\relax 
\providecommand\hyper@newdestlabel[2]{}
\providecommand\HyperFirstAtBeginDocument{\AtBeginDocument}
\HyperFirstAtBeginDocument{\ifx\hyper@anchor\@undefined
\global\let\oldcontentsline\contentsline
\gdef\contentsline#1#2#3#4{\oldcontentsline{#1}{#2}{#3}}
\global\let\oldnewlabel\newlabel
\gdef\newlabel#1#2{\newlabelxx{#1}#2}
\gdef\newlabelxx#1#2#3#4#5#6{\oldnewlabel{#1}{{#2}{#3}}}
\AtEndDocument{\ifx\hyper@anchor\@undefined
\let\contentsline\oldcontentsline
\let\newlabel\oldnewlabel
\fi}
\fi}
\global\let\hyper@last\relax 
\gdef\HyperFirstAtBeginDocument#1{#1}
\providecommand\HyField@AuxAddToFields[1]{}
\providecommand\HyField@AuxAddToCoFields[2]{}
\citation{goodfellow2016deep}
\citation{power2022grokking}
\citation{vaswani2017attention}
\@writefile{toc}{\contentsline {section}{\numberline {1}Introduction}{1}{section.1}\protected@file@percent }
\newlabel{sec:intro}{{1}{1}{Introduction}{section.1}{}}
\newlabel{sec:intro@cref}{{[section][1][]1}{[1][1][]1}}
\citation{Glorot2010UnderstandingTD}
\@writefile{toc}{\contentsline {section}{\numberline {2}Related Work}{2}{section.2}\protected@file@percent }
\newlabel{sec:related}{{2}{2}{Related Work}{section.2}{}}
\newlabel{sec:related@cref}{{[section][2][]2}{[1][2][]2}}
\citation{He2015DelvingDI}
\citation{Saxe2013ExactST}
\citation{power2022grokking}
\citation{vaswani2017attention}
\citation{vaswani2017attention}
\citation{bahdanau2014neural}
\@writefile{toc}{\contentsline {subsection}{\numberline {2.1}Weight Initialization Strategies}{3}{subsection.2.1}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.2}Grokking Phenomenon}{3}{subsection.2.2}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {2.3}Transformer Training Dynamics}{3}{subsection.2.3}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {3}Background}{3}{section.3}\protected@file@percent }
\newlabel{sec:background}{{3}{3}{Background}{section.3}{}}
\newlabel{sec:background@cref}{{[section][3][]3}{[1][3][]3}}
\citation{goodfellow2016deep}
\citation{Glorot2010UnderstandingTD}
\citation{He2015DelvingDI}
\citation{Saxe2013ExactST}
\citation{power2022grokking}
\citation{loshchilov2017adamw}
\citation{kingma2014adam}
\@writefile{toc}{\contentsline {subsection}{\numberline {3.1}Problem Setting}{4}{subsection.3.1}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {4}Method}{5}{section.4}\protected@file@percent }
\newlabel{sec:method}{{4}{5}{Method}{section.4}{}}
\newlabel{sec:method@cref}{{[section][4][]4}{[1][4][]5}}
\@writefile{toc}{\contentsline {section}{\numberline {5}Experimental Setup}{5}{section.5}\protected@file@percent }
\newlabel{sec:experimental}{{5}{5}{Experimental Setup}{section.5}{}}
\newlabel{sec:experimental@cref}{{[section][5][]5}{[1][5][]5}}
\@writefile{toc}{\contentsline {subsection}{\numberline {5.1}Dataset}{5}{subsection.5.1}\protected@file@percent }
\@writefile{loa}{\contentsline {algorithm}{\numberline {1}{\ignorespaces Experimental Procedure\relax }}{6}{algorithm.1}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {5.2}Model Architecture}{6}{subsection.5.2}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {5.3}Training Details}{6}{subsection.5.3}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {5.4}Initialization Strategies}{6}{subsection.5.4}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {5.5}Evaluation Metrics}{7}{subsection.5.5}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {5.6}Implementation Details}{7}{subsection.5.6}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {6}Results}{7}{section.6}\protected@file@percent }
\newlabel{sec:results}{{6}{7}{Results}{section.6}{}}
\newlabel{sec:results@cref}{{[section][6][]6}{[1][7][]7}}
\providecommand*\caption@xref[2]{\@setref\relax\@undefined{#1}}
\newlabel{fig:train_acc_x_plus_y}{{1a}{7}{Training Accuracy\relax }{figure.caption.1}{}}
\newlabel{fig:train_acc_x_plus_y@cref}{{[subfigure][1][1]1a}{[1][7][]7}}
\newlabel{sub@fig:train_acc_x_plus_y}{{a}{7}{Training Accuracy\relax }{figure.caption.1}{}}
\newlabel{sub@fig:train_acc_x_plus_y@cref}{{[subfigure][1][1]1a}{[1][7][]7}}
\newlabel{fig:val_acc_x_plus_y}{{1b}{7}{Validation Accuracy\relax }{figure.caption.1}{}}
\newlabel{fig:val_acc_x_plus_y@cref}{{[subfigure][2][1]1b}{[1][7][]7}}
\newlabel{sub@fig:val_acc_x_plus_y}{{b}{7}{Validation Accuracy\relax }{figure.caption.1}{}}
\newlabel{sub@fig:val_acc_x_plus_y@cref}{{[subfigure][2][1]1b}{[1][7][]7}}
\@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces Training and Validation Accuracy for x\_plus\_y task across different initialization methods\relax }}{7}{figure.caption.1}\protected@file@percent }
\newlabel{fig:acc_x_plus_y}{{1}{7}{Training and Validation Accuracy for x\_plus\_y task across different initialization methods\relax }{figure.caption.1}{}}
\newlabel{fig:acc_x_plus_y@cref}{{[figure][1][]1}{[1][7][]7}}
\newlabel{fig:train_loss_x_minus_y}{{2a}{8}{Training Loss\relax }{figure.caption.2}{}}
\newlabel{fig:train_loss_x_minus_y@cref}{{[subfigure][1][2]2a}{[1][7][]8}}
\newlabel{sub@fig:train_loss_x_minus_y}{{a}{8}{Training Loss\relax }{figure.caption.2}{}}
\newlabel{sub@fig:train_loss_x_minus_y@cref}{{[subfigure][1][2]2a}{[1][7][]8}}
\newlabel{fig:val_loss_x_minus_y}{{2b}{8}{Validation Loss\relax }{figure.caption.2}{}}
\newlabel{fig:val_loss_x_minus_y@cref}{{[subfigure][2][2]2b}{[1][7][]8}}
\newlabel{sub@fig:val_loss_x_minus_y}{{b}{8}{Validation Loss\relax }{figure.caption.2}{}}
\newlabel{sub@fig:val_loss_x_minus_y@cref}{{[subfigure][2][2]2b}{[1][7][]8}}
\@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces Training and Validation Loss for x\_minus\_y task across different initialization methods\relax }}{8}{figure.caption.2}\protected@file@percent }
\newlabel{fig:loss_x_minus_y}{{2}{8}{Training and Validation Loss for x\_minus\_y task across different initialization methods\relax }{figure.caption.2}{}}
\newlabel{fig:loss_x_minus_y@cref}{{[figure][2][]2}{[1][7][]8}}
\newlabel{fig:train_acc_x_div_y}{{3a}{8}{Training Accuracy\relax }{figure.caption.3}{}}
\newlabel{fig:train_acc_x_div_y@cref}{{[subfigure][1][3]3a}{[1][7][]8}}
\newlabel{sub@fig:train_acc_x_div_y}{{a}{8}{Training Accuracy\relax }{figure.caption.3}{}}
\newlabel{sub@fig:train_acc_x_div_y@cref}{{[subfigure][1][3]3a}{[1][7][]8}}
\newlabel{fig:val_acc_x_div_y}{{3b}{8}{Validation Accuracy\relax }{figure.caption.3}{}}
\newlabel{fig:val_acc_x_div_y@cref}{{[subfigure][2][3]3b}{[1][7][]8}}
\newlabel{sub@fig:val_acc_x_div_y}{{b}{8}{Validation Accuracy\relax }{figure.caption.3}{}}
\newlabel{sub@fig:val_acc_x_div_y@cref}{{[subfigure][2][3]3b}{[1][7][]8}}
\@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces Training and Validation Accuracy for x\_div\_y task across different initialization methods\relax }}{8}{figure.caption.3}\protected@file@percent }
\newlabel{fig:acc_x_div_y}{{3}{8}{Training and Validation Accuracy for x\_div\_y task across different initialization methods\relax }{figure.caption.3}{}}
\newlabel{fig:acc_x_div_y@cref}{{[figure][3][]3}{[1][7][]8}}
\newlabel{fig:train_loss_permutation}{{4a}{8}{Training Loss\relax }{figure.caption.4}{}}
\newlabel{fig:train_loss_permutation@cref}{{[subfigure][1][4]4a}{[1][8][]8}}
\newlabel{sub@fig:train_loss_permutation}{{a}{8}{Training Loss\relax }{figure.caption.4}{}}
\newlabel{sub@fig:train_loss_permutation@cref}{{[subfigure][1][4]4a}{[1][8][]8}}
\newlabel{fig:val_loss_permutation}{{4b}{8}{Validation Loss\relax }{figure.caption.4}{}}
\newlabel{fig:val_loss_permutation@cref}{{[subfigure][2][4]4b}{[1][8][]8}}
\newlabel{sub@fig:val_loss_permutation}{{b}{8}{Validation Loss\relax }{figure.caption.4}{}}
\newlabel{sub@fig:val_loss_permutation@cref}{{[subfigure][2][4]4b}{[1][8][]8}}
\@writefile{lof}{\contentsline {figure}{\numberline {4}{\ignorespaces Training and Validation Loss for permutation task across different initialization methods\relax }}{8}{figure.caption.4}\protected@file@percent }
\newlabel{fig:loss_permutation}{{4}{8}{Training and Validation Loss for permutation task across different initialization methods\relax }{figure.caption.4}{}}
\newlabel{fig:loss_permutation@cref}{{[figure][4][]4}{[1][8][]8}}
\citation{power2022grokking}
\@writefile{lot}{\contentsline {table}{\numberline {1}{\ignorespaces 95\% Confidence Intervals for Steps to 99\% Validation Accuracy ($S_{99}$)\relax }}{9}{table.caption.5}\protected@file@percent }
\newlabel{tab:confidence_intervals}{{1}{9}{95\% Confidence Intervals for Steps to 99\% Validation Accuracy ($S_{99}$)\relax }{table.caption.5}{}}
\newlabel{tab:confidence_intervals@cref}{{[table][1][]1}{[1][8][]9}}
\@writefile{toc}{\contentsline {section}{\numberline {7}Conclusions}{9}{section.7}\protected@file@percent }
\newlabel{sec:conclusion}{{7}{9}{Conclusions}{section.7}{}}
\newlabel{sec:conclusion@cref}{{[section][7][]7}{[1][9][]9}}
\bibstyle{iclr2024_conference}
\bibdata{references}
\bibcite{bahdanau2014neural}{{1}{2014}{{Bahdanau et~al.}}{{Bahdanau, Cho, and Bengio}}}
\bibcite{Glorot2010UnderstandingTD}{{2}{2010}{{Glorot \& Bengio}}{{Glorot and Bengio}}}
\bibcite{goodfellow2016deep}{{3}{2016}{{Goodfellow et~al.}}{{Goodfellow, Bengio, Courville, and Bengio}}}
\bibcite{He2015DelvingDI}{{4}{2015}{{He et~al.}}{{He, Zhang, Ren, and Sun}}}
\bibcite{kingma2014adam}{{5}{2014}{{Kingma \& Ba}}{{Kingma and Ba}}}
\bibcite{loshchilov2017adamw}{{6}{2017}{{Loshchilov \& Hutter}}{{Loshchilov and Hutter}}}
\bibcite{power2022grokking}{{7}{2022}{{Power et~al.}}{{Power, Burda, Edwards, Babuschkin, and Misra}}}
\bibcite{Saxe2013ExactST}{{8}{2013}{{Saxe et~al.}}{{Saxe, McClelland, and Ganguli}}}
\bibcite{vaswani2017attention}{{9}{2017}{{Vaswani et~al.}}{{Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, and Polosukhin}}}
\ttl@finishall
\gdef \@abspage@last{10}