Indexed text attachment
topos-actions-in-machine-learning-and-ai.tex
Source for Topos Actions in Machine Learning and Artificial Intelligence
View paper · Download original file
Lines 21–120 of 1207. Source text is displayed literally.
21\definecolor{paperaccent}{HTML}{145A51}22\definecolor{papermuted}{HTML}{51655F}23\definecolor{paperrule}{HTML}{B6CBC4}24\hypersetup{25 pdftitle={Topos Actions in Machine Learning and Artificial Intelligence},26 pdfauthor={Anonymous},27 pdfsubject={Prior work and models for local symmetry},28 colorlinks=true,linkcolor=paperaccent,citecolor=paperaccent,urlcolor=paperaccent29}30\AtBeginDocument{\color{paperink}}31\setlength{\columnsep}{.29in}32\setlength{\parindent}{1em}33\setlength{\parskip}{1pt plus .5pt}34\setlength{\emergencystretch}{1.7em}35\linespread{1.035}36\allowdisplaybreaks[1]37\raggedbottom38\clubpenalty=100039\widowpenalty=100040 41\titleformat{\section}{\large\bfseries\raggedright}42 {\color{paperaccent}\thesection}{.55em}{}43\titlespacing*{\section}{0pt}{14pt plus 3pt minus 2pt}{5pt}44\titleformat{\subsection}{\normalsize\bfseries\raggedright}45 {\color{paperaccent}\thesubsection}{.55em}{}46\titlespacing*{\subsection}{0pt}{10pt plus 2pt minus 1pt}{4pt}47\captionsetup{font=small,labelfont={bf,color=paperaccent}}48 49\pagestyle{fancy}50\fancyhf{}51\fancyhead[L]{\footnotesize\sffamily\color{papermuted}Topos actions in machine learning and artificial intelligence}52\fancyfoot[R]{\footnotesize\sffamily\color{papermuted}\thepage}53\renewcommand{\headrulewidth}{0pt}54\fancypagestyle{plain}{%55 \fancyhf{}%56 \fancyfoot[R]{\footnotesize\sffamily\color{papermuted}\thepage}%57 \renewcommand{\headrulewidth}{0pt}%58}59 60\theoremstyle{plain}61\newtheorem{theorem}{Theorem}[section]62\newtheorem{proposition}[theorem]{Proposition}63\newtheorem{lemma}[theorem]{Lemma}64\newtheorem{corollary}[theorem]{Corollary}65\theoremstyle{definition}66\newtheorem{definition}[theorem]{Definition}67\newtheorem{example}[theorem]{Example}68\theoremstyle{remark}69\newtheorem{remark}[theorem]{Remark}70\DeclareMathOperator{\Map}{Map}71\DeclareMathOperator{\Tot}{Tot}72\DeclareMathOperator{\Fun}{Fun}73\DeclareMathOperator{\Hom}{Hom}74\DeclareMathOperator{\grad}{grad}75\DeclareMathOperator{\Ext}{Ext}76\DeclareMathOperator{\Res}{Res}77\DeclareMathOperator{\Kl}{Kl}78\DeclareMathOperator{\id}{id}79\DeclareMathOperator{\rank}{rank}80\newcommand{\E}{\mathcal E}81\newcommand{\Spc}{\mathcal S}82\newcommand{\R}{\mathbb R}83\newcommand{\C}{\mathbb C}84\newcommand{\Z}{\mathbb Z}85\newcommand{\overB}{\E_{/B}}86\tikzcdset{every label/.append style={font=\small,/tikz/auto=false,description}}87\begin{document}88\twocolumn[89\begin{@twocolumnfalse}90\noindent\textcolor{paperaccent}{\rule{\textwidth}{1pt}}\par91\vspace{11pt}92{\noindent\fontsize{25}{30}\selectfont\bfseries93Topos Actions in Machine Learning\\94and Artificial Intelligence\par}95\vspace{8pt}96{\noindent\large\color{papermuted}97Prior work and models for local symmetry\par}98\vspace{10pt}99{\noindent\small November 2026\par}100\vspace{9pt}101\noindent\begin{minipage}{\textwidth}\small102\textbf{Abstract.}103We review uses of topoi, group actions, and descent in machine learning104and artificial intelligence. Existing work models neural networks by105presheaves and stacks, learns automata with nominal symmetries, and106builds architectures from cellular sheaves or groupoid-equivariant107kernels. The literature identified here gives little direct treatment108of learning with a monoidal Grothendieck topos acting by geometric109morphisms. We develop two applications in that setting and its110internal counterpart. A central action gives a quotient-type criterion111for nuisance-invariant layers and distinguishes quotient features from112linear averaging. A nonconstant group sheaf over the circle yields113learnable filters whose coefficient bundle has a M\"obius summand.114Descent determines which parameter choices and update rules give115consistent models across contexts. The results are structural and116are illustrated by exact learning examples.117\par\medskip118\noindent\textbf{Keywords.}\enspace119Topos actions, equivariant learning, descent, sheaf neural networks,120nominal automata, nuisance invariance.