{
    "Summary": "This paper introduces an adaptive dual-scale denoising approach for low-dimensional diffusion models, addressing the challenge of balancing global structure and local detail in generated samples. The proposed architecture incorporates two parallel branches (global and local) with a learnable, timestep-conditioned weighting mechanism. The method is evaluated on four 2D datasets: circle, dino, line, and moons, showing improvements in sample quality, with KL divergence reductions of up to 12.8% compared to a baseline model.",
    "Strengths": [
        "Novel dual-scale denoising architecture for low-dimensional diffusion models.",
        "Learnable, timestep-conditioned weighting mechanism to balance global and local features.",
        "Comprehensive empirical evaluations on four different 2D datasets.",
        "Significant improvements in sample quality, as evidenced by reductions in KL divergence."
    ],
    "Weaknesses": [
        "Increased computational complexity, with training times approximately doubling.",
        "Performance improvements are dataset-specific and not consistently superior across all datasets.",
        "Limited evaluation metrics; reliance on KL divergence and visual inspection may not fully capture model performance.",
        "Insufficient discussion of potential negative societal impacts.",
        "Limited to low-dimensional data, which restricts the broader impact.",
        "Dense and could benefit from clearer organization."
    ],
    "Originality": 3,
    "Quality": 3,
    "Clarity": 3,
    "Significance": 3,
    "Questions": [
        "How does the model perform on higher-dimensional datasets?",
        "Can the computational complexity be reduced without significantly affecting performance?",
        "What is the impact of hyperparameter tuning on the model's performance?",
        "Can the authors provide more details on the potential generalizability of their approach to higher-dimensional data?",
        "How does the learnable weighting mechanism perform in scenarios with significantly different noise schedules or datasets?",
        "Could the authors clarify the computational trade-offs in more detail, particularly in terms of scalability?",
        "Can the authors provide more insights into the increased computational cost? Are there any potential ways to mitigate this?",
        "Could the authors elaborate on the potential reasons for performance variability on the dino dataset?",
        "Have the authors considered evaluating the model on higher-dimensional datasets or more complex 2D datasets?",
        "Could the authors provide a more detailed breakdown of the training and inference time increases per dataset?",
        "Can the authors provide more details on how the upscaling process affects the local branch's input?",
        "What are the computational requirements and memory usage for the proposed dual-scale model?",
        "Could the authors elaborate on the choice of hyperparameters and any tuning that was performed?"
    ],
    "Limitations": [
        "Increased computational cost, which may limit practical applications.",
        "Potential inconsistency in performance, especially on more complex datasets (e.g., dino).",
        "The trade-off between improved sample quality and computational complexity needs careful consideration.",
        "Limited evaluation on low-dimensional data only.",
        "Performance variability on complex datasets indicates potential instability.",
        "Limited hyperparameter tuning."
    ],
    "Ethical Concerns": false,
    "Soundness": 3,
    "Presentation": 3,
    "Contribution": 3,
    "Overall": 5,
    "Confidence": 4,
    "Decision": "Reject"
}