Independent scholarship · Open accessThe monthly digest

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 947–1046 of 1207. Source text is displayed literally.

947global section group alone would impose no commutation948conditions and would allow all nine entries of a local949matrix. Requiring internal equivariance leaves three950local coefficients. Choosing flat global coefficients951ties the two shifted terms together, leaving two.952For smooth models the third coefficient $b(t)$ can953be any smooth function on the lifted angle satisfying954$b(t+2\pi)=-b(t)$. For example955$b(t)=\cos(t/2)$ gives a well-defined global filter956$b(t)(R-R^2)$. Every continuous such $b$ has a zero:957if $b(t)$ is nonzero, the values at $t$ and958$t+2\pi$ have opposite signs. Hence a model that959requires this coefficient to stay nonzero can work over960the orientation double cover, where the pulled-back961line is trivial. Learning this coefficient in the962bundle and learning one untwisted scalar are different963parameterization problems.964 965\begin{figure}[tbp]966\centering967\(968\begin{tikzcd}[column sep=4.5em,row sep=2.2em]969 V_t \arrow[r,"A_t"] \arrow[d,"S"'] & V_t \arrow[d,"S"]\\970 V_{t+2\pi}\arrow[r,"A_{t+2\pi}"'] & V_{t+2\pi}971\end{tikzcd}972\)973\caption{A filter descends when it intertwines the change of974frame. Conjugation by the transition $S$ reverses the975coefficient of $R-R^2$.}976\label{fig:twisted-filter}977\end{figure}978 979A coordinatewise activation commutes with the local shift980and the frame permutation. Consequently one may stack981these linear filters with such activations to obtain a982global nonlinear layer. A sum of the three output983coordinates then gives a scalar readout, invariant under984both the local group and the frame transition.985 986A simple regression calculation gives a training987interpretation with a specified observation law.988In a local frame draw $x=e_j$ uniformly over $j$,989draw a hidden sign $\varepsilon\in\{\pm1\}$ fairly990and independently, and observe the target991$y=R^\varepsilon x$. Inverting the frame sends992$(x,y,\varepsilon)$ to $(Sx,Sy,-\varepsilon)$, so993the law descends even though a deterministic forward994shift $R$ does not define a flat global filter.995The least-squares estimator conditional on $x$ is996$(R+R^2)x/2$, an admissible flat filter.997The two conditional target vectors are distinct basis998vectors, so each lies squared distance $1/2$ from999their mean. The best expected squared error using1000only $x$ is therefore $1/2$. The rule $(x,\varepsilon)\mapsto R^\varepsilon x$ has zero error1001if the sign is supplied with its frame transition. The ambiguity here1002is a stated property of the observation law.1003Monodromy supplies the transformation of the hidden1004sign and the constraint on flat filters.1005\par1006\section{Conclusion}1007 1008The literature provides several operational cases of learning1009with symmetry-sensitive structure. Nominal automata support1010active queries over infinitely many names with finitely1011described orbits. Sheaf neural networks use local incidence1012geometry, and groupoid-equivariant networks use transformations1013that are defined only on parts of the input space. Topos and1014stack semantics also supply formal descriptions of network1015behavior and symbolic reasoning. The type of action differs1016across these examples.1017 1018The two models in Section~\ref{sec:applications} give concrete1019consequences of following the type of action through to the1020parameters and observations. A central quotient identifies1021the input orbit as the information available to every1022nuisance-invariant predictor. The circle group sheaf imposes1023a transition law on one learnable coefficient, even though1024its global group of sections is trivial. An equivariant1025optimizer requires compatible metric, loss, and data transport1026in addition to compatible layers. The topological statements1027describe admissible realizations. Prediction error and1028optimization depend on the observation law and update rule1029specified for the model.1030\par1031\begin{thebibliography}{99}1032\small\raggedright1033 1034\bibitem{abramsky-contextuality}1035S. Abramsky and A. Brandenburger. The sheaf-theoretic structure1036of non-locality and contextuality.1037\emph{New J. Phys.} 13 (2011), 113036.1038\href{https://doi.org/10.1088/1367-2630/13/11/113036}{DOI}.1039 1040\bibitem{amari-natural}1041S.-i. Amari. Natural gradient works efficiently in learning.1042\emph{Neural Computation} 10(2) (1998), 251--276.1043\href{https://doi.org/10.1162/089976698300017746}{DOI}.1044 1045\bibitem{topos-actions}1046Anonymous. \emph{Actions of Toposes on Toposes: