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 407–506 of 1207. Source text is displayed literally.

407 408A rank-one example is a triangle with two positive unit409transports and one negative unit transport. The product410around the cycle is $-1$, so the only real global section411is zero. By comparison, using three positive unit412transports gives the line of constant sections. The413underlying incidence graph is unchanged. This separates414the effect of the transport from that of connectivity.415When the incidence maps are isometries, each residual416has norm $\|x_v-T_e x_u\|$. In that case the operator is417a discrete connection Laplacian in the sense relevant418to vector diffusion maps \cite{singer-wu-vdm}. For orthogonal419changes of stalk coordinates, let $Q_0$ and $Q_1$ be the block420diagonal changes at vertices and edges, respectively. The new421incidence matrices are $r'_{ev}=Q_e r_{ev}Q_v^{-1}$, giving422$\delta'=Q_1\delta Q_0^{-1}$ and $L'=Q_0 L Q_0^{-1}$.423Thus the fixed diffusion is compatible with these changes of424frames even when the incidence maps are noninvertible.425 426The same argument constrains learnable linear parameters.427For two representations $V,W$ of the graph path428groupoid, a family of local linear maps429$A_v:V_v\to W_v$ is compatible with transport exactly430when431\[432  A_v T_e^V = T_e^W A_u,\qquad e:u\to v .433\]434Equivalently, $A$ is a parallel section of the bundle of435linear maps, with transport436$A\mapsto T_e^W A(T_e^V)^{-1}$. On a connected graph437its parameter space is therefore the space of438intertwiners of the two holonomy representations at439one vertex. For sign and trivial line holonomies440around the same cycle, a scalar parameter is forced441to vanish. For two copies of sign holonomy, any scalar442at the chosen vertex extends. Weight tying here is443dictated by representation type as well as by444the connectivity of the data. The invertibility445hypothesis matters: ordinary cellular sheaves also446permit noninvertible incidence maps and unequal stalk447dimensions, for which the incidence equation still448makes sense and there need not be transport along449groupoid arrows.450 451\subsection{From global groups to local bisections}452 453Global group actions determine the shared weights in454group-equivariant convolutions \cite{cohen-welling}.455Cohen, Geiger, and Weiler characterize equivariant456linear maps between fields on homogeneous spaces by457their convolution kernels \cite{cohen-geiger-weiler}.458On meshes, de Haan et al.\ use parallel transport to459define orientation-sensitive convolutions that respect460changes of local frame \cite{dehaan-mesh}.461 462For a finite data set carrying a group action, let463$K_{x,y}:V_y\to W_x$ be the blocks of a linear464layer and let $\rho_V(g;y)$ and $\rho_W(g;x)$ be the465feature transports. Equivariance is equivalent to466\[467 K_{gx,gy}468 =\rho_W(g;x)K_{x,y}\rho_V(g;y)^{-1}.469\]470It follows by comparing coefficients of a feature471supported at $y$. Each diagonal pair orbit has one472free block in the intertwiner space of its joint473stabilizer, since two transports from a representative474to the same pair differ by a stabilizer element.475Conversely an intertwiner at each representative476propagates to a well-defined equivariant kernel.477 478Ibort, Jimenez-Vazquez, and Perez-Pardo extend this type479of kernel analysis to groupoid representations with a480chosen pseudogroup of local bisections, measure, and481input and output bundles. Their preprint treats bounded482domains, where a geometric transformation can be483defined on part of the data space and pair types can484depend on the boundary \cite{ibort-groupoid-cnn}. The485choice of which bisections to act by and the integration486and support conditions enter the analytical layer487construction. In the groupoid descent picture, the488representation bundles supply transport data on the489features and on the kernel's source and target490variables. The kernel then acts on spaces of sections491and can mix distinct sites. A morphism of feature492representations over the identity base, such as $A$493above, acts fiberwise. This distinction separates useful494precedents for learning with groupoid descent from495studies of a monoidal Grothendieck topos acting on496another topos by geometric morphisms.497\par498\section{Compositional training and inference}499\label{sec:categorical-ml}500 501The categorical literature on training and probability supplies two502ingredients for learning in a setting with topos actions. Parametrized503maps provide a compositional description of local models, and504stochastic kernels provide a description of local inference.505To obtain a model over a groupoid of domains or in an action topos,506one supplies additional transport for the parameters, inputs, and