Make paper arXiv-compatible with pdflatex

- Switch from fontspec/TeX Gyre to standard fonts (mathpazo, helvet, courier)
- Replace emoji package with no-op (title is just "TinyTorch")
- Switch from biblatex/biber to natbib/bibtex
- Works with both lualatex (local) and pdflatex (arXiv)

tinytorch/paper/paper.tex

@@ -10,18 +10,11 @@
 \tolerance=1000            % Allow slightly looser spacing to avoid hyphens
 
 % Essential packages
-\usepackage{fontspec}
-% Use TeX Gyre fonts (available in TeX Live, should work on arXiv with xelatex)
-\setmainfont{TeX Gyre Pagella}[
-  Ligatures=TeX,
-  Numbers=Lining
-]
-\setsansfont{TeX Gyre Heros}[
-  Scale=MatchLowercase
-]
-\setmonofont{TeX Gyre Cursor}[
-  Scale=MatchLowercase
-]
+% Using standard pdflatex fonts for arXiv compatibility
+\usepackage[T1]{fontenc}
+\usepackage{mathpazo}      % Palatino for main text and math
+\usepackage[scaled=0.9]{helvet}  % Helvetica for sans-serif
+\usepackage{courier}       % Courier for monospace
 \usepackage{microtype}
 \usepackage{graphicx}
 \usepackage{amsmath}
@@ -31,23 +24,9 @@
 \usepackage{xcolor}
 \usepackage{listings}
 \PassOptionsToPackage{hyphens,spaces,obeyspaces}{url}
-% Use biblatex for better author truncation control
-\usepackage[
-  backend=biber,
-  style=authoryear,
-  natbib=true,
-  maxbibnames=999,
-  maxcitenames=2,
-  minbibnames=999,
-  mincitenames=1,
-  uniquelist=false,
-  giveninits=true,
-  doi=false,
-  isbn=false,
-  url=false,
-  eprint=false
-]{biblatex}
-\addbibresource{references.bib}
+% Use natbib for arXiv compatibility
+\usepackage[round,authoryear]{natbib}
+\bibliographystyle{plainnat}
 \usepackage{hyperref}
 \usepackage{cleveref}
 
@@ -55,7 +34,8 @@
 \makeatletter
 \g@addto@macro{\UrlBreaks}{\UrlOrds}
 \makeatother
-\usepackage{emoji}
+% emoji package not available on arXiv - title will be just "TinyTorch"
+\newcommand{\emoji}[1]{}
 \usepackage{tikz}
 \usetikzlibrary{shapes,arrows,positioning,shadows,calc,backgrounds,decorations.pathreplacing}
 \usepackage{subcaption}
@@ -1113,13 +1093,13 @@ Sutton's Bitter Lesson teaches that general methods leveraging computation ultim
 
 \section*{Acknowledgments}
 
-TinyTorch emerged from CS249r: Machine Learning Systems at Harvard University. We thank the students across multiple offerings whose questions, struggles, and feedback shaped the curriculum into its current form. Their experiences revealed which concepts needed scaffolding, which modules needed restructuring, and where systems intuition failed to transfer. We also thank Colby Banbury and Zishen Wan for their feedback on this work, and the global community around \texttt{mlsysbook.ai} whose engagement continues to refine these materials.
+TinyTorch emerged from CS249r: Tiny Machine Learning at Harvard University, where teaching ML systems on constrained devices revealed that the same principles (memory management, computational efficiency, optimization tradeoffs) apply at every scale. The ``Tiny'' in TinyTorch reflects these origins, but the framework teaches general ML systems fundamentals that transfer from microcontrollers to datacenters. We thank the students across multiple offerings whose questions, struggles, and feedback shaped the curriculum into its current form. We also thank Colby Banbury and Zishen Wan for their feedback on this work, and the global community around \texttt{mlsysbook.ai} whose engagement continues to refine these materials.
 
 \subsection*{Use of Generative AI}
 
-In accordance with ACM policy on authorship, we disclose that generative AI tools were used during the development of both the TinyTorch framework and this manuscript, including for code review, debugging assistance, proofreading, and improving clarity of exposition. We view this as aligned with TinyTorch's mission: just as TinyTorch aims to democratize ML systems education by removing infrastructure barriers, AI tools democratize the creation of educational resources by augmenting individual contributors. A curriculum of this scope would traditionally require a team; AI assistance enabled a single author to develop, test, and document 20 interconnected modules. The author takes full responsibility for all content, and the core intellectual contributions (curriculum structure, pedagogical framework, and systems-first methodology) reflect human design decisions informed by teaching experience.
+In accordance with ACM policy on authorship, we disclose that generative AI tools were used during the development of both the TinyTorch framework and this manuscript, including code review, debugging assistance, proofreading, and improving clarity of exposition. We view this as aligned with TinyTorch's mission: just as TinyTorch aims to democratize ML systems education by removing infrastructure barriers, AI tools democratize the creation of educational resources by augmenting individual contributors. A curriculum of this scope would traditionally require a team; AI assistance enabled a single author to develop, test, and document 20 interconnected modules. The author takes full responsibility for all content, and the core intellectual contributions (curriculum structure, pedagogical framework, and systems first methodology) reflect human design decisions informed by teaching experience.
 
 % Bibliography
-\printbibliography
+\bibliography{references}
 
 \end{document}