaches. """
1314,"""Local Image-to-Image Translation via Pixel-wise Highway Adaptive Instance Normalization""","['image to image translation', 'image translation', 'exemplar', 'mutlimodal']","""Recently, image-to-image translation has seen a significant success. Among many approaches, image translation based on an exemplar image, which contains the target style information, has been popular, owing to its capability to handle multimodality as well as its suitability for practical use. However, most of the existing methods extract the style information from an entire exemplar and apply it to the entire input image, which introduces excessive image translation in irrelevant image regions. In response, this paper proposes a novel approach that jointly extracts out the local masks of the input image and the exemplar as targeted regions to be involved for image translation. In particular, the main novelty of our model lies in (1) co-segmentation networks for local mask generation and (2) the local mask-based highway adaptive instance normalization technique. We demonstrate the quantitative and the qualitative evaluation results to show the advantages of our proposed approach. Finally, the code is available at pseudo-url""","""The paper received mixed ratings. The proposed idea is quite reasonable but also sounds somewhat incremental. While the idea of separating foreground/background is reasonable, it also limits the applicability of the proposed method (i.e., the method is only demonstrated on aligned face images). In addition, combining AdaIn with foreground mask is a reasonable idea but doesnt sound groundbreakingly novel. The comparison against StarGAN looks quite anecdotal and the proposed method seems to cause only hairstyle changes (but transfer with other attributes are not obvious). In addition, please refer to detailed reviewers comments for other concerns. Overall, it sounds like a good engineering paper that might be better fit to computer vision venue, but experimental validation seems somewhat preliminary and its unclear how much novel insight and general technical contributions that this work provides. """
1315,"""Neural Network Regression with Beta, Dirichlet, and Dirichlet-Multinomial Outputs""","['regression', 'uncertainty', 'deep learning']","""We propose a method for quantifying uncertainty in neural network regression models when the targets are real values on a pseudo-formula -dimensional simplex, such as probabilities. We show that each target can be modeled as a sample from a Dirichlet distribution, where the parameters of the Dirichlet are provided by the output of a neural network, and that the combined model can be trained using the gradient of the data likelihood. This approach provides interpretable predictions in the form of multidimensional distributions, rather than point estimates, from which one can obtain confidence intervals or quantify risk in decision making. Furthermore, we show that the same approach can be used to model targets in the form of empirical counts as samples from the Dirichlet-multinomial compound distribution. In experiments, we verify that our approach provides these benefits without harming the performance of the point estimate predictions on two diverse applications: (1) distilling deep convolutional networks trained on CIFAR-100, and (2) predicting the location of particle collisions in the XENON1T Dark Matter detector.""","""This paper proposes to quantify the uncertainty of neural network models with Beta, Dirichlet and Dirichlet-Multinomial likelihood. This paper is clearly written with a sound main idea. However, it is a common practice to model different types of data with different likelihood, although the proposed distributions are not usually used for network output. All the reviewers therefore considered this paper to be of limited novelty. Reviewer 2 also had a concern about the mixed experimental results of the proposed method. Reviewer 3 raised the concern that this paper did not model the uncertainty of prediction from the uncertainty of the model parameters. It is a common consideration in a Bayesian approach and I encourage the authors to discussed different sources of uncertainty in future revisions."""
1316,"""KnockoffGAN: Generating Knockoffs for Feature Selection using Generative Adversarial Networks""","['Knockoff model', 'Feature selection', 'False discovery rate control', 'Generative Adversarial networks']","""Feature selection is a pervasive problem. The discovery of relevant features can be as important for performing a particular task (such as to avoid overfitting in prediction) as it can be for understanding the underlying processes governing the true label (such as discovering relevant genetic factors for a disease). Machine learning driven feature selection can enable discovery from large, high-dimensional, non-linear observational datasets by creating a subset of features for experts to focus on. In order to use expert time most efficiently, we need a principled methodology capable of controlling the False Discovery Rate. In this work, we build on the promising Knockoff framework by developing a flexible knockoff generation model. We adapt the Generative Adversarial Networks framework to allow us to generate knockoffs with no assumptions on the feature distribution. Our model consists of 4 networks, a generator, a discriminator, a stability network and a power network. We demonstrate the capability of our model to perform feature selection, showing that it performs as well as the originally proposed knockoff generation model in the Gaussian setting and that it outperforms the original model in non-Gaussian settings, including on a real-world dataset.""","""The paper presents a novel strategy for statistically motivated feature selection i.e. aimed at controlling the false discovery rate. This is achieved by extending knockoffs to complex predictive models and complex distributions via (multiple) generative adversarial networks. The reviewers and ACs noted weakness in the original submission which seems to have been fixed after the rebuttal period -- primary related to missing experimental details. There was also some concern (as is common with inferential papers) that the claims are difficult to evaluate on real data, as the ground truth is unknown. To this end, the authors provide empirical results with simulated data that address this issue. There is also some concern that more complex predictive models are not evaluated. Overall the reviewers and AC have a positive opinion of this paper and recommend acceptance."""
1317,"""How Training Data Affect the Accuracy and Robustness of Neural Networks for Image Classification""","['Adversarial attacks', 'Robustness', 'CW', 'I-FGSM']","""Recent work has demonstrated the lack of robustness of well-trained deep neural networks (DNNs) to adversarial examples. For example, visually indistinguishable perturbations, when mixed with an original image, can easily lead deep learning models to misclassifications. In light of a recent study on the mutual influence between robustness and accuracy over 18 different ImageNet models, this paper investigates how training data affect the accuracy and robustness of deep neural networks. We conduct extensive experiments on four different datasets, including CIFAR-10, MNIST, STL-10, and Tiny ImageNet, with several representative neural networks. Our results reveal previously unknown phenomena that exist between the size of training data and characteristics of the resulting models. In particular, besides confirming that the model accuracy improves as the amount of training data increases, we also observe that the model robustness improves initially, but there exists a turning point after which robustness starts to decrease. How and when such turning points occur vary for different neural networks and different datasets.""","""The reviewers conclude the paper does not bring an important contribution compared to existing work. The experimental study can also be improved. """
1318,"""Learning data-derived privacy preserving representations from information metrics""","['Machine learning', 'privacy', 'adversarial training', 'information theory', 'data-driven privacy']","""It is clear that users should own and control their data and privacy. Utility providers are also becoming more interested in guaranteeing data privacy. Therefore, users and providers can and should collaborate in privacy protecting challenges, and this paper addresses this new paradigm. We propose a framework where the user controls what characteristics of the data they want to share (utility) and what they want to keep private (secret), without necessarily asking the utility provider to change its existing machine learning algorithms. We first analyze the space of privacy-preserving representations and derive natural information-theoretic bounds on the utility-privacy trade-off when disclosing a sanitized version of the data X. We present explicit learning architectures to learn privacy-preserving representations that approach this bound in a data-driven fashion. We describe important use-case scenarios where the utility providers are willing to collaborate with the sanitization process. We study space-preserving transformations where the utility provider can use the same algorithm on original and sanitized data, a critical and novel attribute to help service providers accommodate varying privacy requirements with a single set of utility algorithms. We illustrate this framework through the implementation of three use cases; subject-within-subject, where we tackle the problem of having a face identity detector that works only on a consenting subset of users, an important application, for example, for mobile devices activated by face recognition; gender-and-subject, where we preserve facial verification while hiding the gender attribute for users who choose to do so; and emotion-and-gender, where we hide independent variables, as is the case of hiding gender while preserving emotion detection.""","""This paper addresses data sanitization, using a KL-divergence-based notion of privacy. While an interesting goal, the use of average-case as opposed to worst-case privacy misses the point of privacy guarantees, which must protect all individuals. (Otherwise, individuals with truly anomalous private values may be the only ones who opt for the highest levels of privacy, yet this situation will itself leak some information about their private values). """
1319,"""Classifier-agnostic saliency map extraction""","['saliency maps', 'explainable AI', 'convolutional neural networks', 'generative adversarial training', 'classification']","""Extracting saliency maps, which indicate parts of the image important to classification, requires many tricks to achieve satisfactory performance when using classifier-dependent methods. Instead, we propose classifier-agnostic saliency map extraction, which finds all parts of the image that any classifier could use, not just one given in advance. We observe that the proposed approach extracts higher quality saliency maps and outperforms existing weakly-supervised localization techniques, setting the new state of the art result on the ImageNet dataset.""","""{418}; {Classifier-agnostic saliency map extraction}; {Avg: 4.33}; {} 1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. The paper is well-written and the method is simple, effective, and well-justified. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. 1. The introduction, in particular the last row of pg 1, implies that this work is the first to show that a class-agnostic saliency estimation method can produce higher-quality saliency maps than class-dependent ones. However, Fan et al. have already shown this. For this reason, AR1 recommended that the authors reword the introduction to reflect prior work on this aspect but the authors declined to do so. The AC would have liked to see a discussion of how the different points of view of the two works (robustness to corruption vs class-agnosticism) both address the same issue (poor segmentation of the salient image regions). 2. The work of Fan et al has a very similar approach and a deeper comparison is needed. While the authors dedicated two paragraphs of discussion to this work, they should have gone further. For example, the work of Fan et al. uses a very simple saliency map extraction network and it's unclear how much this impacts their performance when compared to the proposed method, which uses ResNet50. The AC agrees with the authors that re-implementing the method of Fan et al. is asking a lot but a discussion of the potential impact would have sufficed. 3. The authors didn't mention at all the vast body of work on salient object detection (for a somewhat recent review see Borji et al. ""Salient object detection: A benchmark."" IEEE TIP). The differences to this line of work should have been discussed. Points 1 and 2 were particularly salient for the final decision. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. Two major points of contention were: - The discussion of differences between the proposed method and the method of Fan et al. - The fairness of the comparison to Fan et al. AR1 felt that the paper was deficient on both counts (AR2 had similar concerns) and the authors disagreed, arguing that the discussion was complete and the quantitative comparison fair. The AC was sympathetic to these concerns and found the authors' responses to be dismissive of those concerns. In particular, the AC agrees that the paper, as currently organized, minimizes the degree to which the work is derived from Fan et al. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be rejected. """
1320,"""Object detection deep learning networks for Optical Character Recognition""","['OCR', 'object detection', 'RCNN', 'Yolo']","""In this article, we show how we applied a simple approach coming from deep learning networks for object detection to the task of optical character recognition in order to build image features taylored for documents. In contrast to scene text reading in natural images using networks pretrained on ImageNet, our document reading is performed with small networks inspired by MNIST digit recognition challenge, at a small computational budget and a small stride. The object detection modern frameworks allow a direct end-to-end training, with no other algorithm than the deep learning and the non-max-suppression algorithm to filter the duplicate predictions. The trained weights can be used for higher level models, such as, for example, document classification, or document segmentation. ""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. The paper tackles an interesting and relevant problem for ICLR: optical character recognition in document images. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The authors propose to use small networks to localize text in document images, claiming that for document images smaller networks work better than standard SOTA networks for scene text. As pointed out in the reviews, the authors didn't make any comparisons to SOTA object detection networks (trained either on scene text or on document images) so their central claim has not been experimentally verified. - The reviewers were unanimous that the work lacks novelty as object detection pipelines have already been used for OCR so a contribution of considering smaller detection networks is minor. - There were serious issues with formatting and clarity. These three issues all informed the final decision. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. There were no major points of contention and no author feedback. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers reached a consensus that the paper should be rejected. """
1321,"""Auto-Encoding Knockoff Generator for FDR Controlled Variable Selection""","['Model-X Knockoff Generator', 'model-free FDR control', 'variable selection']","""A new statistical procedure (Candes,2018) has provided a way to identify important factors using any supervised learning method controlling for FDR. This line of research has shown great potential to expand the horizon of machine learning methods beyond the task of prediction, to serve the broader need for scientific researches for interpretable findings. However, the lack of a practical and flexible method to generate knockoffs remains the major obstacle for wide application of Model-X procedure. This paper fills in the gap by proposing a model-free knockoff generator which approximates the correlation structure between features through latent variable representation. We demonstrate our proposed method can achieve FDR control and better power than two existing methods in various simulated settings and a real data example for finding mutations associated with drug resistance in HIV-1 patients. ""","""The paper presents a novel strategy for statistically motivated feature selection i.e. aimed at controlling the false discovery rate. This is achieved by extending knockoffs to complex predictive models and complex distributions; specifically using a variational auto-encoder to generate conditionally independent data samples with the same joint distribution. The reviewers and ACs noted weakness in the original submission related to the clarity of the presentation, relationship to already published work, and concerns about the correctness of some main claims (this mostly seems to have been fixed after the rebuttal period). There are additional concerns about a thorough evaluation of the claimed results, as the ground truth is unknown. The authors (and reviewers) also note a similar paper submitted to ICLR with the same goal but implemented using GANs. Nevertheless, there remain significant concerns about the clarity of the presentation."""
1322,"""Identifying Bias in AI using Simulation""","['Bias', 'Simulation', 'Optimization', 'Face Detection']","""Machine learned models exhibit bias, often because the datasets used to train them are biased. This presents a serious problem for the deployment of such technology, as the resulting models might perform poorly on populations that are minorities within the training set and ultimately present higher risks to them. We propose to use high-fidelity computer simulations to interrogate and diagnose biases within ML classifiers. We present a framework that leverages Bayesian parameter search to efficiently characterize the high dimensional feature space and more quickly identify weakness in performance. We apply our approach to an example domain, face detection, and show that it can be used to help identify demographic biases in commercial face application programming interfaces (APIs).""",""" The paper addresses an important problem of detecting biases in classifiers (e.g. in face detection), using simulation tools with Bayesian parameter search. While the direction of research and the presented approach seem to be practically useful, there were several concerns raised by the reviewers regarding strengthening the results (e.g., beyond single avatar, etc), and suggestions on possibly a more applied conference as a better venue. While thourough rebuttals by the authors addressed some of these concerns, which increased some ratings, overall, the paper was still in the borderline range. We hope the suggestions and comments of the reviewers can help to improve the paper. """
1323,"""The Case for Full-Matrix Adaptive Regularization""","['adaptive regularization', 'non-convex optimization']","""Adaptive regularization methods pre-multiply a descent direction by a preconditioning matrix. Due to the large number of parameters of machine learning problems, full-matrix preconditioning methods are prohibitively expensive. We show how to modify full-matrix adaptive regularization in order to make it practical and effective. We also provide novel theoretical analysis for adaptive regularization in non-convex optimization settings. The core of our algorithm, termed GGT, consists of efficient inverse computation of square roots of low-rank matrices. Our preliminary experiments underscore improved convergence rate of GGT across a variety of synthetic tasks and standard deep learning benchmarks.""","""This paper shows how to implement a low-rank version of the Adagrad preconditioner in a GPU-friendly manner. A theoretical analysis of a ""hard-window"" version of the proposed algorithm demonstrates that it is not worse than SGD at finding a first-order stationary point in the nonconvex setting. Experiments on CIFAR-10 classification using a ConvNet and Penn Treebank character-level language modeling using an LSTM show that the proposed algorithm improves training loss faster than SGD, Adagrad, and Adam (measuring time in epochs) and has better generalization performance on the language modeling task. However, if wall-clock time is used to measure time, there is no speedup for the ConvNet model, but there is for the recurrent model. The reviewers liked the simplicity of the approach and greatly appreciated the elegant visualization of the eigenspectrum in Figure 4. But, even after discussion, critical concerns remained about the need for more focus on the practical tradeoffs between per-iteration improvement and per-second improvement in the loss and the need for a more careful analysis of the relationship of this method to stochastic L-BFGS. A more minor concern is that the term ""full-matrix regularization"" seems somewhat deceptive when the actual regularization is low rank. The AC also suggests that, if the authors plan to revise this paper and submit it to another venue, they consider the relationship between GGT and the various stochastic natural gradient optimization algorithms in the literature that differ from GGT primarily in the exponent on the Gram matrix."""
1324,"""Feature Intertwiner for Object Detection""","['feature learning', 'computer vision', 'deep learning']","""A well-trained model should classify objects with unanimous score for every category. This requires the high-level semantic features should be alike among samples, despite a wide span in resolution, texture, deformation, etc. Previous works focus on re-designing the loss function or proposing new regularization constraints on the loss. In this paper, we address this problem via a new perspective. For each category, it is assumed that there are two sets in the feature space: one with more reliable information and the other with less reliable source. We argue that the reliable set could guide the feature learning of the less reliable set during training - in spirit of student mimicking teachers behavior and thus pushing towards a more compact class centroid in the high-dimensional space. Such a scheme also benefits the reliable set since samples become more closer within the same category - implying that it is easilier for the classifier to identify. We refer to this mutual learning process as feature intertwiner and embed the spirit into object detection. It is well-known that objects of low resolution are more difficult to detect due to the loss of detailed information during network forward pass. We thus regard objects of high resolution as the reliable set and objects of low resolution as the less reliable set. Specifically, an intertwiner is achieved by minimizing the distribution divergence between two sets. We design a historical buffer to represent all previous samples in the reliable set and utilize them to guide the feature learning of the less reliable set. The design of obtaining an effective feature representation for the reliable set is further investigated, where we introduce the optimal transport (OT) algorithm into the framework. Samples in the less reliable set are better aligned with the reliable set with aid of OT metric. Incorporated with such a plug-and-play intertwiner, we achieve an evident improvement over previous state-of-the-arts on the COCO object detection benchmark.""","""The paper proposes an interesting idea (using ""reliable"" samples to guide the learning of ""less reliable"" samples). The experimental results and detailed analysis show clear improvement in object detection, especially small objects. On the weak side, the paper seems to focus quite heavily on the object detection problem, and how to divide the data into reliable/less-reliable samples is domain-specific (it makes sense for object detection tasks, but it's unclear how to do this for general scenarios). As the authors promise, it will make more sense to change the title to ""Feature Intertwiner for Object Detection"" to alleviate such criticisms. Given this said, I think this paper is over the acceptance threshold and would be of interest to many researchers. """
1325,"""Localized random projections challenge benchmarks for bio-plausible deep learning""","['deep learning', 'bio-plausibility', 'random projections', 'spiking networks', 'unsupervised learning', 'MNIST', 'spike timing dependent plasticity']","""Similar to models of brain-like computation, artificial deep neural networks rely on distributed coding, parallel processing and plastic synaptic weights. Training deep neural networks with the error-backpropagation algorithm, however, is considered bio-implausible. An appealing alternative to training deep neural networks is to use one or a few hidden layers with fixed random weights or trained with an unsupervised, local learning rule and train a single readout layer with a supervised, local learning rule. We find that a network of leaky-integrate-andfire neurons with fixed random, localized receptive fields in the hidden layer and spike timing dependent plasticity to train the readout layer achieves 98.1% test accuracy on MNIST, which is close to the optimal result achievable with error-backpropagation in non-convolutional networks of rate neurons with one hidden layer. To support the design choices of the spiking network, we systematically compare the classification performance of rate networks with a single hidden layer, where the weights of this layer are either random and fixed, trained with unsupervised Principal Component Analysis or Sparse Coding, or trained with the backpropagation algorithm. This comparison revealed, first, that unsupervised learning does not lead to better performance than fixed random projections for large hidden layers on digit classification (MNIST) and object recognition (CIFAR10); second, networks with random projections and localized receptive fields perform significantly better than networks with all-to-all connectivity and almost reach the performance of networks trained with the backpropagation algorithm. The performance of these simple random projection networks is comparable to most current models of bio-plausible deep learning and thus provides an interesting benchmark for future approaches.""","""This paper presents a biologically plausible architecture and learning algorithm for deep neural networks. The authors then go on to show that the proposed approach achieves competitive results on the MNIST dataset. In general, the reviewers found that the paper was well written and the motivation compelling. However, they were not convinced by the experiments, analysis or comparison to existing literature. In particular, they did not find MNIST to be a particularly interesting problem and had questions about the novelty of this approach over past literature. Perhaps the paper would be more impactful and convincing if the authors demonstrated competitive performance on a more challenging problem (e.g. machine translation, speech recognition or imagenet) using a biologically plausible approach. """
1326,"""Variational Bayesian Phylogenetic Inference""","['Bayesian phylogenetic inference', 'Variational inference', 'Subsplit Bayesian networks']","""Bayesian phylogenetic inference is currently done via Markov chain Monte Carlo with simple mechanisms for proposing new states, which hinders exploration efficiency and often requires long runs to deliver accurate posterior estimates. In this paper we present an alternative approach: a variational framework for Bayesian phylogenetic analysis. We approximate the true posterior using an expressive graphical model for tree distributions, called a subsplit Bayesian network, together with appropriate branch length distributions. We train the variational approximation via stochastic gradient ascent and adopt multi-sample based gradient estimators for different latent variables separately to handle the composite latent space of phylogenetic models. We show that our structured variational approximations are flexible enough to provide comparable posterior estimation to MCMC, while requiring less computation due to a more efficient tree exploration mechanism enabled by variational inference. Moreover, the variational approximations can be readily used for further statistical analysis such as marginal likelihood estimation for model comparison via importance sampling. Experiments on both synthetic data and real data Bayesian phylogenetic inference problems demonstrate the effectiveness and efficiency of our methods.""","""The reviewers lean to accept, and the authors clearly put a significant amount of time into their response. I will also lean to accept. However, the comments of reviewer 2 should be taken seriously, and addressed if possible, including an attempt to cut the paper length down."""
1327,"""DANA: Scalable Out-of-the-box Distributed ASGD Without Retuning""","['distributed', 'asynchronous', 'gradient staleness', 'nesterov', 'optimization', 'out-of-the-box', 'stochastic gradient descent', 'sgd', 'imagenet', 'distributed training', 'neural networks', 'deep learning']","""Distributed computing can significantly reduce the training time of neural networks. Despite its potential, however, distributed training has not been widely adopted: scaling the training process is difficult, and existing SGD methods require substantial tuning of hyperparameters and learning schedules to achieve sufficient accuracy when increasing the number of workers. In practice, such tuning can be prohibitively expensive given the huge number of potential hyperparameter configurations and the effort required to test each one. We propose DANA, a novel approach that scales out-of-the-box to large clusters using the same hyperparameters and learning schedule optimized for training on a single worker, while maintaining similar final accuracy without additional overhead. DANA estimates the future value of model parameters by adapting Nesterov Accelerated Gradient to a distributed setting, and so mitigates the effect of gradient staleness, one of the main difficulties in scaling SGD to more workers. Evaluation on three state-of-the-art network architectures and three datasets shows that DANA scales as well as or better than existing work without having to tune any hyperparameters or tweak the learning schedule. For example, DANA achieves 75.73% accuracy on ImageNet when training ResNet-50 with 16 workers, similar to the non-distributed baseline.""","""The paper needs more revisions and better presentation of empirical study."""
1328,"""DiffraNet: Automatic Classification of Serial Crystallography Diffraction Patterns""","['Serial Crystallography', 'Deep Learning', 'Image Classification']","""Serial crystallography is the field of science that studies the structure and properties of crystals via diffraction patterns. In this paper, we introduce a new serial crystallography dataset generated through the use of a simulator; the synthetic images are labeled and they are both scalable and accurate. The resulting synthetic dataset is called DiffraNet, and it is composed of 25,000 512x512 grayscale labeled images. We explore several computer vision approaches for classification on DiffraNet such as standard feature extraction algorithms associated with Random Forests and Support Vector Machines but also an end-to-end CNN topology dubbed DeepFreak tailored to work on this new dataset. All implementations are publicly available and have been fine-tuned using off-the-shelf AutoML optimization tools for a fair comparison. Our best model achieves 98.5% accuracy. We believe that the DiffraNet dataset and its classification methods will have in the long term a positive impact in accelerating discoveries in many disciplines, including chemistry, geology, biology, materials science, metallurgy, and physics.""","""Reviewer ratings varied radically (from a 3 to an 8). However, the reviewer rating the paper as 8 provided extremely little justification for their rating. The reviewers providing lower ratings gave more detailed reviews, and also engaged in discussion with the authors. Ultimately neither decided to champion the paper, and therefore, I cannot recommend acceptance."""
1329,"""Explaining Image Classifiers by Counterfactual Generation""","['Explainability', 'Interpretability', 'Generative Models', 'Saliency Map', 'Machine Learning', 'Deep Learning']","""When an image classifier makes a prediction, which parts of the image are relevant and why? We can rephrase this question to ask: which parts of the image, if they were not seen by the classifier, would most change its decision? Producing an answer requires marginalizing over images that could have been seen but weren't. We can sample plausible image in-fills by conditioning a generative model on the rest of the image. We then optimize to find the image regions that most change the classifier's decision after in-fill. Our approach contrasts with ad-hoc in-filling approaches, such as blurring or injecting noise, which generate inputs far from the data distribution, and ignore informative relationships between different parts of the image. Our method produces more compact and relevant saliency maps, with fewer artifacts compared to previous methods.""","""Important problem (explainable AI); sensible approach, one of the first to propose a method for the counter-factual question (if this part of the input were different, what would the network have predicted). Initially there were some concerns by the reviewers but after the author response and reviewer discussion, all three recommend acceptance (not all of them updated their final scores in the system). """
1330,"""Learning Latent Superstructures in Variational Autoencoders for Deep Multidimensional Clustering""","['latent tree model', 'variational autoencoder', 'deep learning', 'latent variable model', 'bayesian network', 'structure learning', 'stepwise em', 'message passing', 'graphical model', 'multidimensional clustering', 'unsupervised learning']","""We investigate a variant of variational autoencoders where there is a superstructure of discrete latent variables on top of the latent features. In general, our superstructure is a tree structure of multiple super latent variables and it is automatically learned from data. When there is only one latent variable in the superstructure, our model reduces to one that assumes the latent features to be generated from a Gaussian mixture model. We call our model the latent tree variational autoencoder (LTVAE). Whereas previous deep learning methods for clustering produce only one partition of data, LTVAE produces multiple partitions of data, each being given by one super latent variable. This is desirable because high dimensional data usually have many different natural facets and can be meaningfully partitioned in multiple ways.""","""A well-written paper that proposes an original approach for leaning a structured prior for VAEs, as a latent tree model whose structure and parameters are simultaneously learned. It describes a well-principled approach to learning a multifaceted clustering, and is shown empirically to be competitive with other unsupervised clustering models. Reviewers noted that the approach reached a worse log-likelihood than regular VAE (which it should be able to find as a special case), hinting towards potential optimization difficulties (local minimum?). This would benefit form a more in-depth analysis. But reviewers appreciated the gain in interpretability and insights from the model, and unanimously agreed that the paper was an interesting novel contribution worth publishing. """
1331,"""Can I trust you more? Model-Agnostic Hierarchical Explanations""","['interpretability', 'interactions', 'context-dependent', 'context-free']","""Interactions such as double negation in sentences and scene interactions in images are common forms of complex dependencies captured by state-of-the-art machine learning models. We propose Mah, a novel approach to provide Model-Agnostic Hierarchical Explanations of how powerful machine learning models, such as deep neural networks, capture these interactions as either dependent on or free of the context of data instances. Specifically, Mah provides context-dependent explanations by a novel local interpretation algorithm that effectively captures any-order interactions, and obtains context-free explanations through generalizing context-dependent interactions to explain global behaviors. Experimental results show that Mah obtains improved local interaction interpretations over state-of-the-art methods and successfully provides explanations of interactions that are context-free.""","""This paper introduces Mahe, a model-agnostic hierarchical explanation technique, that constructs a hierarchy of explanations, from local, context-dependent ones (like LIME) to global, context-free ones. The reviewers found the proposed work to be a quite interesting application of the neural interaction detection (NID) framework, and overall found the results to be quite extensive and promising. The reviewers and the AC note the following as the primary concerns of the paper: (1) a crucial concern with the proposed work is the clarity of writing in the paper, and (2) the proposed work is quite expensive, computationally, as the exhaustive search is needed over local interactions. The reviewers appreciated the detailed comments and the revision, and felt the revised the manuscript was much improved by the additional editing, details in the papers, and the additional experiments. However, both reviewer 1 and 3 have strong reservations about the computational complexity of the approach, and the additional experiments did not alleviate it. Further, reviewer 1 is still concerned about the clarity of the work, finding much of the proposed work to be unclear, and recommends further revisions. Given these considerations, everyone felt that the idea is strong and most of the experiments are quite promising. However, without further editing and some efficiency strategies, it barely misses the bar of acceptance. """
1332,"""On the Relation Between the Sharpest Directions of DNN Loss and the SGD Step Length""","['optimization', 'generalization', 'theory of deep learning', 'SGD', 'hessian']","""The training of deep neural networks with Stochastic Gradient Descent (SGD) with a large learning rate or a small batch-size typically ends in flat regions of the weight space, as indicated by small eigenvalues of the Hessian of the training loss. This was found to correlate with a good final generalization performance. In this paper we extend previous work by investigating the curvature of the loss surface along the whole training trajectory, rather than only at the endpoint. We find that initially SGD visits increasingly sharp regions, reaching a maximum sharpness determined by both the learning rate and the batch-size of SGD. At this peak value SGD starts to fail to minimize the loss along directions in the loss surface corresponding to the largest curvature (sharpest directions). To further investigate the effect of these dynamics in the training process, we study a variant of SGD using a reduced learning rate along the sharpest directions which we show can improve training speed while finding both sharper and better generalizing solution, compared to vanilla SGD. Overall, our results show that the SGD dynamics in the subspace of the sharpest directions influence the regions that SGD steers to (where larger learning rate or smaller batch size result in wider regions visited), the overall training speed, and the generalization ability of the final model.""","""The reviewers found the paper insightful and the authors explanations well-provided. However the paper would benefit from more systematic empirical evaluation and corresponding theoretical intuition."""
1333,"""Deep Anomaly Detection with Outlier Exposure""","['confidence', 'uncertainty', 'anomaly', 'robustness']","""It is important to detect anomalous inputs when deploying machine learning systems. The use of larger and more complex inputs in deep learning magnifies the difficulty of distinguishing between anomalous and in-distribution examples. At the same time, diverse image and text data are available in enormous quantities. We propose leveraging these data to improve deep anomaly detection by training anomaly detectors against an auxiliary dataset of outliers, an approach we call Outlier Exposure (OE). This enables anomaly detectors to generalize and detect unseen anomalies. In extensive experiments on natural language processing and small- and large-scale vision tasks, we find that Outlier Exposure significantly improves detection performance. We also observe that cutting-edge generative models trained on CIFAR-10 may assign higher likelihoods to SVHN images than to CIFAR-10 images; we use OE to mitigate this issue. We also analyze the flexibility and robustness of Outlier Exposure, and identify characteristics of the auxiliary dataset that improve performance.""","""The paper proposes a new fine-tuning method for improving the performance of existing anomaly detectors. The reviewers and AC note the limitation of novelty beyond existing literature. This is quite a borader line paper, but AC decided to recommend acceptance as comprehensive experimental results (still based on empirical observation though) are interesting."""
1334,"""The Unusual Effectiveness of Averaging in GAN Training""","['Generative Adversarial Networks (GANs)', 'Moving Average', 'Exponential Moving Average', 'Convergence', 'Limit Cycles']","""We examine two different techniques for parameter averaging in GAN training. Moving Average (MA) computes the time-average of parameters, whereas Exponential Moving Average (EMA) computes an exponentially discounted sum. Whilst MA is known to lead to convergence in bilinear settings, we provide the -- to our knowledge -- first theoretical arguments in support of EMA. We show that EMA converges to limit cycles around the equilibrium with vanishing amplitude as the discount parameter approaches one for simple bilinear games and also enhances the stability of general GAN training. We establish experimentally that both techniques are strikingly effective in the non-convex-concave GAN setting as well. Both improve inception and FID scores on different architectures and for different GAN objectives. We provide comprehensive experimental results across a range of datasets -- mixture of Gaussians, CIFAR-10, STL-10, CelebA and ImageNet -- to demonstrate its effectiveness. We achieve state-of-the-art results on CIFAR-10 and produce clean CelebA face images.\footnote{~The code is available at \url{pseudo-url}}""","""This work analyses the use of parameter averaging in GANs. It can mainly be seen as an empirical study (while also a convergence analysis of EMA for a concrete example provides some minor theoretical result) but experimental results are very convincing and could promote using parameter averaging in the GAN community. Therefore, even if the technical novelty is limited, the insights brought by the paper are intesting. """
1335,"""Maximal Divergence Sequential Autoencoder for Binary Software Vulnerability Detection""","['Vulnerabilities Detection', 'Sequential Auto-Encoder', 'Separable Representation']","""Due to the sharp increase in the severity of the threat imposed by software vulnerabilities, the detection of vulnerabilities in binary code has become an important concern in the software industry, such as the embedded systems industry, and in the field of computer security. However, most of the work in binary code vulnerability detection has relied on handcrafted features which are manually chosen by a select few, knowledgeable domain experts. In this paper, we attempt to alleviate this severe binary vulnerability detection bottleneck by leveraging recent advances in deep learning representations and propose the Maximal Divergence Sequential Auto-Encoder. In particular, latent codes representing vulnerable and non-vulnerable binaries are encouraged to be maximally divergent, while still being able to maintain crucial information from the original binaries. We conducted extensive experiments to compare and contrast our proposed methods with the baselines, and the results show that our proposed methods outperform the baselines in all performance measures of interest.""",""" * Strengths This paper applies deep learning to the domain of cybersecurity, which is non-traditional relative to more common domains such as vision and speech. I see this as a strength. Additionally, the paper curates a dataset that may be of broader interest. * Weaknesses While the empirical results are good, there appears to be limited conceptual novelty. However, this is fine for a paper that is providing a new task in an interesting application domain. * Discussion Some reviewers were concerned about whether the dataset is a substantial contribution, as it is created based on existing publicly available data. However, these concerns were addressed by the author responses and all reviewers now agree with accepting the paper."""
1336,"""Small nonlinearities in activation functions create bad local minima in neural networks""","['spurious local minima', 'loss surface', 'optimization landscape', 'neural network']","""We investigate the loss surface of neural networks. We prove that even for one-hidden-layer networks with ""slightest"" nonlinearity, the empirical risks have spurious local minima in most cases. Our results thus indicate that in general ""no spurious local minim"" is a property limited to deep linear networks, and insights obtained from linear networks may not be robust. Specifically, for ReLU(-like) networks we constructively prove that for almost all practical datasets there exist infinitely many local minima. We also present a counterexample for more general activations (sigmoid, tanh, arctan, ReLU, etc.), for which there exists a bad local minimum. Our results make the least restrictive assumptions relative to existing results on spurious local optima in neural networks. We complete our discussion by presenting a comprehensive characterization of global optimality for deep linear networks, which unifies other results on this topic.""","""This is an interesting paper that develops new techniques for analyzing the loss surface of deep networks, allowing the existence of spurious local minima to be established under fairly general conditions. The reviewers responded with uniformly positive opinions."""
1337,"""HAPPIER: Hierarchical Polyphonic Music Generative RNN""","['hierarchical model', 'RNN', 'generative model', 'automatic composing']","""Generating polyphonic music with coherent global structure is a major challenge for automatic composition algorithms. The primary difficulty arises due to the inefficiency of models to recognize underlying patterns beneath music notes across different levels of time scales and remain long-term consistency while composing. Hierarchical architectures can capture and represent learned patterns in different temporal scales and maintain consistency over long time spans, and this corresponds to the hierarchical structure in music. Motivated by this, focusing on leveraging the idea of hierarchical models and improve them to fit the sequence modeling problem, our paper proposes HAPPIER: a novel HierArchical PolyPhonic musIc gEnerative RNN. In HAPPIER, A higher `measure level' learns correlations across measures and patterns for chord progressions, and a lower `note level' learns a conditional distribution over the notes to generate within a measure. The two hierarchies operate at different clock rates: the higher one operates on a longer timescale and updates every measure, while the lower one operates on a shorter timescale and updates every unit duration. The two levels communicate with each other, and thus the entire architecture is trained jointly end-to-end by back-propagation. HAPPIER, profited from the strength of the hierarchical structure, generates polyphonic music with long-term dependencies compared to the state-of-the-art methods.""","""Although all the reviewers find the problem and the approach of using hierarchical models important and interesting, how it has been executed in this submission has not been found favourable by the reviewers."""
1338,"""A Rate-Distortion Theory of Adversarial Examples""","['adversarial examples', 'information bottleneck', 'robustness']","""The generalization ability of deep neural networks (DNNs) is intertwined with model complexity, robustness, and capacity. Through establishing an equivalence between a DNN and a noisy communication channel, we characterize generalization and fault tolerance for unbounded adversarial attacks in terms of information-theoretic quantities. Invoking rate-distortion theory, we suggest that excess capacity is a significant cause of vulnerability to adversarial examples.""","""Both authors and reviewers agree that the ideas in the paper were not presented clearly enough."""
1339,"""Interpolation-Prediction Networks for Irregularly Sampled Time Series""","['irregular sampling', 'multivariate time series', 'supervised learning', 'interpolation', 'missing data']","""In this paper, we present a new deep learning architecture for addressing the problem of supervised learning with sparse and irregularly sampled multivariate time series. The architecture is based on the use of a semi-parametric interpolation network followed by the application of a prediction network. The interpolation network allows for information to be shared across multiple dimensions of a multivariate time series during the interpolation stage, while any standard deep learning model can be used for the prediction network. This work is motivated by the analysis of physiological time series data in electronic health records, which are sparse, irregularly sampled, and multivariate. We investigate the performance of this architecture on both classification and regression tasks, showing that our approach outperforms a range of baseline and recently proposed models. ""","""After much discussion, all reviewers agree that this paper should be accepted. Congratulations!!"""
1340,"""ReNeg and Backseat Driver: Learning from demonstration with continuous human feedback""","['learning from demonstration', 'imitation learning', 'behavioral cloning', 'reinforcement learning', 'off-policy', 'continuous control', 'autonomous vehicles', 'deep learning', 'machine learning', 'policy gradient']","""Reinforcement learning (RL) is a powerful framework for solving problems by exploring and learning from mistakes. However, in the context of autonomous vehicle (AV) control, requiring an agent to make mistakes, or even allowing mistakes, can be quite dangerous and costly in the real world. For this reason, AV RL is generally only viable in simulation. Because these simulations have imperfect representations, particularly with respect to graphics, physics, and human interaction, we find motivation for a framework similar to RL, suitable to the real world. To this end, we formulate a learning framework that learns from restricted exploration by having a human demonstrator do the exploration. Existing work on learning from demonstration typically either assumes the collected data is performed by an optimal expert, or requires potentially dangerous exploration to find the optimal policy. We propose an alternative framework that learns continuous control from only safe behavior. One of our key insights is that the problem becomes tractable if the feedback score that rates the demonstration applies to the atomic action, as opposed to the entire sequence of actions. We use human experts to collect driving data as well as to label the driving data through a framework we call ``Backseat Driver'', giving us state-action pairs matched with scalar values representing the score for the action. We call the more general learning framework ReNeg, since it learns a regression from states to actions given negative as well as positive examples. We empirically validate several models in the ReNeg framework, testing on lane-following with limited data. We find that the best solution in this context outperforms behavioral cloning has strong connections to stochastic policy gradient approaches.""","""The authors consider the interesting and important problem of how to train a robust driving policy without allowing unsafe exploration, an important challenge for real-world training scenarios. They suggest that both good and intentionally bad human demonstrations could be used, with the intuition being that humans can readily produce unsafe exploration such as swerving which can then be learnt using both positive and negative regressions. The reviewers all agree that the paper would not appeal to or have relevance for the wider community. The reviewers also agree that the main ideas are not well presented, that some of the claims are confusing, and that the writing is not technical enough. They also question the thoroughness of the empirical validation. """
1341,"""Using Word Embeddings to Explore the Learned Representations of Convolutional Neural Networks""","['Distributional Semantics', 'word embeddings', 'cnns', 'interpretability']","""As deep neural net architectures minimize loss, they build up information in a hierarchy of learned representations that ultimately serve their final goal. Different architectures tackle this problem in slightly different ways, but all models aim to create representational spaces that accumulate information through the depth of the network. Here we build on previous work that indicated that two very different model classes trained on two very different tasks actually build knowledge representations that have similar underlying representations. Namely, we compare word embeddings from SkipGram (trained to predict co-occurring words) to several CNN architectures (trained for image classification) in order to understand how this accumulation of knowledge behaves in CNNs. We improve upon previous work by including 5 times more ImageNet classes in our experiments, and further expand the scope of the analyses to include a network trained on CIFAR-100. We characterize network behavior in pretrained models, and also during training, misclassification, and adversarial attack. Our work illustrates the power of using one model to explore another, gives new insights for CNN models, and provides a framework for others to perform similar analyses when developing new architectures.""","""The paper aims to study what is learned in the word representations by comparing SkipGram embeddings trained from a text corpus and CNNs trained from ImageNet. Pros: The paper tries to be comprehensive, including analysis of text representations and image representations, and the cases of misclassification and adversarial examples. Cons: The clarity of the paper is a major concern, as noted by all reviwers, and the authors did not come back with rebuttal to address reviewers' quetions. Also, as R1 and R2 pointed out the novelty over recent relevant papers such as (Dharmaretnam & Fyshe, 2018) is not clear. Verdict: Reject due to weak novelty and major clarity issues."""
1342,"""PIE: Pseudo-Invertible Encoder""","['Invertible Mappings', 'Bijectives', 'Dimensionality reduction', 'Autoencoder']","""We consider the problem of information compression from high dimensional data. Where many studies consider the problem of compression by non-invertible trans- formations, we emphasize the importance of invertible compression. We introduce new class of likelihood-based auto encoders with pseudo bijective architecture, which we call Pseudo Invertible Encoders. We provide the theoretical explanation of their principles. We evaluate Gaussian Pseudo Invertible Encoder on MNIST, where our model outperform WAE and VAE in sharpness of the generated images.""","""The presented approach demonstrates an invertible architecture for auto-encoding, which demonstrates improvements in performance relative to VAE and WAE's on MNIST. Pros: + R3: The idea of pseudo-inversion is interesting. + R3: Manuscript is clear. Cons: - R1,2,3: Additional experiments needed on CIFAR, ImageNet, others. - R1: Presentation unclear. Authors have not made any apparent attempt to improve the clarity of the manuscript, though they make their point that the method allows dimensionality reduction in their response. - R1, R2: Main advantages not clear. - R3: Text could be compressed further to allow room for additional experiments. Reviewers lean reject, and authors have not updated experiments. Authors are encouraged to continue to improve the work."""
1343,"""Harmonizing Maximum Likelihood with GANs for Multimodal Conditional Generation""","['conditional GANs', 'conditional image generation', 'multimodal generation', 'reconstruction loss', 'maximum likelihood estimation', 'moment matching']","""Recent advances in conditional image generation tasks, such as image-to-image translation and image inpainting, are largely accounted to the success of conditional GAN models, which are often optimized by the joint use of the GAN loss with the reconstruction loss. However, we reveal that this training recipe shared by almost all existing methods causes one critical side effect: lack of diversity in output samples. In order to accomplish both training stability and multimodal output generation, we propose novel training schemes with a new set of losses named moment reconstruction losses that simply replace the reconstruction loss. We show that our approach is applicable to any conditional generation tasks by performing thorough experiments on image-to-image translation, super-resolution and image inpainting using Cityscapes and CelebA dataset. Quantitative evaluations also confirm that our methods achieve a great diversity in outputs while retaining or even improving the visual fidelity of generated samples.""","""The paper presents new loss functions (which replace the reconstruction part) for the training of conditional GANs. Theoretical considerations and an empirical analysis show that the proposed loss can better handle multimodality of the target distribution than reconstruction based losses while being competitive in terms of image quality."""
1344,"""Deep learning generalizes because the parameter-function map is biased towards simple functions""","['generalization', 'deep learning theory', 'PAC-Bayes', 'Gaussian processes', 'parameter-function map', 'simplicity bias']","""Deep neural networks (DNNs) generalize remarkably well without explicit regularization even in the strongly over-parametrized regime where classical learning theory would instead predict that they would severely overfit. While many proposals for some kind of implicit regularization have been made to rationalise this success, there is no consensus for the fundamental reason why DNNs do not strongly overfit. In this paper, we provide a new explanation. By applying a very general probability-complexity bound recently derived from algorithmic information theory (AIT), we argue that the parameter-function map of many DNNs should be exponentially biased towards simple functions. We then provide clear evidence for this strong simplicity bias in a model DNN for Boolean functions, as well as in much larger fully connected and convolutional networks trained on CIFAR10 and MNIST. As the target functions in many real problems are expected to be highly structured, this intrinsic simplicity bias helps explain why deep networks generalize well on real world problems. This picture also facilitates a novel PAC-Bayes approach where the prior is taken over the DNN input-output function space, rather than the more conventional prior over parameter space. If we assume that the training algorithm samples parameters close to uniformly within the zero-error region then the PAC-Bayes theorem can be used to guarantee good expected generalization for target functions producing high-likelihood training sets. By exploiting recently discovered connections between DNNs and Gaussian processes to estimate the marginal likelihood, we produce relatively tight generalization PAC-Bayes error bounds which correlate well with the true error on realistic datasets such as MNIST and CIFAR10 and for architectures including convolutional and fully connected networks.""","""Dear authors, There was some disagreement among reviewers on the significance of your results, in particular because of the limited experimental section. Despite this issues, which is not minor, your work adds yet another piece of the generalization puzzle. However, I would encourage the authors to make sure they do not oversell their results, either in the title or in their text, for the final version."""
1345,"""Weak contraction mapping and optimization""","['Weak contraction mapping', 'fixed-point theorem', 'non-convex optimization']","""The weak contraction mapping is a self mapping that the range is always a subset of the domain, which admits a unique fixed-point. The iteration of weak contraction mapping is a Cauchy sequence that yields the unique fixed-point. A gradient-free optimization method as an application of weak contraction mapping is proposed to achieve global minimum convergence. The optimization method is robust to local minima and initial point position.""","""This paper proposes an optimization algorithm based on 'weak contraction mapping'. The paper is written poorly without clear definitions and mathematical rigor. Reviewers doubt both the correctness and the usefulness of the proposed method. I strongly suggest authors to rewrite the paper addressing all the reviews before submitting to a different venue. """
1346,"""Fluctuation-dissipation relations for stochastic gradient descent""","['stochastic gradient descent', 'adaptive method', 'loss surface', 'Hessian']","""The notion of the stationary equilibrium ensemble has played a central role in statistical mechanics. In machine learning as well, training serves as generalized equilibration that drives the probability distribution of model parameters toward stationarity. Here, we derive stationary fluctuation-dissipation relations that link measurable quantities and hyperparameters in the stochastic gradient descent algorithm. These relations hold exactly for any stationary state and can in particular be used to adaptively set training schedule. We can further use the relations to efficiently extract information pertaining to a loss-function landscape such as the magnitudes of its Hessian and anharmonicity. Our claims are empirically verified.""","""The paper presents interesting idea, but the reviewers ask for improving further paper clarity - that includes, but is not limited to, providing in-depth explanation of assumptions and also improving the writing that is too heavy and difficult to understand."""
1347,"""SynonymNet: Multi-context Bilateral Matching for Entity Synonyms""","['deep learning', 'entity synonym']","""Being able to automatically discover synonymous entities from a large free-text corpus has transformative effects on structured knowledge discovery. Existing works either require structured annotations, or fail to incorporate context information effectively, which lower the efficiency of information usage. In this paper, we propose a framework for synonym discovery from free-text corpus without structured annotation. As one of the key components in synonym discovery, we introduce a novel neural network model SynonymNet to determine whether or not two given entities are synonym with each other. Instead of using entities features, SynonymNet makes use of multiple pieces of contexts in which the entity is mentioned, and compares the context-level similarity via a bilateral matching schema to determine synonymity. Experimental results demonstrate that the proposed model achieves state-of-the-art results on both generic and domain-specific synonym datasets: Wiki+Freebase, PubMed+UMLS and MedBook+MKG, with up to 4.16% improvement in terms of Area Under the Curve (AUC) and 3.19% in terms of Mean Average Precision (MAP) compare to the best baseline method.""","""This paper presents a model to identify entity mentions that are synonymous. This could have utility in practical scenarios that handle entities. The main criticism of the paper is regarding the baselines used. Most of the baselines that are compared against are extremely simple. There is a significant body of literature that models paraphrase and entailment and many of those baselines are missing (decomposable attention, DIIN, other cross-attention mechanisms). Adding those experiments would make the experimental setup stronger. There is a bit of a disagreement between reviewers, but I agree with the two reviewers who point out the weakness of the experimental setup, and fixing those issues could improve the paper significantly."""
1348,"""Stable Opponent Shaping in Differentiable Games""","['multi-agent learning', 'multiple interacting losses', 'opponent shaping', 'exploitation', 'convergence']","""A growing number of learning methods are actually differentiable games whose players optimise multiple, interdependent objectives in parallel from GANs and intrinsic curiosity to multi-agent RL. Opponent shaping is a powerful approach to improve learning dynamics in these games, accounting for player influence on others updates. Learning with Opponent-Learning Awareness (LOLA) is a recent algorithm that exploits this response and leads to cooperation in settings like the Iterated Prisoners Dilemma. Although experimentally successful, we show that LOLA agents can exhibit arrogant behaviour directly at odds with convergence. In fact, remarkably few algorithms have theoretical guarantees applying across all (n-player, non-convex) games. In this paper we present Stable Opponent Shaping (SOS), a new method that interpolates between LOLA and a stable variant named LookAhead. We prove that LookAhead converges locally to equilibria and avoids strict saddles in all differentiable games. SOS inherits these essential guarantees, while also shaping the learning of opponents and consistently either matching or outperforming LOLA experimentally.""","""This paper provides interesting results on convergence and stability in general differentiable games. The theory appears to be correct, and the paper reasonably well written. The main concern is in connections to an area of related work that has been omitted, with overly strong statements in the paper that there has been little work for general game dynamics. This is a serious omission, since it calls into question some of the novelty of the results because they have not been adequately placed relative to this work. The authors should incorporate a thorough discussion on relations to this work, and adjust claims about novelty (and potentially even results) based on that literature."""
1349,"""Solving the Rubik's Cube with Approximate Policy Iteration""","['reinforcement learning', ""Rubik's Cube"", 'approximate policy iteration', 'deep learning', 'deep reinforcement learning']","""Recently, Approximate Policy Iteration (API) algorithms have achieved super-human proficiency in two-player zero-sum games such as Go, Chess, and Shogi without human data. These API algorithms iterate between two policies: a slow policy (tree search), and a fast policy (a neural network). In these two-player games, a reward is always received at the end of the game. However, the Rubiks Cube has only a single solved state, and episodes are not guaranteed to terminate. This poses a major problem for these API algorithms since they rely on the reward received at the end of the game. We introduce Autodidactic Iteration: an API algorithm that overcomes the problem of sparse rewards by training on a distribution of states that allows the reward to propagate from the goal state to states farther away. Autodidactic Iteration is able to learn how to solve the Rubiks Cube and the 15-puzzle without relying on human data. Our algorithm is able to solve 100% of randomly scrambled cubes while achieving a median solve length of 30 moves less than or equal to solvers that employ human domain knowledge.""","""The paper introduces a version of approximate policy iteration (API), called Autodidactic Iteration (ADI), designed to overcome the problem of sparse rewards. In particular, the policy evaluation step of ADI is trained on a distribution of states that allows the reward to easily propagate from the goal state to states farther away. ADI is applied to successfully solve the Rubik's Cube (together with other existing techniques). This work is an interesting contribution where the ADI idea may be useful in other scenarios. A limitation is that the whole empirical study is on the Rubik's Cube; a controlled experiment on other problems (even if simpler) can be useful to understand the pros & cons of ADI compared to others. Minor: please update the bib entry of Bottou (2011). It's now published in MLJ 2014."""
1350,"""Stackelberg GAN: Towards Provable Minimax Equilibrium via Multi-Generator Architectures""","['generative adversarial nets', 'minimax duality gap', 'equilibrium']","""We study the problem of alleviating the instability issue in the GAN training procedure via new architecture design. The discrepancy between the minimax and maximin objective values could serve as a proxy for the difficulties that the alternating gradient descent encounters in the optimization of GANs. In this work, we give new results on the benefits of multi-generator architecture of GANs. We show that the minimax gap shrinks to \epsilon as the number of generators increases with rate O(1/\epsilon). This improves over the best-known result of O(1/\epsilon^2). At the core of our techniques is a novel application of Shapley-Folkman lemma to the generic minimax problem, where in the literature the technique was only known to work when the objective function is restricted to the Lagrangian function of a constraint optimization problem. Our proposed Stackelberg GAN performs well experimentally in both synthetic and real-world datasets, improving Frechet Inception Distance by 14.61% over the previous multi-generator GANs on the benchmark datasets.""","""This paper proposes new GAN training method with multi generator architecture inspired by Stackelberg competition in game theory. The paper has theoretical results showing that minmax gap scales to \eps for number of generators O(1/\eps), improving over previous bounds. Paper also has some experimental results on Fashion Mnist and CIFAR10 datasets. Reviewers find the theoretical results of the paper interesting. However, reviewers have multiple concerns about comparison with other multi generator architectures, optimization dynamics of the new objective and clarity of writing of the original submission. While authors have addressed some of these concerns in their response reviewers still remain skeptical of the contributions. Perhaps more experiments on imagenet quality datasets with detailed comparison can help make the contributions of the paper clearer. """
1351,"""Noisy Information Bottlenecks for Generalization""","['information theory', 'deep learning', 'generalization', 'information bottleneck', 'variational inference', 'approximate inference']","""We propose Noisy Information Bottlenecks (NIB) to limit mutual information between learned parameters and the data through noise. We show why this benefits generalization and allows mitigation of model overfitting both for supervised and unsupervised learning, even for arbitrarily complex architectures. We reinterpret methods including the Variational Autoencoder, beta-VAE, network weight uncertainty and a variant of dropout combined with weight decay as special cases of our approach, explaining and quantifying regularizing properties and vulnerabilities within information theory.""","""The paper proposes a regularization method that introduces an information bottleneck between parameters and predictions. The reviewers agree that the paper proposes some interesting ideas, but those idea need to be clarified. The paper lacks in clarity. The reviewers also doubt whether the paper is expected to have significant impact in the field."""
1352,"""Generative Adversarial Network Training is a Continual Learning Problem""","['Generative Adversarial Networks', 'Continual Learning', 'Deep Learning']","""Generative Adversarial Networks (GANs) have proven to be a powerful framework for learning to draw samples from complex distributions. However, GANs are also notoriously difficult to train, with mode collapse and oscillations a common problem. We hypothesize that this is at least in part due to the evolution of the generator distribution and the catastrophic forgetting tendency of neural networks, which leads to the discriminator losing the ability to remember synthesized samples from previous instantiations of the generator. Recognizing this, our contributions are twofold. First, we show that GAN training makes for a more interesting and realistic benchmark for continual learning methods evaluation than some of the more canonical datasets. Second, we propose leveraging continual learning techniques to augment the discriminator, preserving its ability to recognize previous generator samples. We show that the resulting methods add only a light amount of computation, involve minimal changes to the model, and result in better overall performance on the examined image and text generation tasks.""","""This paper studies training of the generative adversarial networks (GANs), specifically the discriminator, as a continual learning problem, where the discriminator does not forget previously generated samples. This model can be potentially used for improving GANs training and for generating synthetic datasets for evaluating continual learning methods. All the reviewers and AC agree that showing how continual learning techniques applied to the discriminator can alleviate mode collapse in GANs training is an important direction to study. There is reviewer disagreement on this paper. AC can confirm that all three reviewers have read the author responses and have contributed to the final discussion. While acknowledging that continual learning setting is potentially useful, the reviewers have raised several important concerns: (1) low technical novelty in light of EWC++ and online EWC methods (R1 and R3) -- methodological and empirical comparison to these baselines is required to assess the difference and benefits of the proposed approach; the authors response to these concerns (and also R2s comments in the discussion) were insufficient to assess the scope of the contribution. (2) More diverse/convincing empirical findings would strengthen the evaluation (e.g. assessing whether or not generator could help to overcome forgetting; showing that memory replay strategy by storing sufficient fake examples from previously generated samples cannot prevent mode collapse in GANs training see the R3s comment; showing the benefits of the generated samples for evaluating continual learning methods). (3) R1 left unconvinced that GAN training can be improved via continual learning training, as the relation between the proposed view and the minimax optimization difficulties in GANs is not addressed see R1s comment about this. The authors briefly discussed in their response to the review that the proposed approach is orthogonal to these works. However, a better (possibly theoretical) analysis of GANs training and continual learning would indeed help to evaluate the scope of the contribution of this work. Regarding the available datasets that exhibit a coherent time evolution -- see the Continuous Manifold Based Adaptation for Evolving Visual Domains by Hoffman et al, CVPR 2014. Among (1)-(3): (2) and (3) did not have a substantial impact on the decision, but would be helpful to address in a subsequent revision. However, (1) makes it very difficult to assess the benefits of the proposed approach, and was viewed by AC as a critical issue. AC suggests that in this current state the paper can be considered for a workshop and recommend to prepare a major revision before resubmitting it for the second round of reviews. """
1353,"""EnGAN: Latent Space MCMC and Maximum Entropy Generators for Energy-based Models""","['Energy based model', 'Generative models', 'MCMC', 'GANs']","""Unsupervised learning is about capturing dependencies between variables and is driven by the contrast between the probable vs improbable configurations of these variables, often either via a generative model which only samples probable ones or with an energy function (unnormalized log-density) which is low for probable ones and high for improbable ones. Here we consider learning both an energy function and an efficient approximate sampling mechanism for the corresponding distribution. Whereas the critic (or discriminator) in generative adversarial networks (GANs) learns to separate data and generator samples, introducing an entropy maximization regularizer on the generator can turn the interpretation of the critic into an energy function, which separates the training distribution from everything else, and thus can be used for tasks like anomaly or novelty detection. This paper is motivated by the older idea of sampling in latent space rather than data space because running a Monte-Carlo Markov Chain (MCMC) in latent space has been found to be easier and more efficient, and because a GAN-like generator can convert latent space samples to data space samples. For this purpose, we show how a Markov chain can be run in latent space whose samples can be mapped to data space, producing better samples. These samples are also used for the negative phase gradient required to estimate the log-likelihood gradient of the data space energy function. To maximize entropy at the output of the generator, we take advantage of recently introduced neural estimators of mutual information. We find that in addition to producing a useful scoring function for anomaly detection, the resulting approach produces sharp samples (like GANs) while covering the modes well, leading to high Inception and Frchet scores. ""","""The proposed method is an extension of Kim & Bengio (2016)'s energy-based GAN. The novel contributions are to approximate the entropy regularizer using a mutual information estimator, and to try to clean up the model samples using some Langevin steps. Experiments include mode dropping experiments on toy data, samples from the model on CelebA, and measures of inception score and FID. The paper is well-written, and the proposal seems sensible. But as various reviewers point out, the work is a fairly incremental extension of Kim and Bengio (2016). Most of the new elements, such as Langevin sampling and the gradient penalty, have also been well-explored in the deep generative modeling literature. It's not clear there is a particular contribution here that really stands out. The experimental evidence for improvement is also fairly limited. Generated samples, inception scores, and FID are pretty weak measures for generative models, though I'm willing to go with them since they seem to be standard in the field. But even by these measures, there doesn't seem to be much improvement. I wouldn't expect SOTA results because of computational limitations, but the generated samples and quantitative evaluations seem worse than the WGAN-GP, even though the proposed method includes the gradient penalty and hence should be able to at least match WGAN-GP. The MCMC sampling doesn't appear to have helped, as far as I can tell. Overall, the proposal seems promising, but I don't think this paper is ready for publication at ICLR. """
1354,"""Deep Frank-Wolfe For Neural Network Optimization""","['optimization', 'conditional gradient', 'Frank-Wolfe', 'SVM']","""Learning a deep neural network requires solving a challenging optimization problem: it is a high-dimensional, non-convex and non-smooth minimization problem with a large number of terms. The current practice in neural network optimization is to rely on the stochastic gradient descent (SGD) algorithm or its adaptive variants. However, SGD requires a hand-designed schedule for the learning rate. In addition, its adaptive variants tend to produce solutions that generalize less well on unseen data than SGD with a hand-designed schedule. We present an optimization method that offers empirically the best of both worlds: our algorithm yields good generalization performance while requiring only one hyper-parameter. Our approach is based on a composite proximal framework, which exploits the compositional nature of deep neural networks and can leverage powerful convex optimization algorithms by design. Specifically, we employ the Frank-Wolfe (FW) algorithm for SVM, which computes an optimal step-size in closed-form at each time-step. We further show that the descent direction is given by a simple backward pass in the network, yielding the same computational cost per iteration as SGD. We present experiments on the CIFAR and SNLI data sets, where we demonstrate the significant superiority of our method over Adam, Adagrad, as well as the recently proposed BPGrad and AMSGrad. Furthermore, we compare our algorithm to SGD with a hand-designed learning rate schedule, and show that it provides similar generalization while often converging faster. The code is publicly available at pseudo-url.""","""The paper was judged by the reviewers as providing interesting ideas, well-written and potentially having impact on future research on NN optimization. The authors are asked to make sure they addressed reviewers comments clearly in the paper."""
1355,"""Probabilistic Knowledge Graph Embeddings""","['knowledge graph', 'variational inference', 'probabilistic models', 'representation learning']","""We develop a probabilistic extension of state-of-the-art embedding models for link prediction in relational knowledge graphs. Knowledge graphs are collections of relational facts, where each fact states that a certain relation holds between two entities, such as people, places, or objects. We argue that knowledge graphs should be treated within a Bayesian framework because even large knowledge graphs typically contain only few facts per entity, leading effectively to a small data problem where parameter uncertainty matters. We introduce a probabilistic reinterpretation of the DistMult (Yang et al., 2015) and ComplEx (Trouillon et al., 2016) models and employ variational inference to estimate a lower bound on the marginal likelihood of the data. We find that the main benefit of the Bayesian approach is that it allows for efficient, gradient based optimization over hyperparameters, which would lead to divergences in a non-Bayesian treatment. Models with such learned hyperparameters improve over the state-of-the-art by a significant margin, as we demonstrate on several benchmarks.""","""The paper proposes a Bayesian extension to existing knowledge base embedding methods (like DistMult and ComplEx), which is applied for for hyperparameter learning. While using Bayesian inference for for hyperparameter tuning for embedding methods is not generally novel, it has not been used in the context of knowledge graph modelling before. The paper could be strengthened by comparing the method to other strategies of hyperparameter selection to prove the significance of the advantage brought by the method."""
1356,"""Improving Sentence Representations with Multi-view Frameworks""","['multi-view', 'learning', 'sentence', 'representation']","""Multi-view learning can provide self-supervision when different views are available of the same data. Distributional hypothesis provides another form of useful self-supervision from adjacent sentences which are plentiful in large unlabelled corpora. Motivated by the asymmetry in the two hemispheres of the human brain as well as the observation that different learning architectures tend to emphasise different aspects of sentence meaning, we present two multi-view frameworks for learning sentence representations in an unsupervised fashion. One framework uses a generative objective and the other a discriminative one. In both frameworks, the final representation is an ensemble of two views, in which, one view encodes the input sentence with a Recurrent Neural Network (RNN), and the other view encodes it with a simple linear model. We show that, after learning, the vectors produced by our multi-view frameworks provide improved representations over their single-view learnt counterparts, and the combination of different views gives representational improvement over each view and demonstrates solid transferability on standard downstream tasks.""","""This paper offers a new method for sentence representation learning, fitting loosely into the multi-view learning framework, with fairly strong results. The paper is clearly borderline, with one reviewer arguing for acceptance and another arguing for rejection. While it is a tough decision, I have to argue for rejection in this case. There was a robust discussion and the authors revised the paper, so none of the remaining technical issues strike me as fatal. My primary concern is simply that the reviewers could not reach a consensus in favor of the paper. In particular, two reviewers expressed concerns that this paper makes too small an advance in NLP to be of interest to non-NLP researchers. I think it should be possible to broaden the scope of the paper and resubmit it to another general ML venue, and (as one reviewer suggested explicitly), this paper may have a better chance at an NLP-specific venue. While neither of these factors was crucial in the decision, I'd encourage the authors (i) to put more effort into comparing properly with the Subramanian and Radford baselines, and (ii) to clarify the points about the human brain. For the second point: While none of the claims about the brain are false *or misleading*, as far as I know, the authors do not make a convincing case that the claims about the brain are actually relevant to the work being done here."""
1357,"""High Resolution and Fast Face Completion via Progressively Attentive GANs""","['Face Completion', 'progressive GANs', 'Attribute Control', 'Frequency-oriented Attention']","""Face completion is a challenging task with the difficulty level increasing significantly with respect to high resolution, the complexity of ""holes"" and the controllable attributes of filled-in fragments. Our system addresses the challenges by learning a fully end-to-end framework that trains generative adversarial networks (GANs) progressively from low resolution to high resolution with conditional vectors encoding controllable attributes. We design a novel coarse-to-fine attentive module network architecture. Our model is encouraged to attend on finer details while the network is growing to a higher resolution, thus being capable of showing progressive attention to different frequency components in a coarse-to-fine way. We term the module Frequency-oriented Attentive Module (FAM). Our system can complete faces with large structural and appearance variations using a single feed-forward pass of computation with mean inference time of 0.54 seconds for images at 1024x1024 resolution. A pilot human study shows our approach outperforms state-of-the-art face completion methods. The code will be released upon publication. ""","""All reviewers gave a 5 rating. The author rebuttal was not able to alter the consensus view of reviewers. See below for details."""
1358,"""signSGD via Zeroth-Order Oracle""","['nonconvex optimization', 'zeroth-order algorithm', 'black-box adversarial attack']","""In this paper, we design and analyze a new zeroth-order (ZO) stochastic optimization algorithm, ZO-signSGD, which enjoys dual advantages of gradient-free operations and signSGD. The latter requires only the sign information of gradient estimates but is able to achieve a comparable or even better convergence speed than SGD-type algorithms. Our study shows that ZO signSGD requires pseudo-formula times more iterations than signSGD, leading to a convergence rate of pseudo-formula under mild conditions, where pseudo-formula is the number of optimization variables, and pseudo-formula is the number of iterations. In addition, we analyze the effects of different types of gradient estimators on the convergence of ZO-signSGD, and propose two variants of ZO-signSGD that at least achieve pseudo-formula convergence rate. On the application side we explore the connection between ZO-signSGD and black-box adversarial attacks in robust deep learning. Our empirical evaluations on image classification datasets MNIST and CIFAR-10 demonstrate the superior performance of ZO-signSGD on the generation of adversarial examples from black-box neural networks.""","""This is a solid paper that proposes and analyzes a sound approach to zero order optimization, covering a variants of a simple base algorithm. After resolving some issues during the response period, the reviewers concluded with a unanimous recommendation of acceptance. Some concerns regarding the necessity for such algorithms persisted, but the connection to adversarial examples provides an interesting motivation."""
1359,"""Towards Consistent Performance on Atari using Expert Demonstrations""","['Reinforcement Learning', 'Atari', 'RL', 'Demonstrations']","""Despite significant advances in the field of deep Reinforcement Learning (RL), today's algorithms still fail to learn human-level policies consistently over a set of diverse tasks such as Atari 2600 games. We identify three key challenges that any algorithm needs to master in order to perform well on all games: processing diverse reward distributions, reasoning over long time horizons, and exploring efficiently. In this paper, we propose an algorithm that addresses each of these challenges and is able to learn human-level policies on nearly all Atari games. A new transformed Bellman operator allows our algorithm to process rewards of varying densities and scales; an auxiliary temporal consistency loss allows us to train stably using a discount factor of 0.999 (instead of 0.99) extending the effective planning horizon by an order of magnitude; and we ease the exploration problem by using human demonstrations that guide the agent towards rewarding states. When tested on a set of 42 Atari games, our algorithm exceeds the performance of an average human on 40 games using a common set of hyper parameters.""","""This paper proposes a combination of three techniques to improve the learning performance of Atari games. Good performance was shown in the paper with all three techniques together applied to DQN. However, it is hard to justify the integration of these techniques. It is also not clear why the specific decisions were made when combining them. More comprehensive experiments, such as a more systematic ablation study, are required to convince the benefits of individual components. Furthermore, it seems very hard to tell whether the improvement of existing approaches, such as Ape-X DQN, was from using the proposed techniques or a deeper architecture (Tables 1&2&4&5). Overall, this paper is not ready for publication. """
1360,"""Stacking for Transfer Learning""","['data diversification', 'domain adaptation', 'transfer learning', 'stacked generalization']","""In machine learning tasks, overtting frequently crops up when the number of samples of target domain is insufcient, for the generalization ability of the classier is poor in this circumstance. To solve this problem, transfer learning utilizes the knowledge of similar domains to improve the robustness of the learner. The main idea of existing transfer learning algorithms is to reduce the dierence between domains by sample selection or domain adaptation. However, no matter what transfer learning algorithm we use, the difference always exists and the hybrid training of source and target data leads to reducing tting capability of the learner on target domain. Moreover, when the relatedness between domains is too low, negative transfer is more likely to occur. To tackle the problem, we proposed a two-phase transfer learning architecture based on ensemble learning, which uses the existing transfer learning algorithms to train the weak learners in the rst stage, and uses the predictions of target data to train the nal learner in the second stage. Under this architecture, the tting capability and generalization capability can be guaranteed at the same time. We evaluated the proposed method on public datasets, which demonstrates the effectiveness and robustness of our proposed method.""","""This work proposes a method for both instance and feature based transfer learning. The reviewers agree that the approach in current form lacks sufficient technical novelty for publication. The paper would benefit from experiments on larger datasets and with more analysis into the different aspects of the proposed model. """
1361,"""Equi-normalization of Neural Networks""","['convolutional neural networks', 'Normalization', 'Sinkhorn', 'Regularization']","""Modern neural networks are over-parametrized. In particular, each rectified linear hidden unit can be modified by a multiplicative factor by adjusting input and out- put weights, without changing the rest of the network. Inspired by the Sinkhorn-Knopp algorithm, we introduce a fast iterative method for minimizing the l2 norm of the weights, equivalently the weight decay regularizer. It provably converges to a unique solution. Interleaving our algorithm with SGD during training improves the test accuracy. For small batches, our approach offers an alternative to batch- and group- normalization on CIFAR-10 and ImageNet with a ResNet-18.""","""The proposed ENorm procedure is a normalization scheme for neural nets whereby the weights are rescaled in a way that minimizes the sum of L_p norms while maintaining functional equivalence. An algorithm is given which provably converges to the globally optimal solution. Experiments show it is complementary to, and perhaps slightly better than, other normalization schemes. Normalization issues are important for DNN training, and normalization schemes like batch norm, weight norm, etc. have the unsatisfying property that they entangle multiple issues such as normalization, stochastic regularization, and effective learning rates. ENorm is a conceptually cleaner (if more algorithmically complicated) approach. It's a nice addition to the set of normalization schemes, and possibly complementary to the existing ones. After a revision which included various new experiments, the reviewers are generally happy with the paper. While there's still some controversy over whether it's really better than things like batch norm, I think the paper would be worth publishing even if the results came out negative, since it is a very natural idea which took some algorithmic insight in order to actually execute. """
1362,"""Learning what you can do before doing anything""","['unsupervised learning', 'vision', 'motion', 'action space', 'video prediction', 'variational models']","""Intelligent agents can learn to represent the action spaces of other agents simply by observing them act. Such representations help agents quickly learn to predict the effects of their own actions on the environment and to plan complex action sequences. In this work, we address the problem of learning an agents action space purely from visual observation. We use stochastic video prediction to learn a latent variable that captures the scene's dynamics while being minimally sensitive to the scene's static content. We introduce a loss term that encourages the network to capture the composability of visual sequences and show that it leads to representations that disentangle the structure of actions. We call the full model with composable action representations Composable Learned Action Space Predictor (CLASP). We show the applicability of our method to synthetic settings and its potential to capture action spaces in complex, realistic visual settings. When used in a semi-supervised setting, our learned representations perform comparably to existing fully supervised methods on tasks such as action-conditioned video prediction and planning in the learned action space, while requiring orders of magnitude fewer action labels. Project website: pseudo-url""","""The reviewers had some concerns regarding clarity and evaluation but in general liked various aspects of the paper. The authors did a good job of addressing the reviewers' concerns so acceptance is recommended."""
1363,"""Global-to-local Memory Pointer Networks for Task-Oriented Dialogue""","['pointer networks', 'memory networks', 'task-oriented dialogue systems', 'natural language processing']","""End-to-end task-oriented dialogue is challenging since knowledge bases are usually large, dynamic and hard to incorporate into a learning framework. We propose the global-to-local memory pointer (GLMP) networks to address this issue. In our model, a global memory encoder and a local memory decoder are proposed to share external knowledge. The encoder encodes dialogue history, modifies global contextual representation, and generates a global memory pointer. The decoder first generates a sketch response with unfilled slots. Next, it passes the global memory pointer to filter the external knowledge for relevant information, then instantiates the slots via the local memory pointers. We empirically show that our model can improve copy accuracy and mitigate the common out-of-vocabulary problem. As a result, GLMP is able to improve over the previous state-of-the-art models in both simulated bAbI Dialogue dataset and human-human Stanford Multi-domain Dialogue dataset on automatic and human evaluation.""","""Interesting paper applying memory networks that encode external knowledge (represented in the form of triples) and conversation context for task oriented dialogues. Experiments demonstrate improvements over the state of the art on two public datasets. Notation and presentation in the first version of the paper were not very clear, hence many question and answers were exchanged during the reviews. """
1364,"""Skip-gram word embeddings in hyperbolic space""","['word embeddings', 'hyperbolic', 'skip-gram']","""Embeddings of tree-like graphs in hyperbolic space were recently shown to surpass their Euclidean counterparts in performance by a large margin. Inspired by these results, we present an algorithm for learning word embeddings in hyperbolic space from free text. An objective function based on the hyperbolic distance is derived and included in the skip-gram negative-sampling architecture from word2vec. The hyperbolic word embeddings are then evaluated on word similarity and analogy benchmarks. The results demonstrate the potential of hyperbolic word embeddings, particularly in low dimensions, though without clear superiority over their Euclidean counterparts. We further discuss subtleties in the formulation of the analogy task in curved spaces.""","""although the proposed method could be considered an interesting application to recently popular hypobolic space to word embeddings, it is unclear why this needs to be done so. experiments also do not support why or whether the application of hyperbolic space to word embedding is necessary."""
1365,"""Learning Goal-Conditioned Value Functions with one-step Path rewards rather than Goal-Rewards""","['Floyd-Warshall', 'Reinforcement learning', 'goal conditioned value functions', 'multi-goal']","""Multi-goal reinforcement learning (MGRL) addresses tasks where the desired goal state can change for every trial. State-of-the-art algorithms model these problems such that the reward formulation depends on the goals, to associate them with high reward. This dependence introduces additional goal reward resampling steps in algorithms like Hindsight Experience Replay (HER) that reuse trials in which the agent fails to reach the goal by recomputing rewards as if reached states were psuedo-desired goals. We propose a reformulation of goal-conditioned value functions for MGRL that yields a similar algorithm, while removing the dependence of reward functions on the goal. Our formulation thus obviates the requirement of reward-recomputation that is needed by HER and its extensions. We also extend a closely related algorithm, Floyd-Warshall Reinforcement Learning, from tabular domains to deep neural networks for use as a baseline. Our results are competetive with HER while substantially improving sampling efficiency in terms of reward computation. ""","""This manuscript presents a reinterpretation of hindsight experience replay which aims to avoid recomputing the reward function, and investigates Floyd-Warshall RL in the function approximation setting. The paper was judged as relatively clear. The authors report a slight improvement in computational cost, which some reviewers called into question. However, all of the reviewers pointed out that the experimental evidence for the method's superiority is weak. Two reviewers additionally raised that this wasn't significantly different than the standard formulation of Hindsight Experience Replay, which doesn't require the computation of rewards for relabeled goals. Ultimately, reviewers were in agreement that the novelty of the method and quality of the obtained results rendered the work insufficient for publication. The Area Chair concurs, and urges the authors to consider the reviewers' pointers to the existing literature in order to clarify their contribution for subsequent submission."""
1366,"""Improving Sequence-to-Sequence Learning via Optimal Transport""","['NLP', 'optimal transport', 'sequence to sequence', 'natural language processing']","""Sequence-to-sequence models are commonly trained via maximum likelihood estimation (MLE). However, standard MLE training considers a word-level objective, predicting the next word given the previous ground-truth partial sentence. This procedure focuses on modeling local syntactic patterns, and may fail to capture long-range semantic structure. We present a novel solution to alleviate these issues. Our approach imposes global sequence-level guidance via new supervision based on optimal transport, enabling the overall characterization and preservation of semantic features. We further show that this method can be understood as a Wasserstein gradient flow trying to match our model to the ground truth sequence distribution. Extensive experiments are conducted to validate the utility of the proposed approach, showing consistent improvements over a wide variety of NLP tasks, including machine translation, abstractive text summarization, and image captioning.""","""The paper proposes the idea of using optimal transport to evaluate the semantic correspondence between two sets of words predicted by the model and ground truth sequences. Strong empirical results are presented which support the use of optimal transport in conjunction with log-likelihood for training sequence models. I appreciate the improvements to the manuscript during the review process, and I encourage the authors to address the rest of the comments in the final version."""
1367,"""Dynamic Early Terminating of Multiply Accumulate Operations for Saving Computation Cost in Convolutional Neural Networks""","['Convolutional neural network', 'Early terminating', 'Dynamic model optimization']","""Deep learning has been attracting enormous attention from academia as well as industry due to its great success in many artificial intelligence applications. As more applications are developed, the need for implementing a complex neural network model on an energy-limited edge device becomes more critical. To this end, this paper proposes a new optimization method to reduce the computation efforts of convolutional neural networks. The method takes advantage of the fact that some convolutional operations are actually wasteful since their outputs are pruned by the following activation or pooling layers. Basically, a convolutional filter conducts a series of multiply-accumulate (MAC) operations. We propose to set a checkpoint in the MAC process to determine whether a filter could terminate early based on the intermediate result. Furthermore, a fine-tuning process is conducted to recover the accuracy drop due to the applied checkpoints. The experimental results show that the proposed method can save approximately 50% MAC operations with less than 1% accuracy drop for CIFAR-10 example model and Network in Network on the CIFAR-10 and CIFAR-100 datasets. Additionally, compared with the state-of- the-art method, the proposed method is more effective on the CIFAR-10 dataset and is competitive on the CIFAR-100 dataset.""","""This paper proposes a new method for speeding up convolutional neural networks. It uses the idea of early terminating the computation of convolutional layers. It saves FLOPs, but the reviewers raised a critical concern that it doesn't save wall-clock time. The time overhead is about 4 or 5 times of the original model. There is not any reduced execution time but much longer. The authors agreed that ""the overhead on the inference time is certainly an issue of our method"". The work is not mature and practical. recommend for rejection. """
1368,"""On Learning Heteroscedastic Noise Models within Differentiable Bayes Filters""","['bayesian filtering', 'heteroscedastic noise', 'deep learning']","""In many robotic applications, it is crucial to maintain a belief about the state of a system, like the location of a robot or the pose of an object. These state estimates serve as input for planning and decision making and provide feedback during task execution. Recursive Bayesian Filtering algorithms address the state estimation problem, but they require a model of the process dynamics and the sensory observations as well as noise estimates that quantify the accuracy of these models. Recently, multiple works have demonstrated that the process and sensor models can be learned by end-to-end training through differentiable versions of Recursive Filtering methods. However, even if the predictive models are known, finding suitable noise models remains challenging. Therefore, many practical applications rely on very simplistic noise models. Our hypothesis is that end-to-end training through differentiable Bayesian Filters enables us to learn more complex heteroscedastic noise models for the system dynamics. We evaluate learning such models with different types of filtering algorithms and on two different robotic tasks. Our experiments show that especially for sampling-based filters like the Particle Filter, learning heteroscedastic noise models can drastically improve the tracking performance in comparison to using constant noise models.""","""This paper shows experiments in favor of learning and using heteroscedastic noise models for differentiable Bayes filter. Reviewers agree that this is interesting and also very useful for the community. However, they have also found plenty of issues with the presentation, execution and evaluations shown in the paper. Post rebuttal, one of the reviewer increased their score, but the other has reduced the score. Overall, the reviewers are in agreement that more work is required before this work can be accepted. Some of existing work on variational inference has not been included which, I agree, is problematic. Simple methods have been compared but then why these methods were chosen and not the other ones, is not completely clear. The paper definitely can improve on this aspect, clearly discussing relationships to many existing methods and then picking important methods to clearly bring some useful insights about learning heteroscedastic noise. Such insights are currently missing in the paper. Reviewers have given many useful feedback in their review, and I believe this can be helpful for the authors to improve their work. In its current form, the paper is not ready to be accepted and I recommend rejection. I encourage the authors to resubmit this work. """
1369,"""Aligning Artificial Neural Networks to the Brain yields Shallow Recurrent Architectures""","['Computational Neuroscience', 'Brain-Inspired', 'Neural Networks', 'Simplified Models', 'Recurrent Neural Networks', 'Computer Vision']","""Deep artificial neural networks with spatially repeated processing (a.k.a., deep convolutional ANNs) have been established as the best class of candidate models of visual processing in the primate ventral visual processing stream. Over the past five years, these ANNs have evolved from a simple feedforward eight-layer architecture in AlexNet to extremely deep and branching NASNet architectures, demonstrating increasingly better object categorization performance. Here we ask, as ANNs have continued to evolve in performance, are they also strong candidate models for the brain? To answer this question, we developed Brain-Score, a composite of neural and behavioral benchmarks for determining how brain-like a model is, together with an online platform where models can receive a Brain-Score and compare against other models. Despite high scores, typical deep models from the machine learning community are often hard to map onto the brain's anatomy due to their vast number of layers and missing biologically-important connections, such as recurrence. To further map onto anatomy and validate our approach, we built CORnet-S: an ANN guided by Brain-Score with the anatomical constraints of compactness and recurrence. Although a shallow model with four anatomically mapped areas and recurrent connectivity, CORnet-S is a top model on Brain-Score and outperforms similarly compact models on ImageNet. Analyzing CORnet-S circuitry variants revealed recurrence as the main predictive factor of both Brain-Score and ImageNet top-1 performance. ""","""This work provides two contributions: 1) Brain-Score, that quantifies how a given network's responses compare to responses from natural systems; 2) CORnet-S, an architecture trained to optimize Brain-Score, that performs well on Imagenet. As noted by all reviewers, this work is interesting and shows a promising approach to quantifying how brain-like an architecture is, with the limitations inherent to the fact that there is a lot about natural visual processing that we don't fully understand. However, the work here starts from the premise that being more similar to current metrics of brain processes is by itself a good thing -- without a better understanding of what features of brain processing are responsible for good performance and which are mere by-products, this premise is not one that would appeal to most of ICLR audience. In fact, the best performing architectures on imagenet are not the best scoring for Brain-Score. Overall, this work is quite intriguing and well presented, but as pointed out by some reviewers, requires a ""leap of faith"" in matching signatures of brain processes that most of the ICLR audience is unlikely to be willing to take."""
1370,"""Efficient Multi-Objective Neural Architecture Search via Lamarckian Evolution""","['Neural Architecture Search', 'AutoML', 'AutoDL', 'Deep Learning', 'Evolutionary Algorithms', 'Multi-Objective Optimization']","""Architecture search aims at automatically finding neural architectures that are competitive with architectures designed by human experts. While recent approaches have achieved state-of-the-art predictive performance for image recognition, they are problematic under resource constraints for two reasons: (1) the neural architectures found are solely optimized for high predictive performance, without penalizing excessive resource consumption; (2)most architecture search methods require vast computational resources. We address the first shortcoming by proposing LEMONADE, an evolutionary algorithm for multi-objective architecture search that allows approximating the Pareto-front of architectures under multiple objectives, such as predictive performance and number of parameters, in a single run of the method. We address the second shortcoming by proposing a Lamarckian inheritance mechanism for LEMONADE which generates children networks that are warmstarted with the predictive performance of their trained parents. This is accomplished by using (approximate) network morphism operators for generating children. The combination of these two contributions allows finding models that are on par or even outperform different-sized NASNets, MobileNets, MobileNets V2 and Wide Residual Networks on CIFAR-10 and ImageNet64x64 within only one week on eight GPUs, which is about 20-40x less compute power than previous architecture search methods that yield state-of-the-art performance.""","""The paper proposes an evolutionary architecture search method which uses weight inheritance through network morphism to avoid training candidate models from scratch. The method can optimise multiple objectives (e.g. accuracy and inference time), which is relevant for practical applications, and the results are promising and competitive with the state of the art. All reviewers are generally positive about the paper. Reviewers feedback on improving presentation and adding experiments with a larger number of objectives has been addressed in the new revision. I strongly encourage the authors to add experiments on the full ImageNet dataset (not just 64x64) and/or language modelling -- the two benchmarks widely used in neural architecture search field."""
1371,"""Gaussian-gated LSTM: Improved convergence by reducing state updates""","['time gate', 'faster convergence', 'trainability', 'rnn', 'computational budget']","""Recurrent neural networks can be difficult to train on long sequence data due to the well-known vanishing gradient problem. Some architectures incorporate methods to reduce RNN state updates, therefore allowing the network to preserve memory over long temporal intervals. To address these problems of convergence, this paper proposes a timing-gated LSTM RNN model, called the Gaussian-gated LSTM (g-LSTM). The time gate controls when a neuron can be updated during training, enabling longer memory persistence and better error-gradient flow. This model captures long-temporal dependencies better than an LSTM and the time gate parameters can be learned even from non-optimal initialization values. Because the time gate limits the updates of the neuron state, the number of computes needed for the network update is also reduced. By adding a computational budget term to the training loss, we can obtain a network which further reduces the number of computes by at least 10x. Finally, by employing a temporal curriculum learning schedule for the g-LSTM, we can reduce the convergence time of the equivalent LSTM network on long sequences.""","""perhaps the biggest issue with the proposed approach is that the proposed approach, which supposedly addresses the issue of capturing long-term dependency with a faster convergence, was only tested on problems with largely fixed length. with the proposed k_n gate being defined as a gaussian with a single mean (per unit?) and variance, it is important and interesting to know how this network would cope with examples of vastly varying lengths. in addition, r3 made good points about comparison against conventional LSTM and how it should be done with careful hyperparameter tuning and based on conventional known setups. this submission will be greatly strengthened with more experiments using a better set of benchmarks and by more carefully placing its contribution w.r.t. other recent advances."""
1372,"""Overcoming Catastrophic Forgetting for Continual Learning via Model Adaptation""","['overcoming forgetting', 'model adaptation', 'continual learning']","""Learning multiple tasks sequentially is important for the development of AI and lifelong learning systems. However, standard neural network architectures suffer from catastrophic forgetting which makes it difficult for them to learn a sequence of tasks. Several continual learning methods have been proposed to address the problem. In this paper, we propose a very different approach, called Parameter Generation and Model Adaptation (PGMA), to dealing with the problem. The proposed approach learns to build a model, called the solver, with two sets of parameters. The first set is shared by all tasks learned so far and the second set is dynamically generated to adapt the solver to suit each test example in order to classify it. Extensive experiments have been carried out to demonstrate the effectiveness of the proposed approach.""","""This paper presents a promising model to avoid catastrophic forgetting in continual learning. The model consists of a) a data generator to be used at training time to replay past examples (and removes the need for storage of data or labels), b) a dynamic parameter generator that given a test input produces the parameters of a classification model, and c) a solver (the actual classifier). The advantages of such combination is that no parameter increase or network expansion is needed to learn a new task, and no previous data needs to be stored for memory replay. There is reviewer disagreement on this paper. AC can confirm that all three reviewers have read the author responses and have significantly contributed to the revision of the manuscript. All three reviewers and AC note the following potential weaknesses: (1) presentation clarity needed substantial improvement. Notably, the authors revised the paper several times while incorporating the reviewers suggestions regarding presentation clarity. R2 has raised the final rating from 4 to 5 while retaining doubts about clarity. (2) weak empirical evidence: evaluation with more than three tasks and using more recent/stronger baseline methods would substantially strengthen the evaluation (R2, R3). AC would like to report the authors added an experiment with five tasks and provided a verbal comparison with ""Riemannian Walk for Incremental Learning: Understanding Forgetting and Intransigence"", ECCV-2018 by reporting the authors results on the MNIST dataset. (3) as noted by R2, an ablation study of different model components could strengthen the evaluation. The authors included such ablation study in Table 4 of the revised paper. (4) reproducibility of the model could be difficult (R1). In their response, the authors promised to make the code publicly available. AC can confirm that all three reviewers have contributed to the final discussion. Given the effort of the reviewers and authors in revising this work and its potential novelty, the AC decided that the paper could be accepted, but the authors are strongly urged to further improve presentation clarity in the final revision if possible. """
1373,"""Heated-Up Softmax Embedding""",[],"""Metric learning aims at learning a distance which is consistent with the semantic meaning of the samples. The problem is generally solved by learning an embedding, such that the samples of the same category are close (compact) while samples from different categories are far away (spread-out) in the embedding space. One popular way of generating such embeddings is to use the second-to-last layer of a deep neural network trained as a classifier with the softmax cross-entropy loss. In this paper, we show that training classifiers with different temperatures of the softmax function lead to different distributions of the embedding space. And finding a balance between the compactness, 'spread-out' and the generalization ability of the feature is critical in metric learning. Leveraging these insights, we propose a 'heating-up' strategy to train a classifier with increasing temperatures. Extensive experiments show that the proposed method achieves state-of-the-art embeddings on a variety of metric learning benchmarks. ""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The method and justification are clear - The quantitative results are promising. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. - The contribution is minor - Analysis of the properties of the method is lacking. The first point was the major factor in the final decision. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. Reviewer opinion was quite divergent but both AR1 and AR2 had concerns about the 2 weaknesses mentioned in the previous section (which remained after the author rebuttal). 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. No consensus was reached. The source of disagreement was on how to weigh the pros vs the cons. The final decision was aligned with the lower ratings. The AC agrees that the contribution is minor. """
1374,"""Detecting Out-Of-Distribution Samples Using Low-Order Deep Features Statistics""","['computer vision', 'out-of-distribution detection', 'image classification']","""The ability to detect when an input sample was not drawn from the training distribution is an important desirable property of deep neural networks. In this paper, we show that a simple ensembling of first and second order deep feature statistics can be exploited to effectively differentiate in-distribution and out-of-distribution samples. Specifically, we observe that the mean and standard deviation within feature maps differs greatly between in-distribution and out-of-distribution samples. Based on this observation, we propose a simple and efficient plug-and-play detection procedure that does not require re-training, pre-processing or changes to the model. The proposed method outperforms the state-of-the-art by a large margin in all standard benchmarking tasks, while being much simpler to implement and execute. Notably, our method improves the true negative rate from 39.6% to 95.3% when 95% of in-distribution (CIFAR-100) are correctly detected using a DenseNet and the out-of-distribution dataset is TinyImageNet resize. The source code of our method will be made publicly available.""","""The paper proposes a simple method for detecting out-of-distribution samples. The authors' major finding is that mean and standard deviation within feature maps can be used as an input for classifying out-of-distribution (OOD) samples. The proposed method is simple and practical. The reviewers and AC note the following potential weaknesses: (1) limited novelty and somewhat ad-hoc approach, i.e., it is not too surprising to expect that such statistics can be useful for the purpose. Some theoretical justification might help. (2) arguable experimental settings, i.e., the performance highly varies depending on validation (even in the revised draft), and sometimes irrationally good. It also depends on the choice of classifier. For (2), I think the whole evaluation should be done assuming that we don't know how it looks the OOD set. Under the setting, the authors should compare the proposed method and existing ones for fair comparisons. AC understands the authors follows the same experimental settings of some previous work addressing this problem, but it's time that this is changed. Indeed, a recent paper by Lee at al. 2018 considers such a setting for detecting more general types of abnormal samples including OOD. In overall, the proposed idea is simple and easy to use. However, AC decided that the authors need more significant works to publish the work. """
1375,"""Learning Robust Representations by Projecting Superficial Statistics Out""","['domain generalization', 'robustness']","""Despite impressive performance as evaluated on i.i.d. holdout data, deep neural networks depend heavily on superficial statistics of the training data and are liable to break under distribution shift. For example, subtle changes to the background or texture of an image can break a seemingly powerful classifier. Building on previous work on domain generalization, we hope to produce a classifier that will generalize to previously unseen domains, even when domain identifiers are not available during training. This setting is challenging because the model may extract many distribution-specific (superficial) signals together with distribution-agnostic (semantic) signals. To overcome this challenge, we incorporate the gray-level co-occurrence matrix (GLCM) to extract patterns that our prior knowledge suggests are superficial: they are sensitive to the texture but unable to capture the gestalt of an image. Then we introduce two techniques for improving our networks' out-of-sample performance. The first method is built on the reverse gradient method that pushes our model to learn representations from which the GLCM representation is not predictable. The second method is built on the independence introduced by projecting the model's representation onto the subspace orthogonal to GLCM representation's. We test our method on the battery of standard domain generalization data sets and, interestingly, achieve comparable or better performance as compared to other domain generalization methods that explicitly require samples from the target distribution for training.""","""The paper presents a new approach for domain generalization whereby the original supervised model is trained with an explicit objective to ignore so called superficial statistics present in the training set but which may not be present in future test sets. The paper proposes using a differentiable variant of gray-level co-occurrence matrix to capture the textural information and then experiments with two techniques for learning feature invariance. All reviewers agree the approach is novel, unique, and potentially high impact to the community. The main issues center around reproducibility as well as the intended scope of problems this approach addresses. The authors have offered to include further discussions in the final version to address these points. Doing so will strengthen the paper and aid the community in building upon this work. """
1376,"""Graph Convolutional Network with Sequential Attention For Goal-Oriented Dialogue Systems""","['Goal-oriented Dialogue Systems', 'Graph Convolutional Networks']","""Domain specific goal-oriented dialogue systems typically require modeling three types of inputs, viz., (i) the knowledge-base associated with the domain, (ii) the history of the conversation, which is a sequence of utterances and (iii) the current utterance for which the response needs to be generated. While modeling these inputs, current state-of-the-art models such as Mem2Seq typically ignore the rich structure inherent in the knowledge graph and the sentences in the conversation context. Inspired by the recent success of structure-aware Graph Convolutional Networks (GCNs) for various NLP tasks such as machine translation, semantic role labeling and document dating, we propose a memory augmented GCN for goal-oriented dialogues. Our model exploits (i) the entity relation graph in a knowledge-base and (ii) the dependency graph associated with an utterance to compute richer representations for words and entities. Further, we take cognizance of the fact that in certain situations, such as, when the conversation is in a code-mixed language, dependency parsers may not be available. We show that in such situations we could use the global word co-occurrence graph and use it to enrich the representations of utterances. We experiment with the modified DSTC2 dataset and its recently released code-mixed versions in four languages and show that our method outperforms existing state-of-the-art methods, using a wide range of evaluation metrics.""","""This paper describes a graph convolutional network (GCN) approach to capture relational information in natural language as well as knowledge sources for goal-oriented dialogue systems. Relational information is captured by dependency parses, and when there is code switching in the input language, word co-occurrence information is used instead. Experiments on the modified DSTC2 dataset show significant improvements over baselines. The original version of the paper lacked comparison to some SOTA baselines as also raised by the reviewers, these are included in the revised version. Although the results show improvements over other approaches, it is arguable BLEU and ROUGE scores are not good enough for this task. Inclusion of human evaluation in the results would be very useful. """
1377,"""Fake Sentence Detection as a Training Task for Sentence Encoding""",[],""" Sentence encoders are typically trained on generative language modeling tasks with large unlabeled datasets. While these encoders achieve strong results on many sentence-level tasks, they are difficult to train with long training cycles. We introduce fake sentence detection as a new discriminative training task for learning sentence encoders. We automatically generate fake sentences by corrupting original sentences from a source collection and train the encoders to produce representations that are effective at detecting fake sentences. This binary classification task turns to be quite efficient for training sentence encoders. We compare a basic BiLSTM encoder trained on this task with strong sentence encoding models (Skipthought and FastSent) trained on a language modeling task. We find that the BiLSTM trains much faster on fake sentence detection (20 hours instead of weeks) using smaller amounts of data (1M instead of 64M sentences). Further analysis shows the learned representations also capture many syntactic and semantic properties expected from good sentence representations.""","""This paper presents a new unsupervised training objective for sentence-to-vector encoding, and shows that it produces representations that often work slightly better than those produced by some prominent earlier work. The reviewers have some concerns about presentation, but the main issuewhich all three reviewers pointed towas the lack of strong recent baselines. Sentence-to-vector representation learning is a fairly active field with an accepted approach to evaluation, and this paper seems to omit conspicuous promising baselines. This includes labeled-data pretraining methods which are known to work well for English (including results from the cited Conneau paper)while these may be difficult to generalize beyond English, this paper does not attempt such a generalization. This also includes more recent unlabeled-data methods like ULMFiT or Radford et al.'s Transformer which could be easily trained on the same sources of data used here. The authors argue in the comments that these language models tend to use more parameters, but these additional parameters are only used during pretraining, so I don't find this objection compelling enough to warrant leaving out baselines of this kind. Baselines of both kinds have been known for at least a year and come with distributed models and code for close comparison."""
1378,"""Physiological Signal Embeddings (PHASE) via Interpretable Stacked Models""","['Representation learning', 'transfer learning', 'health', 'machine learning', 'physiological signals', 'interpretation', 'feature attributions', 'shapley values', 'univariate embeddings', 'LSTMs', 'XGB', 'neural networks', 'stacked models', 'model pipelines', 'interpretable stacked models']","""In health, machine learning is increasingly common, yet neural network embedding (representation) learning is arguably under-utilized for physiological signals. This inadequacy stands out in stark contrast to more traditional computer science domains, such as computer vision (CV), and natural language processing (NLP). For physiological signals, learning feature embeddings is a natural solution to data insufficiency caused by patient privacy concerns -- rather than share data, researchers may share informative embedding models (i.e., representation models), which map patient data to an output embedding. Here, we present the PHASE (PHysiologicAl Signal Embeddings) framework, which consists of three components: i) learning neural network embeddings of physiological signals, ii) predicting outcomes based on the learned embedding, and iii) interpreting the prediction results by estimating feature attributions in the ""stacked"" models (i.e., feature embedding model followed by prediction model). PHASE is novel in three ways: 1) To our knowledge, PHASE is the first instance of transferal of neural networks to create physiological signal embeddings. 2) We present a tractable method to obtain feature attributions through stacked models. We prove that our stacked model attributions can approximate Shapley values -- attributions known to have desirable properties -- for arbitrary sets of models. 3) PHASE was extensively tested in a cross-hospital setting including publicly available data. In our experiments, we show that PHASE significantly outperforms alternative embeddings -- such as raw, exponential moving average/variance, and autoencoder -- currently in use. Furthermore, we provide evidence that transferring neural network embedding/representation learners between distinct hospitals still yields performant embeddings and offer recommendations when transference is ineffective.""","""Authors present a technique to learn embeddings over physiological signals independently using univariate LSTMs tasked to predict future values. Supervised methods are them employed over these embeddings. Univariate approach is taken to improve transferability across institutions, and Shapley values are used to provide interpretable insight. The work is interesting, and authors have made a good attempt at answering reviewers' concerns, but more work remains to be done. Pros: - R1 & R3: Well written. - R3: Transferrable embeddings are useful in this domain, and not often researched. Cons: - R3: Method builds embeddings that assume that future task will be relevant to drops in signals. Authors confirm. - R3: Performance improvement is marginal versus baselines. Authors essentially confirm that the small improvement is the accurate number. - R2 & R3: Interpretability evaluation is not sufficient. Medical expert should rate interpretability of results. Authors did not include or revise according to suggestion. """
1379,"""Unsupervised Learning via Meta-Learning""","['unsupervised learning', 'meta-learning']","""A central goal of unsupervised learning is to acquire representations from unlabeled data or experience that can be used for more effective learning of downstream tasks from modest amounts of labeled data. Many prior unsupervised learning works aim to do so by developing proxy objectives based on reconstruction, disentanglement, prediction, and other metrics. Instead, we develop an unsupervised meta-learning method that explicitly optimizes for the ability to learn a variety of tasks from small amounts of data. To do so, we construct tasks from unlabeled data in an automatic way and run meta-learning over the constructed tasks. Surprisingly, we find that, when integrated with meta-learning, relatively simple task construction mechanisms, such as clustering embeddings, lead to good performance on a variety of downstream, human-specified tasks. Our experiments across four image datasets indicate that our unsupervised meta-learning approach acquires a learning algorithm without any labeled data that is applicable to a wide range of downstream classification tasks, improving upon the embedding learned by four prior unsupervised learning methods.""","""Reviewers largely agree that the paper proposes a novel and interesting idea for unsupervised learning through meta learning and the empirical evaluation does a convincing job in demonstrating its effectiveness. There were some concerns on clarity/readability of the paper which seem to have been addressed by the authors. I recommend acceptance. """
1380,"""Posterior Attention Models for Sequence to Sequence Learning""","['posterior inference', 'attention', 'seq2seq learning', 'translation']","""Modern neural architectures critically rely on attention for mapping structured inputs to sequences. In this paper we show that prevalent attention architectures do not adequately model the dependence among the attention and output tokens across a predicted sequence. We present an alternative architecture called Posterior Attention Models that after a principled factorization of the full joint distribution of the attention and output variables, proposes two major changes. First, the position where attention is marginalized is changed from the input to the output. Second, the attention propagated to the next decoding stage is a posterior attention distribution conditioned on the output. Empirically on five translation and two morphological inflection tasks the proposed posterior attention models yield better BLEU score and alignment accuracy than existing attention models.""","""The reviewers of this paper agreed that it has done a stellar job of presenting a novel and principled approach to attention as a latent variable, providing a new and sound set of inference techniques to this end. This builds on top of a discussion of the limitations of existing deterministic approaches to attention, and frames the contribution well in relation to other recurrent and stochastic approaches to attention. While there are a few issues with clarity surrounding some aspects of the proposed method, which the authors are encouraged to fine-tune in their final version, paying careful attention to the review comments, this paper is more or less ready for publication with a few tweaks. It makes a clear, significant, and well-evaluate contribution to the field of attention models in sequence to sequence architectures, and will be of great interest to many attendees at ICLR."""
1381,"""Learnable Embedding Space for Efficient Neural Architecture Compression""","['Network Compression', 'Neural Architecture Search', 'Bayesian Optimization', 'Architecture Embedding']","""We propose a method to incrementally learn an embedding space over the domain of network architectures, to enable the careful selection of architectures for evaluation during compressed architecture search. Given a teacher network, we search for a compressed network architecture by using Bayesian Optimization (BO) with a kernel function defined over our proposed embedding space to select architectures for evaluation. We demonstrate that our search algorithm can significantly outperform various baseline methods, such as random search and reinforcement learning (Ashok et al., 2018). The compressed architectures found by our method are also better than the state-of-the-art manually-designed compact architecture ShuffleNet (Zhang et al., 2018). We also demonstrate that the learned embedding space can be transferred to new settings for architecture search, such as a larger teacher network or a teacher network in a different architecture family, without any training.""","""The authors propose a method to learn a neural network architecture which achieves the same accuracy as a reference network, with fewer parameters through Bayesian Optimization. The search is carried out on embeddings of the neural network architecture using a train bi-directional LSTM. The reviewers generally found the work to be clearly written, and well motivated, with thorough experimentation, particularly in the revised version. Given the generally positive reviews from the authors, the AC recommends that the paper be accepted. """
1382,"""Learning to Remember More with Less Memorization""","['memory-augmented neural networks', 'writing optimization']","""Memory-augmented neural networks consisting of a neural controller and an external memory have shown potentials in long-term sequential learning. Current RAM-like memory models maintain memory accessing every timesteps, thus they do not effectively leverage the short-term memory held in the controller. We hypothesize that this scheme of writing is suboptimal in memory utilization and introduces redundant computation. To validate our hypothesis, we derive a theoretical bound on the amount of information stored in a RAM-like system and formulate an optimization problem that maximizes the bound. The proposed solution dubbed Uniform Writing is proved to be optimal under the assumption of equal timestep contributions. To relax this assumption, we introduce modifications to the original solution, resulting in a solution termed Cached Uniform Writing. This method aims to balance between maximizing memorization and forgetting via overwriting mechanisms. Through an extensive set of experiments, we empirically demonstrate the advantages of our solutions over other recurrent architectures, claiming the state-of-the-arts in various sequential modeling tasks. ""","""Well-written paper that motivates through theoretical analysis new memory writing methods in memory augmented neural networks. Extensive experimental analysis support and demonstrate the advantages of the new solutions over other recurrent architectures. Reviewers suggested extension and clarification of the analysis presented in the paper, for example, for different memory sizes. The paper was revised accordingly. Another important suggestion was considering ACT as a baseline. Authors explained clearly why it wasn't considered as a baseline, and updated the paper to include references and explanations in the paper as well."""
1383,"""TequilaGAN: How To Easily Identify GAN Samples""","['Generative Adversarial Networks', 'Deep Learning']","""In this paper we show strategies to easily identify fake samples generated with the Generative Adversarial Network framework. One strategy is based on the statistical analysis and comparison of raw pixel values and features extracted from them. The other strategy learns formal specifications from the real data and shows that fake samples violate the specifications of the real data. We show that fake samples produced with GANs have a universal signature that can be used to identify fake samples. We provide results on MNIST, CIFAR10, music and speech data.""","""The paper points out a statistical properties of GAN samples which allows their identification as synthetic. The paper was praised by one reviewer as well-written, easy to follow, and addressing an interesting topic. Another added that the authors did an excellent job of ""probing into different statistical perspectives"", and the fact that they did not confine their investigation to images. Two reviewers leveraged the criticism that various properties discovered are not surprising given the loss functions and associated metrics as well as the inductive biases of continuous-valued generator networks. Tests employed were criticized as ad hoc, and reviewers felt that their generality was limited given their reduced sensitivity on certain modalities. (While Figure 10b is raised by the authors several times in the discussion, and the test statistics of samples are noted to be closer to the test data than to the random baseline, the test falsely rejects the null [p-value ~= 0.0] for non-synthetic test data.) I would encourage the authors to continue this line of inquiry as it is overall agreed to be an interesting topic of relevance and increasing importance, however based on the criticisms of reviewers and the content of the ensuing discussion I do not recommend acceptance at this time."""
1384,"""Hierarchical Attention: What Really Counts in Various NLP Tasks""","['attention', 'hierarchical', 'machine reading comprehension', 'poem generation']","""Attention mechanisms in sequence to sequence models have shown great ability and wonderful performance in various natural language processing (NLP) tasks, such as sentence embedding, text generation, machine translation, machine reading comprehension, etc. Unfortunately, existing attention mechanisms only learn either high-level or low-level features. In this paper, we think that the lack of hierarchical mechanisms is a bottleneck in improving the performance of the attention mechanisms, and propose a novel Hierarchical Attention Mechanism (Ham) based on the weighted sum of different layers of a multi-level attention. Ham achieves a state-of-the-art BLEU score of 0.26 on Chinese poem generation task and a nearly 6.5% averaged improvement compared with the existing machine reading comprehension models such as BIDAF and Match-LSTM. Furthermore, our experiments and theorems reveal that Ham has greater generalization and representation ability than existing attention mechanisms. ""","""The authors propose a hierarchical attention layer which combines intermediate layers of multi-level attention. While this is a simple idea, and the authors show some improvements over the baselines, the authors raised a number of concerns about the validity of the chosen baselines, and the lack of more detailed evaluations on additional tasks and analysis of the results. Given the incremental nature of the work, and the significant concerns raised by the reviewers, the AC is recommending that this paper be rejected."""
1385,"""Benchmarking Neural Network Robustness to Common Corruptions and Perturbations""","['robustness', 'benchmark', 'convnets', 'perturbations']","""In this paper we establish rigorous benchmarks for image classifier robustness. Our first benchmark, ImageNet-C, standardizes and expands the corruption robustness topic, while showing which classifiers are preferable in safety-critical applications. Then we propose a new dataset called ImageNet-P which enables researchers to benchmark a classifier's robustness to common perturbations. Unlike recent robustness research, this benchmark evaluates performance on common corruptions and perturbations not worst-case adversarial perturbations. We find that there are negligible changes in relative corruption robustness from AlexNet classifiers to ResNet classifiers. Afterward we discover ways to enhance corruption and perturbation robustness. We even find that a bypassed adversarial defense provides substantial common perturbation robustness. Together our benchmarks may aid future work toward networks that robustly generalize.""","""The reviewers have all recommended accepting this paper thus I am as well. Based on the reviews and the selectivity of the single track for oral presentations, I am only recommending acceptance as a poster."""
1386,"""Improving the Generalization of Adversarial Training with Domain Adaptation""","['adversarial training', 'domain adaptation', 'adversarial example', 'deep learning']","""By injecting adversarial examples into training data, adversarial training is promising for improving the robustness of deep learning models. However, most existing adversarial training approaches are based on a specific type of adversarial attack. It may not provide sufficiently representative samples from the adversarial domain, leading to a weak generalization ability on adversarial examples from other attacks. Moreover, during the adversarial training, adversarial perturbations on inputs are usually crafted by fast single-step adversaries so as to scale to large datasets. This work is mainly focused on the adversarial training yet efficient FGSM adversary. In this scenario, it is difficult to train a model with great generalization due to the lack of representative adversarial samples, aka the samples are unable to accurately reflect the adversarial domain. To alleviate this problem, we propose a novel Adversarial Training with Domain Adaptation (ATDA) method. Our intuition is to regard the adversarial training on FGSM adversary as a domain adaption task with limited number of target domain samples. The main idea is to learn a representation that is semantically meaningful and domain invariant on the clean domain as well as the adversarial domain. Empirical evaluations on Fashion-MNIST, SVHN, CIFAR-10 and CIFAR-100 demonstrate that ATDA can greatly improve the generalization of adversarial training and the smoothness of the learned models, and outperforms state-of-the-art methods on standard benchmark datasets. To show the transfer ability of our method, we also extend ATDA to the adversarial training on iterative attacks such as PGD-Adversial Training (PAT) and the defense performance is improved considerably.""","""The paper presents an interesting idea for increasing the robustness of adversarial defenses by combining with existing domain adaptation approaches. All reviewers agree that the paper is well written and clearly articulates the approach and contribution. The main areas of weakness is that the experiments focus on small datasets, namely CiFAR and MNIST. That being said, the algorithm is reasonably ablated on the data explored and the authors provided valuable new experimental evidence during the rebuttal phase and in response to the public comment. """
1387,"""Asynchronous SGD without gradient delay for efficient distributed training""","['SGD', 'distributed asynchronous training', 'deep learning', 'optimisation']","""Asynchronous distributed gradient descent algorithms for training of deep neural networks are usually considered as inefficient, mainly because of the Gradient delay problem. In this paper, we propose a novel asynchronous distributed algorithm that tackles this limitation by well-thought-out averaging of model updates, computed by workers. The algorithm allows computing gradients along the process of gradient merge, thus, reducing or even completely eliminating worker idle time due to communication overhead, which is a pitfall of existing asynchronous methods. We provide theoretical analysis of the proposed asynchronous algorithm, and show its regret bounds. According to our analysis, the crucial parameter for keeping high convergence rate is the maximal discrepancy between local parameter vectors of any pair of workers. As long as it is kept relatively small, the convergence rate of the algorithm is shown to be the same as the one of a sequential online learning. Furthermore, in our algorithm, this discrepancy is bounded by an expression that involves the staleness parameter of the algorithm, and is independent on the number of workers. This is the main differentiator between our approach and other solutions, such as Elastic Asynchronous SGD or Downpour SGD, in which that maximal discrepancy is bounded by an expression that depends on the number of workers, due to gradient delay problem. To demonstrate effectiveness of our approach, we conduct a series of experiments on image classification task on a cluster with 4 machines, equipped with a commodity communication switch and with a single GPU card per machine. Our experiments show a linear scaling on 4-machine cluster without sacrificing the test accuracy, while eliminating almost completely worker idle time. Since our method allows using commodity communication switch, it paves a way for large scale distributed training performed on commodity clusters.""","""Improving the staleness of asynchronous SGD is an important topic. This paper proposed an algorithm to restrict the staleness and provided theoretical analysis. However, the reviewers did not consider the proposed algorithm a significant contribution. The paper still did not solve the staleness problem, and it was lack of discussion or experimental comparison with the state of the art ASGD algorithms. Reviewer 3 also found the explanation of the algorithm hard to follow."""
1388,"""MERCI: A NEW METRIC TO EVALUATE THE CORRELATION BETWEEN PREDICTIVE UNCERTAINTY AND TRUE ERROR""","['evaluation metric', 'predictive uncertainty', 'deep learning']","""As deep learning applications are becoming more and more pervasive, the question of evaluating the reliability of a prediction becomes a central question in the machine learning community. This domain, known as predictive uncertainty, has come under the scrutiny of research groups developing Bayesian approaches to deep learning such as Monte Carlo Dropout. Unfortunately, for the time being, the real goal of predictive uncertainty has been swept under the rug. Indeed, Bayesian approaches are solely evaluated in terms of raw performance of the prediction, while the quality of the estimated uncertainty is not assessed. One contribution of this article is to draw attention on existing metrics developed in the forecast community, designed to evaluate both the sharpness and the calibration of predictive uncertainty. Sharpness refers to the concentration of the predictive distributions and calibration to the consistency between the predicted uncertainty level and the actual errors. We further analyze the behavior of these metrics on regression problems when deep convolutional networks are involved and for several current predictive uncertainty approaches. A second contribution of this article is to propose an alternative metric that is more adapted to the evaluation of relative uncertainty assessment and directly applicable to regression with deep learning. This metric is evaluated and compared with existing ones on a toy dataset as well as on the problem of monocular depth estimation. ""","""Reviewers are in a consensus and recommended to reject after engaging with the authors. Please take reviewers' comments into consideration to improve your submission should you decide to resubmit. """
1389,"""Scaling shared model governance via model splitting""","['deep learning', 'reinforcement learning', 'multi-party computation']","""Currently the only techniques for sharing governance of a deep learning model are homomorphic encryption and secure multiparty computation. Unfortunately, neither of these techniques is applicable to the training of large neural networks due to their large computational and communication overheads. As a scalable technique for shared model governance, we propose splitting deep learning model between multiple parties. This paper empirically investigates the security guarantee of this technique, which is introduced as the problem of model completion: Given the entire training data set or an environment simulator, and a subset of the parameters of a trained deep learning model, how much training is required to recover the models original performance? We define a metric for evaluating the hardness of the model completion problem and study it empirically in both supervised learning on ImageNet and reinforcement learning on Atari and DeepMind Lab. Our experiments show that (1) the model completion problem is harder in reinforcement learning than in supervised learning because of the unavailability of the trained agents trajectories, and (2) its hardness depends not primarily on the number of parameters of the missing part, but more so on their type and location. Our results suggest that model splitting might be a feasible technique for shared model governance in some settings where training is very expensive.""","""As all the reviewers have highlighted, there is some interesting analysis in this paper on understanding which models can be easier to complete. The experiments are quite thorough, and seem reproducible. However, the biggest limitation---and the ones that is making it harder for the reviewers to come to a consensus---is the fact that the motivation seems mismatched with the provided approach. There is quite a lot of focus on security, and being robust to an adversary. Model splitting is proposed as a reasonable solution. However, the Model Completion hardness measure proposed is insufficiently justified, both in that its not clear what security guarantees it provides nor is it clear why training time was chosen over other metrics (like number of samples, as mentioned by a reviewer). If this measure had been previously proposed, and the focus of this paper was to provide empirical insight, that might be fine, but that does not appear to be the case. This mismatch is evident also in the writing in the paper. After the introduction, the paper largely reads as understanding how retrainable different architectures are under which problem settings, when replacing an entire layer, with little to no mention of security or privacy. In summary, this paper has some interesting ideas, but an unclear focus. The proposed strategy should be better justified. Or, maybe even better for the larger ICLR audience, the provided analysis could be motivated for other settings, such as understanding convergence rates or trainability in neural networks."""
1390,"""Faster Training by Selecting Samples Using Embeddings""","['Machine Learning', 'Embeddings', 'Training Time', 'Optimization', 'Autoencoders']","""Long training times have increasingly become a burden for researchers by slowing down the pace of innovation, with some models taking days or weeks to train. In this paper, a new, general technique is presented that aims to speed up the training process by using a thinned-down training dataset. By leveraging autoencoders and the unique properties of embedding spaces, we are able to filter training datasets to include only those samples that matter the most. Through evaluation on a standard CIFAR-10 image classification task, this technique is shown to be effective. With this technique, training times can be reduced with a minimal loss in accuracy. Conversely, given a fixed training time budget, the technique was shown to improve accuracy by over 50%. This technique is a practical tool for achieving better results with large datasets and limited computational budgets.""","""The paper proposes a filtering technique to use less training examples in order to train faster; the filtering step is done with an autoencoder. Experiments are done on CIFAR-10. Reviewers point to a lack of convincing experiments, weak evidence, lack of experimental details. Overall, all reviewers converge to reject this paper, and I agree with them."""
1391,"""Learning from Noisy Demonstration Sets via Meta-Learned Suitability Assessor""","['Imitation Learning', 'Noisy Demonstration Set', 'Meta-Learning']","""A noisy and diverse demonstration set may hinder the performances of an agent aiming to acquire certain skills via imitation learning. However, state-of-the-art imitation learning algorithms often assume the optimality of the given demonstration set. In this paper, we address such optimal assumption by learning only from the most suitable demonstrations in a given set. Suitability of a demonstration is estimated by whether imitating it produce desirable outcomes for achieving the goals of the tasks. For more efficient demonstration suitability assessments, the learning agent should be capable of imitating a demonstration as quick as possible, which shares similar spirit with fast adaptation in the meta-learning regime. Our framework, thus built on top of Model-Agnostic Meta-Learning, evaluates how desirable the imitated outcomes are, after adaptation to each demonstration in the set. The resulting assessments hence enable us to select suitable demonstration subsets for acquiring better imitated skills. The videos related to our experiments are available at: pseudo-url""","""The reviewers raised a number of major concerns including the incremental novelty of the proposed (if any), insufficient explanation, and, most importantly, insufficient and inadequate experimental evaluation presented. The authors did not provide any rebuttal. Hence, I cannot suggest this paper for presentation at ICLR."""
1392,"""Understanding and Improving Interpolation in Autoencoders via an Adversarial Regularizer""","['autoencoders', 'interpolation', 'unsupervised learning', 'representation learning', 'adversarial learning']","""Autoencoders provide a powerful framework for learning compressed representations by encoding all of the information needed to reconstruct a data point in a latent code. In some cases, autoencoders can ""interpolate"": By decoding the convex combination of the latent codes for two datapoints, the autoencoder can produce an output which semantically mixes characteristics from the datapoints. In this paper, we propose a regularization procedure which encourages interpolated outputs to appear more realistic by fooling a critic network which has been trained to recover the mixing coefficient from interpolated data. We then develop a simple benchmark task where we can quantitatively measure the extent to which various autoencoders can interpolate and show that our regularizer dramatically improves interpolation in this setting. We also demonstrate empirically that our regularizer produces latent codes which are more effective on downstream tasks, suggesting a possible link between interpolation abilities and learning useful representations.""","""The reviewers have reached a consensus that this paper is very interesting and add insights into interpolation in autoencoders."""
1393,"""Accelerated Gradient Flow for Probability Distributions""","['Optimal transportation', 'Mean-field optimal control', 'Wasserstein gradient flow', 'Markov-chain Monte-Carlo']","""This paper presents a methodology and numerical algorithms for constructing accelerated gradient flows on the space of probability distributions. In particular, we extend the recent variational formulation of accelerated gradient methods in wibisono2016 from vector valued variables to probability distributions. The variational problem is modeled as a mean-field optimal control problem. The maximum principle of optimal control theory is used to derive Hamilton's equations for the optimal gradient flow. The Hamilton's equation are shown to achieve the accelerated form of density transport from any initial probability distribution to a target probability distribution. A quantitative estimate on the asymptotic convergence rate is provided based on a Lyapunov function construction, when the objective functional is displacement convex. Two numerical approximations are presented to implement the Hamilton's equations as a system of N interacting particles. The continuous limit of the Nesterov's algorithm is shown to be a special case with N=1. The algorithm is illustrated with numerical examples. ""","""This paper developed an accelerated gradient flow in the space of probability measures. Unfortunately, the reviewers think the practical usefulness of the proposed approach is not sufficiently supported by realistic experiments, and the clarity of the paper need to be significantly improved. The authors' rebuttal resolved some of the confusion the reviewers had, but we believe further substantial improvement will make this work a much stronger contribution. """
1394,"""Manifold regularization with GANs for semi-supervised learning""","['semi-supervised learning', 'generative adversarial networks', 'manifold regularization']","""Generative Adversarial Networks are powerful generative models that can model the manifold of natural images. We leverage this property to perform manifold regularization by approximating a variant of the Laplacian norm using a Monte Carlo approximation that is easily computed with the GAN. When incorporated into the semi-supervised feature-matching GAN we achieve state-of-the-art results for semi-supervised learning on CIFAR-10 benchmarks when few labels are used, with a method that is significantly easier to implement than competing methods. We find that manifold regularization improves the quality of generated images, and is affected by the quality of the GAN used to approximate the regularizer.""","""The paper proposes a method to perform manifold regularization for semi-supervised learning using GANs. Although the SSL results in the paper are competitive with existing methods, R1 and R3 are concerned about the novelty of the work in the light of recent manifold regularization SSL papers with GANs, a point that the AC agrees with. Given the borderline reviews and limited novelty of the core method, the paper just falls short of the acceptance threshold for ICLR. """
1395,"""BIGSAGE: unsupervised inductive representation learning of graph via bi-attended sampling and global-biased aggregating""","['network embedding', 'unsupervised learning', 'inductive learning']","""Different kinds of representation learning techniques on graph have shown significant effect in downstream machine learning tasks. Recently, in order to inductively learn representations for graph structures that is unobservable during training, a general framework with sampling and aggregating (GraphSAGE) was proposed by Hamilton and Ying and had been proved more efficient than transductive methods on fileds like transfer learning or evolving dataset. However, GraphSAGE is uncapable of selective neighbor sampling and lack of memory of known nodes that've been trained. To address these problems, we present an unsupervised method that samples neighborhood information attended by co-occurring structures and optimizes a trainable global bias as a representation expectation for each node in the given graph. Experiments show that our approach outperforms the state-of-the-art inductive and unsupervised methods for representation learning on graphs.""","""AR2 is concerned about the marginal novelty, weak experiments and very high complexity of the algorithm. AR3 is concerned about lack of theoretical analysis and parameter setting. AR4 is concerned that the proposed method is useful in very restricted settings and the paper is incremental. Unfortunately, with strong critique from reviewers regarding the novelty, complexity, poor presentation and restricted setting, this draft cannot be accepted by ICLR."""
1396,"""Multi-Task Learning for Semantic Parsing with Cross-Domain Sketch""","['semantic parsing', 'natural language understanding', 'machine learning']","""Semantic parsing which maps a natural language sentence into a formal machine-readable representation of its meaning, is highly constrained by the limited annotated training data. Inspired by the idea of coarse-to-fine, we propose a general-to-detailed neural network(GDNN) by incorporating cross-domain sketch(CDS) among utterances and their logic forms. For utterances in different domains, the General Network will extract CDS using an encoder-decoder model in a multi-task learning setup. Then for some utterances in a specific domain, the Detailed Network will generate the detailed target parts using sequence-to-sequence architecture with advanced attention to both utterance and generated CDS. Our experiments show that compared to direct multi-task learning, CDS has improved the performance in semantic parsing task which converts users' requests into meaning representation language(MRL). We also use experiments to illustrate that CDS works by adding some constraints to the target decoding process, which further proves the effectiveness and rationality of CDS.""","""Interesting approach aiming to leverage cross domain schemas and generic semantic parsing (based on meaning representation language, MRL) for language understanding. Experiments have been performed on the recently released SNIPS corpus and comparisons have been made with multiple recent multi-task learning approaches. Unfortunately, the proposed approach falls short in comparison to the slot gated attention work by Goo et al. The motivation and description of the cross domain schemas can be improved in the paper, and for replication of experiments it would be useful to include how the annotations are extended for this purpose. Experimental results could be extended to the other available corpora mentioned in the paper (ATIS and GEO). """
1397,"""A Max-Affine Spline Perspective of Recurrent Neural Networks""","['RNN', 'max-affine spline operators']","""We develop a framework for understanding and improving recurrent neural networks (RNNs) using max-affine spline operators (MASOs). We prove that RNNs using piecewise affine and convex nonlinearities can be written as a simple piecewise affine spline operator. The resulting representation provides several new perspectives for analyzing RNNs, three of which we study in this paper. First, we show that an RNN internally partitions the input space during training and that it builds up the partition through time. Second, we show that the affine slope parameter of an RNN corresponds to an input-specific template, from which we can interpret an RNN as performing a simple template matching (matched filtering) given the input. Third, by carefully examining the MASO RNN affine mapping, we prove that using a random initial hidden state corresponds to an explicit L2 regularization of the affine parameters, which can mollify exploding gradients and improve generalization. Extensive experiments on several datasets of various modalities demonstrate and validate each of the above conclusions. In particular, using a random initial hidden states elevates simple RNNs to near state-of-the-art performers on these datasets. ""","""While the reformulation of RNNs is not practical as it is missing sigmoids and tanhs that are common in LSTMs it does provide an interesting analysis of traditional RNNs and a technique that's novel for many in the ICLR community. """
1398,"""Fixup Initialization: Residual Learning Without Normalization""","['deep learning', 'residual networks', 'initialization', 'batch normalization', 'layer normalization']","""Normalization layers are a staple in state-of-the-art deep neural network architectures. They are widely believed to stabilize training, enable higher learning rate, accelerate convergence and improve generalization, though the reason for their effectiveness is still an active research topic. In this work, we challenge the commonly-held beliefs by showing that none of the perceived benefits is unique to normalization. Specifically, we propose fixed-update initialization (Fixup), an initialization motivated by solving the exploding and vanishing gradient problem at the beginning of training via properly rescaling a standard initialization. We find training residual networks with Fixup to be as stable as training with normalization -- even for networks with 10,000 layers. Furthermore, with proper regularization, Fixup enables residual networks without normalization to achieve state-of-the-art performance in image classification and machine translation.""","""The paper explores the effect of normalization and initialization in residual networks, motivated by the need to avoid exploding and vanishing activations and gradients. Based on some theoretical analysis of stepsizes in SGD, the authors propose a sensible but effective way of initializing a network that greatly increases training stability. In a nutshell, the method comes down to initializing the residual layers such that a single step of SGD results in a change in activations that is invariant to the depth of the network. The experiments in the paper provide supporting evidence for the benefits; the authors were able to train networks of up to 10,000 layers deep. The experiments have sufficient depth to support the claims. Overall, the method seems to be a simple but effective technique for learning very deep residual networks. While some aspects of the network have been used in earlier work, such as initializing residual branches to output zeros, these earlier methods lacked the rescaling aspect, which seems crucial to the performance of this network. The reviewers agree that the papers provides interesting ideas and significant theoretical and empirical contributions. The main concerns by the reviewers were addressed by the author responses. The AC finds that the remaining concerns raised by the reviewers are minor and insufficient for rejection of the paper. """
1399,"""Integrated Steganography and Steganalysis with Generative Adversarial Networks""","['Steganography', 'Steganography', 'Security', 'Generative Adversarial Networks']","""Recently, generative adversarial network is the hotspot in research areas and industrial application areas. It's application on data generation in computer vision is most common usage. This paper extends its application to data hiding and security area. In this paper, we propose the novel framework to integrate steganography and steganalysis processes. The proposed framework applies generative adversarial networks as the core structure. The discriminative model simulate the steganalysis process, which can help us understand the sensitivity of cover images to semantic changes. The steganography generative model is to generate stego image which is aligned with the original cover image, and attempts to confuse steganalysis discriminative model. The introduction of cycle discriminative model and inconsistent loss can help to enhance the quality and security of generated stego image in the iterative training process. Training dataset is mixed with intact images as well as intentional attacked images. The mix training process can further improve the robustness and security of new framework. Through the qualitative, quantitative experiments and analysis, this novel framework shows compelling performance and advantages over the current state-of-the-art methods in steganography and steganalysis benchmarks.""","""1. Describe the strengths of the paper. As pointed out by the reviewers and based on your expert opinion. - The problem and approach, steganography via GANs, is interesting. - The results seem promising. 2. Describe the weaknesses of the paper. As pointed out by the reviewers and based on your expert opinion. Be sure to indicate which weaknesses are seen as salient for the decision (i.e., potential critical flaws), as opposed to weaknesses that the authors can likely fix in a revision. The original submission was imprecise and difficult to follow and, while the AC acknowledges that the authors made significant improvements, the current version still needs some work before it's clear enough to be acceptable for publication. 3. Discuss any major points of contention. As raised by the authors or reviewers in the discussion, and how these might have influenced the decision. If the authors provide a rebuttal to a potential reviewer concern, its a good idea to acknowledge this and note whether it influenced the final decision or not. This makes sure that author responses are addressed adequately. Concerns varied by reviewer and there was no main point of contention. 4. If consensus was reached, say so. Otherwise, explain what the source of reviewer disagreement was and why the decision on the paper aligns with one set of reviewers or another. The reviewers did not reach a consensus. The final decision is aligned with the less positive reviewers, one of whom was very confident in his/her review. The AC agrees that the paper should be made clearer and more precise. """
1400,"""A Mean Field Theory of Batch Normalization""","['theory', 'batch normalization', 'mean field theory', 'trainability']","""We develop a mean field theory for batch normalization in fully-connected feedforward neural networks. In so doing, we provide a precise characterization of signal propagation and gradient backpropagation in wide batch-normalized networks at initialization. Our theory shows that gradient signals grow exponentially in depth and that these exploding gradients cannot be eliminated by tuning the initial weight variances or by adjusting the nonlinear activation function. Indeed, batch normalization itself is the cause of gradient explosion. As a result, vanilla batch-normalized networks without skip connections are not trainable at large depths for common initialization schemes, a prediction that we verify with a variety of empirical simulations. While gradient explosion cannot be eliminated, it can be reduced by tuning the network close to the linear regime, which improves the trainability of deep batch-normalized networks without residual connections. Finally, we investigate the learning dynamics of batch-normalized networks and observe that after a single step of optimization the networks achieve a relatively stable equilibrium in which gradients have dramatically smaller dynamic range. Our theory leverages Laplace, Fourier, and Gegenbauer transforms and we derive new identities that may be of independent interest.""","""This paper provides a mean-field-theory analysis of batch normalization. First there is a negative result as to the necessity of gradient explosion when using batch normalization in a fully connected network. They then provide further insights as to what can be done about this, along with experiments to confirm their theoretical predictions. The reviewers (and random commenters) found this paper very interesting. The reviewers were unanimous in their vote to accept."""
1401,"""Predictive Local Smoothness for Stochastic Gradient Methods""","['stochastic gradient method', 'local smoothness', 'linear system', 'AMSGrad']","""Stochastic gradient methods are dominant in nonconvex optimization especially for deep models but have low asymptotical convergence due to the fixed smoothness. To address this problem, we propose a simple yet effective method for improving stochastic gradient methods named predictive local smoothness (PLS). First, we create a convergence condition to build a learning rate varied adaptively with local smoothness. Second, the local smoothness can be predicted by the latest gradients. Third, we use the adaptive learning rate to update the stochastic gradients for exploring linear convergence rates. By applying the PLS method, we implement new variants of three popular algorithms: PLS-stochastic gradient descent (PLS-SGD), PLS-accelerated SGD (PLS-AccSGD), and PLS-AMSGrad. Moreover, we provide much simpler proofs to ensure their linear convergence. Empirical results show that our variants have better performance gains than the popular algorithms, such as, faster convergence and alleviating explosion and vanish of gradients.""","""Dear authors, All reviewers pointed to severe issues with the analysis, making the paper unsuitable for publication to ICLR. Please take their comments into account should you decide to resubmit this work."""
1402,""" The relativistic discriminator: a key element missing from standard GAN""","['AI', 'deep learning', 'generative models', 'GAN']","""In standard generative adversarial network (SGAN), the discriminator estimates the probability that the input data is real. The generator is trained to increase the probability that fake data is real. We argue that it should also simultaneously decrease the probability that real data is real because 1) this would account for a priori knowledge that half of the data in the mini-batch is fake, 2) this would be observed with divergence minimization, and 3) in optimal settings, SGAN would be equivalent to integral probability metric (IPM) GANs. We show that this property can be induced by using a relativistic discriminator which estimate the probability that the given real data is more realistic than a randomly sampled fake data. We also present a variant in which the discriminator estimate the probability that the given real data is more realistic than fake data, on average. We generalize both approaches to non-standard GAN loss functions and we refer to them respectively as Relativistic GANs (RGANs) and Relativistic average GANs (RaGANs). We show that IPM-based GANs are a subset of RGANs which use the identity function. Empirically, we observe that 1) RGANs and RaGANs are significantly more stable and generate higher quality data samples than their non-relativistic counterparts, 2) Standard RaGAN with gradient penalty generate data of better quality than WGAN-GP while only requiring a single discriminator update per generator update (reducing the time taken for reaching the state-of-the-art by 400%), and 3) RaGANs are able to generate plausible high resolutions images (256x256) from a very small sample (N=2011), while GAN and LSGAN cannot; these images are of significantly better quality than the ones generated by WGAN-GP and SGAN with spectral normalization. The code is freely available on pseudo-url.""","""All authors agree that the relativistic discriminator is an interesting idea, and a useful proposal to improve the stability and sample quality of GANs. In earlier drafts there were some clarity issues and missing details, but those have been fixed to the satisfaction of the reviewers. Both R1 and R3 expressed a desire for a more theoretical justification of why the relativistic discriminator should work better, but the empirical results are strong enough that this can be left for future work."""
1403,"""Multi-Objective Value Iteration with Parameterized Threshold-Based Safety Constraints""","['reinforcement learning', 'Markov decision processes', 'safety constraints', 'multi-objective optimization', 'geometric analysis']","""We consider an environment with multiple reward functions. One of them represents goal achievement and the others represent instantaneous safety conditions. We consider a scenario where the safety rewards should always be above some thresholds. The thresholds are parameters with values that differ between users. %The thresholds are not known at the time the policy is being designed. We efficiently compute a family of policies that cover all threshold-based constraints and maximize the goal achievement reward. We introduce a new parameterized threshold-based scalarization method of the reward vector that encodes our objective. We present novel data structures to store the value functions of the Bellman equation that allow their efficient computation using the value iteration algorithm. We present results for both discrete and continuous state spaces. ""","""The main issue with the work in its current form is a lack of motivation and some clarity issues. The paper presents some interesting ideas, and will be much stronger when it incorporates a more clear discussion on motivation, both for the problem setting and the proposed solutions. The writing itself could also be significantly improved. """
1404,"""Graph Wavelet Neural Network""","['graph convolution', 'graph wavelet transform', 'graph Fourier transform', 'semi-supervised learning']","""We present graph wavelet neural network (GWNN), a novel graph convolutional neural network (CNN), leveraging graph wavelet transform to address the shortcomings of previous spectral graph CNN methods that depend on graph Fourier transform. Different from graph Fourier transform, graph wavelet transform can be obtained via a fast algorithm without requiring matrix eigendecomposition with high computational cost. Moreover, graph wavelets are sparse and localized in vertex domain, offering high efficiency and good interpretability for graph convolution. The proposed GWNN significantly outperforms previous spectral graph CNNs in the task of graph-based semi-supervised classification on three benchmark datasets: Cora, Citeseer and Pubmed.""","""AR1 and AR3 have found this paper interesting in terms of replacing the spectral operations in GCN by wavelet operations. However, AR4 was more critical about the poor complexity of the proposed method compared to approximations in Hammond et al. AR4 was also right to find the proposed work similar to Chebyshev approximations in ChebNet and to highlight that the proposed approach is only marginally better than GCN. On balance, all reviewers find some merit in this work thus AC advocates an accept. The authors are asked to keep the contents of the final draft as agreed with AR4 (and other reviewers) during rebuttal without making any further theoretical changes/brushing over various new claims/ideas unsolicited by the reviewers (otherwise such changes would require passing the draft again through reviewers)."""
1405,"""Relational Forward Models for Multi-Agent Learning""","['multi-agent reinforcement learning', 'relational reasoning', 'forward models']","""The behavioral dynamics of multi-agent systems have a rich and orderly structure, which can be leveraged to understand these systems, and to improve how artificial agents learn to operate in them. Here we introduce Relational Forward Models (RFM) for multi-agent learning, networks that can learn to make accurate predictions of agents' future behavior in multi-agent environments. Because these models operate on the discrete entities and relations present in the environment, they produce interpretable intermediate representations which offer insights into what drives agents' behavior, and what events mediate the intensity and valence of social interactions. Furthermore, we show that embedding RFM modules inside agents results in faster learning systems compared to non-augmented baselines. As more and more of the autonomous systems we develop and interact with become multi-agent in nature, developing richer analysis tools for characterizing how and why agents make decisions is increasingly necessary. Moreover, developing artificial agents that quickly and safely learn to coordinate with one another, and with humans in shared environments, is crucial.""",""" pros: - interesting application of graph networks for relational inference in MARL, allowing interpretability and, as the results show, increasing performance - better learning curves in several games - somewhat better forward prediction than baselines cons: - perhaps some lingering confusion about the amount of improvement over the LSTM+MLP baseline Many of the reviewer's other issues have been addressed in revision and I recommend acceptance."""
1406,"""DARTS: Differentiable Architecture Search""","['deep learning', 'autoML', 'neural architecture search', 'image classification', 'language modeling']","""This paper addresses the scalability challenge of architecture search by formulating the task in a differentiable manner. Unlike conventional approaches of applying evolution or reinforcement learning over a discrete and non-differentiable search space, our method is based on the continuous relaxation of the architecture representation, allowing efficient search of the architecture using gradient descent. Extensive experiments on CIFAR-10, ImageNet, Penn Treebank and WikiText-2 show that our algorithm excels in discovering high-performance convolutional architectures for image classification and recurrent architectures for language modeling, while being orders of magnitude faster than state-of-the-art non-differentiable techniques.""","""This paper introduces a very simple but effective method for the neural architecture search problem. The key idea of the method is a particular continuous relaxation of the architecture representation to enable gradient descent-like differentiable optimization. Results are quite good. Source code is also available. A concern of the approach is the (possibly large) integrality gap between the continuous solution and the discretized architecture. The solution provided in the paper is a heuristic without guarantees. Overall, this is a good paper. I recommend acceptance."""
1407,"""Select Via Proxy: Efficient Data Selection For Training Deep Networks""","['data selection', 'deep learning', 'uncertainty sampling']","""At internet scale, applications collect a tremendous amount of data by logging user events, analyzing text, and collecting images. This data powers a variety of machine learning models for tasks such as image classification, language modeling, content recommendation, and advertising. However, training large models over all available data can be computationally expensive, creating a bottleneck in the development of new machine learning models. In this work, we develop a novel approach to efficiently select a subset of training data to achieve faster training with no loss in model predictive performance. In our approach, we first train a small proxy model quickly, which we then use to estimate the utility of individual training data points, and then select the most informative ones for training the large target model. Extensive experiments show that our approach leads to a 1.6x and 1.8x speed-up on CIFAR10 and SVHN by selecting 60% and 50% subsets of the data, while maintaining the predictive performance of the model trained on the entire dataset.""","""There reviewers unanimously recommend rejecting this paper and, although I believe the submission is close to something that should be accepted, I concur with their recommendation. This paper should be improved and published elsewhere, but the improvements needed are too extensive to justify accepting it in this conference. I believe the authors are studying a very promising algorithm and it is irrelevant that the algorithm is a relatively obvious one. Ideally, the contribution would be a clear experimental investigation of the utility of this algorithm in realistic conditions. Unfortunately, the existing experiments are not quite there. I agree with reviewer 2 that the method is not particularly novel. However, I disagree that this is a problem, so it was not a factor in my decision. Novelty can be overrated and it would be fine if the experiments were sufficiently insightful and comprehensive. I believe experiments that train for a single epoch on the reduced dataset are absolutely essential in order to understand the potential usefulness of the algorithm. Although it would of course be better, I do not think it is necessary to find datasets traditionally trained in a single pass. You can do single epoch training on other datasets even though it will likely degrade the final validation error reached. This is the type of small scale experiment the paper should include, additional apples-to-apples baselines just need to be added. Also, there are many large language modeling datasets where it is reasonable to make only a single pass over the training set. The goal should be to simulate, as closely as is possible, the sort of conditions that would actually justify using the algorithm in practice. Another issue with the experimental protocol is that, when claiming a potential speedup, one must tune the baseline to get a particular result in the fewest steps. Most baselines get tuned to produce the best final validation error given a fixed number of steps. But when studying training speed, we should fix a competitive goal error rate and then tune for speed. Careful attention to these experimental protocol issues would be important. """
1408,"""Predictive Uncertainty through Quantization""","['variational inference', 'information bottleneck', 'bayesian deep learning', 'latent variable models', 'amortized variational inference', 'uncertainty', 'learning non-linearities']","""High-risk domains require reliable confidence estimates from predictive models. Deep latent variable models provide these, but suffer from the rigid variational distributions used for tractable inference, which err on the side of overconfidence. We propose Stochastic Quantized Activation Distributions (SQUAD), which imposes a flexible yet tractable distribution over discretized latent variables. The proposed method is scalable, self-normalizing and sample efficient. We demonstrate that the model fully utilizes the flexible distribution, learns interesting non-linearities, and provides predictive uncertainty of competitive quality. ""","""The reviewers agree this paper is not good enough for ICLR."""
1409,"""State-Denoised Recurrent Neural Networks""","['recurrent nets', 'attractor nets', 'denoising', 'sequence processing']","""Recurrent neural networks (RNNs) are difficult to train on sequence processing tasks, not only because input noise may be amplified through feedback, but also because any inaccuracy in the weights has similar consequences as input noise. We describe a method for denoising the hidden state during training to achieve more robust representations thereby improving generalization performance. Attractor dynamics are incorporated into the hidden state to `clean up' representations at each step of a sequence. The attractor dynamics are trained through an auxillary denoising loss to recover previously experienced hidden states from noisy versions of those states. This state-denoised recurrent neural network (SDRNN) performs multiple steps of internal processing for each external sequence step. On a range of tasks, we show that the SDRNN outperforms a generic RNN as well as a variant of the SDRNN with attractor dynamics on the hidden state but without the auxillary loss. We argue that attractor dynamics---and corresponding connectivity constraints---are an essential component of the deep learning arsenal and should be invoked not only for recurrent networks but also for improving deep feedforward nets and intertask transfer.""","""The paper is well written and develops a novel and original architecture and technique for RNNs to learn attractors for their hidden states (based on an auxiliary denoising training of an attractor network). All reviewers and AC found the idea very interesting and a promising direction of research for RNNs. However all also agreed that the experimental validation was currently too limited, in type and size of task and data, as in scope. Reviewers demand experimental comparisons with other (simpler) denoising / regularization techniques; more in depth experimental validation and analysis of the state-denoising behaviour; as well as experiments on larger datasets and more ambitious tasks. """
1410,"""Learning What to Remember: Long-term Episodic Memory Networks for Learning from Streaming Data""","['Memory Network', 'Lifelong Learning']","""Current generation of memory-augmented neural networks has limited scalability as they cannot efficiently process data that are too large to fit in the external memory storage. One example of this is lifelong learning scenario where the model receives unlimited length of data stream as an input which contains vast majority of uninformative entries. We tackle this problem by proposing a memory network fit for long-term lifelong learning scenario, which we refer to as Long-term Episodic Memory Networks (LEMN), that features a RNN-based retention agent that learns to replace less important memory entries based on the retention probability generated on each entry that is learned to identify data instances of generic importance relative to other memory entries, as well as its historical importance. Such learning of retention agent allows our long-term episodic memory network to retain memory entries of generic importance for a given task. We validate our model on a path-finding task as well as synthetic and real question answering tasks, on which our model achieves significant improvements over the memory augmented networks with rule-based memory scheduling as well as an RL-based baseline that does not consider relative or historical importance of the memory.""",""" Pros: - This is an interesting and relevant topic - It is well motivated and mostly clear Cons: - The motivation, large amounts of data such as occur in lifelong learning, is not well examined in the evaluation which focuses on quite small problems. For an example of work which addresses the lifelong memory management issue (though does not learn a memory management policy) see [1]. - In general the evaluation is not adequate to the claims. - Reviewer 2 is concerned with the use of a bi-directional RNN for the comparison of memory entries since it may overfit to order. - Reviewer 1 is somewhat concerned with novelty over other memory management schemes. [1] Scalable Recollections for Continual Lifelong Learning. pseudo-url"""
1411,"""Holographic and other Point Set Distances for Machine Learning""","['point set', 'set', 'permutation-invariant', 'loss function']","""We introduce an analytic distance function for moderately sized point sets of known cardinality that is shown to have very desirable properties, both as a loss function as well as a regularizer for machine learning applications. We compare our novel construction to other point set distance functions and show proof of concept experiments for training neural networks end-to-end on point set prediction tasks such as object detection.""","""This paper proposes permutation invariant loss functions which depend on the distance of sets. This has an interesting interpretation as the roots of a polynomial, and potentially leads to a more efficient method. It is not clear, however, whether the method works well in practice for multiple reasons: (i) the experiments are performed in a limited setting, and the rebuttal specifically declined to consider more realistic datasets, (ii) there is an open question about the stability of the resulting gradients, which has been pointed out both in the paper and the reviews. There was initially a majority vote for rejection. After author response, the only reviewer recommending acceptance wrote ""As the other reviews (and my original review) say, the experimental results are not totally convincing. So I would not champion the paper in its present form."""""
1412,"""Big-Little Net: An Efficient Multi-Scale Feature Representation for Visual and Speech Recognition""","['CNN', 'multi-scale', 'efficiency', 'object recognition', 'speech recognition']","""In this paper, we propose a novel Convolutional Neural Network (CNN) architecture for learning multi-scale feature representations with good tradeoffs between speed and accuracy. This is achieved by using a multi-branch network, which has different computational complexity at different branches with different resolutions. Through frequent merging of features from branches at distinct scales, our model obtains multi-scale features while using less computation. The proposed approach demonstrates improvement of model efficiency and performance on both object recognition and speech recognition tasks, using popular architectures including ResNet, ResNeXt and SEResNeXt. For object recognition, our approach reduces computation by 1/3 while improving accuracy significantly over 1% point than the baselines, and the computational savings can be higher up to 1/2 without compromising the accuracy. Our model also surpasses state-of-the-art CNN acceleration approaches by a large margin in terms of accuracy and FLOPs. On the task of speech recognition, our proposed multi-scale CNNs save 30% FLOPs with slightly better word error rates, showing good generalization across domains.""","""This paper propose a novel CNN architecture for learning multi-scale feature representations with good tradeoffs between speed and accuracy. reviewers generally arrived at a consensus on accept."""
1413,"""Characterizing Audio Adversarial Examples Using Temporal Dependency""","['audio adversarial example', 'mitigation', 'detection', 'machine learning']","""Recent studies have highlighted adversarial examples as a ubiquitous threat to different neural network models and many downstream applications. Nonetheless, as unique data properties have inspired distinct and powerful learning principles, this paper aims to explore their potentials towards mitigating adversarial inputs. In particular, our results reveal the importance of using the temporal dependency in audio data to gain discriminate power against adversarial examples. Tested on the automatic speech recognition (ASR) tasks and three recent audio adversarial attacks, we find that (i) input transformation developed from image adversarial defense provides limited robustness improvement and is subtle to advanced attacks; (ii) temporal dependency can be exploited to gain discriminative power against audio adversarial examples and is resistant to adaptive attacks considered in our experiments. Our results not only show promising means of improving the robustness of ASR systems, but also offer novel insights in exploiting domain-specific data properties to mitigate negative effects of adversarial examples.""","""The authors present a study characterizing adversarial examples in the audio domain. They highlight the importance of temporal dependency when defining defense against adversarial attacks. Strengths - The work presents an interesting analysis of properties of audio adversarial examples, and contrasts it with those in vision literature. - Proposes a novel defense mechanism that is based on the idea of temporal dependency. Weaknesses - The technique identifies adversarial examples but is not able to make the correct prediction. - The reviewers raised issue around clarity, but the authors took the effort to improve the section during the revision process. The reviewers agree that the contribution is significant and useful for the community. There are still some concerns about clarity, which the authors should consider improving in the final version. Overall, the paper received positive reviews and therefore, is recommended to be accepted to the conference."""
1414,"""On Tighter Generalization Bounds for Deep Neural Networks: CNNs, ResNets, and Beyond""","['deep learning', 'generalization error bound', 'convolutional neural networks']","""We propose a generalization error bound for a general family of deep neural networks based on the depth and width of the networks, as well as the spectral norm of weight matrices. Through introducing a novel characterization of the Lipschitz properties of neural network family, we achieve a tighter generalization error bound. We further obtain a result that is free of linear dependence on norms for bounded losses. Besides the general deep neural networks, our results can be applied to derive new bounds for several popular architectures, including convolutional neural networks (CNNs), residual networks (ResNets), and hyperspherical networks (SphereNets). When achieving same generalization errors with previous arts, our bounds allow for the choice of much larger parameter spaces of weight matrices, inducing potentially stronger expressive ability for neural networks.""","""I'm quite concerned by the conversation with Anonymous, entitled ""Why is the dependence..."". My issues concern the empirical Rademacher complexity (ERC) and in particular the choice of the loss class for which the ERC is being computed. This class is obviously data dependent, but the Reviewers concerns centers on the nature of its data dependence. It is not valid to define the classes by the Jacobian's norm on the input data, as this _structure_ over the space of classes is data dependent, which is not kosher. The reviewer was gently pushing the authors towards a very strong assumption... i'm guessing that the jacobian norm over all data sets was bounded by a particular constant. This seems like a whopping assumption. The fact that I can so easily read this concern off of the reviewer's comments and the authors seem to not be able to understand what the reviewer is getting at, concerns me. Besides this concern, it seems that this paper has undergone a rather significant revision. I'm not convinced the new version has been properly reviewed. For a theory paper, I'm concerned about letting work through that's not properly vetted, and I'm really not certain this has been. I suggest the authors consider sending it to COLT."""
1415,"""Variance Networks: When Expectation Does Not Meet Your Expectations""","['deep learning', 'variational inference', 'variational dropout']","""Ordinary stochastic neural networks mostly rely on the expected values of their weights to make predictions, whereas the induced noise is mostly used to capture the uncertainty, prevent overfitting and slightly boost the performance through test-time averaging. In this paper, we introduce variance layers, a different kind of stochastic layers. Each weight of a variance layer follows a zero-mean distribution and is only parameterized by its variance. It means that each object is represented by a zero-mean distribution in the space of the activations. We show that such layers can learn surprisingly well, can serve as an efficient exploration tool in reinforcement learning tasks and provide a decent defense against adversarial attacks. We also show that a number of conventional Bayesian neural networks naturally converge to such zero-mean posteriors. We observe that in these cases such zero-mean parameterization leads to a much better training objective than more flexible conventional parameterizations where the mean is being learned.""","""The authors describe a very counterintuitive type of layer: one with mean zero Gaussian weights. They show that various Bayesian deep learning algorithms tend to converge to layers of this variety. This work represents a step forward in our understanding of bayesian deep learning methods and potentially may shine light on how to improve those methods. """
1416,"""ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware""","['Neural Architecture Search', 'Efficient Neural Networks']","""Neural architecture search (NAS) has a great impact by automatically designing effective neural network architectures. However, the prohibitive computational demand of conventional NAS algorithms (e.g. 10 4 GPU hours) makes it difficult to directly search the architectures on large-scale tasks (e.g. ImageNet). Differentiable NAS can reduce the cost of GPU hours via a continuous representation of network architecture but suffers from the high GPU memory consumption issue (grow linearly w.r.t. candidate set size). As a result, they need to utilize proxy tasks, such as training on a smaller dataset, or learning with only a few blocks, or training just for a few epochs. These architectures optimized on proxy tasks are not guaranteed to be optimal on the target task. In this paper, we present ProxylessNAS that can directly learn the architectures for large-scale target tasks and target hardware platforms. We address the high memory consumption issue of differentiable NAS and reduce the computational cost (GPU hours and GPU memory) to the same level of regular training while still allowing a large candidate set. Experiments on CIFAR-10 and ImageNet demonstrate the effectiveness of directness and specialization. On CIFAR-10, our model achieves 2.08% test error with only 5.7M parameters, better than the previous state-of-the-art architecture AmoebaNet-B, while using 6 fewer parameters. On ImageNet, our model achieves 3.1% better top-1 accuracy than MobileNetV2, while being 1.2 faster with measured GPU latency. We also apply ProxylessNAS to specialize neural architectures for hardware with direct hardware metrics (e.g. latency) and provide insights for efficient CNN architecture design.""","""This paper integrates a bunch of existing approaches for neural architecture search, including OneShot/DARTS, BinaryConnect, REINFORCE, etc. Although the novelty of the paper may be limited, empirical performance seems impressive. The source code is not available. I think this is a borderline paper but maybe good enough for acceptance. """
1417,"""Max-MIG: an Information Theoretic Approach for Joint Learning from Crowds""","['crowdsourcing', 'information theory']","""Eliciting labels from crowds is a potential way to obtain large labeled data. Despite a variety of methods developed for learning from crowds, a key challenge remains unsolved: \emph{learning from crowds without knowing the information structure among the crowds a priori, when some people of the crowds make highly correlated mistakes and some of them label effortlessly (e.g. randomly)}. We propose an information theoretic approach, Max-MIG, for joint learning from crowds, with a common assumption: the crowdsourced labels and the data are independent conditioning on the ground truth. Max-MIG simultaneously aggregates the crowdsourced labels and learns an accurate data classifier. Furthermore, we devise an accurate data-crowds forecaster that employs both the data and the crowdsourced labels to forecast the ground truth. To the best of our knowledge, this is the first algorithm that solves the aforementioned challenge of learning from crowds. In addition to the theoretical validation, we also empirically show that our algorithm achieves the new state-of-the-art results in most settings, including the real-world data, and is the first algorithm that is robust to various information structures. Codes are available at pseudo-url . ""","""This paper proposes an interesting approach to leveraging crowd-sourced labels, along with an ML model learned from the data itself. The reviewers were unanimous in their vote to accept."""
1418,"""Meta-Learning with Individualized Feature Space for Few-Shot Classification""","['few-shot classification', 'meta-learning', 'individualized feature space']","""Meta-learning provides a promising learning framework to address few-shot classification tasks. In existing meta-learning methods, the meta-learner is designed to learn about model optimization, parameter initialization, or similarity metric. Differently, in this paper, we propose to learn how to create an individualized feature embedding specific to a given query image for better classifying, i.e., given a query image, a specific feature embedding tailored for its characteristics is created accordingly, leading to an individualized feature space in which the query image can be more accurately classified. Specifically, we introduce a kernel generator as meta-learner to learn to construct feature embedding for query images. The kernel generator acquires meta-knowledge of generating adequate convolutional kernels for different query images during training, which can generalize to unseen categories without fine-tuning. In two standard few-shot classification data sets, i.e. Omniglot, and \emph{mini}ImageNet, our method shows highly competitive performance. ""","""The reviewers all appreciate the idea, and the competitive performance, however the consensus is that this is a simple extension of the work of Han et al. and therefore the current submission contains little novelty. There are also numerous issues regarding clarity that the reviewers have pointed out. It is unfortunate that the authors have not engaged in discussion with the reviewers to resolve these, however they are encouraged to consider the reviewer feedback in order to improve the paper."""
1419,"""Spread Divergences""","['Generative Adversarial Network', 'Divergence']","""For distributions pseudo-formula and pseudo-formula with different support, the divergence pseudo-formula generally will not exist. We define a spread divergence pseudo-formula on modified pseudo-formula and pseudo-formula and describe sufficient conditions for the existence of such a divergence. We give examples of using a spread divergence to train implicit generative models, including linear models (Principal Components Analysis and Independent Components Analysis) and non-linear models (Deep Generative Networks).""","""This manuscript proposes spread divergences as a technique for extending f-divergences to distributions with different supports. This is achieved by convolving with a noise distribution. This is an important topic worth further study in the community, particularly as it related to training generative models. The reviewers and AC opinions were mixed, with reviewers either being unconvinced about the novelty of the proposed work, or expressing issues about the clarity of the presentation. Further improvement of the clarity, combined with additional convincing experiments would significantly strengthen this submission."""