Indexed text attachment
differential-topology-of-parametrized-neural-networks.tex
Source for Differential Topology of Parametrized Neural Networks
View paper · Download original file
Lines 921–1020 of 1074. Source text is displayed literally.
921For \(m\ge n+1\), exact ambient recovery is possible.922If the decoder is constrained to take values in \(S^n\)923on all of \(\mathbb R^m\), the uniform error is \(2\)924for every width \(m\).925\end{proposition}926 927\begin{proof}928For \(m\le n\), apply Borsuk--Ulam to the composite of the929encoder and the coordinate inclusion of \(\mathbb R^m\) in930\(\mathbb R^n\). It gives an \(x\) with \(e(x)=e(-x)\)931\cite{matousek-borsuk}.932If \(u\) is their common decoded value,933\(2\le\|x-u\|+\|-x-u\|\). The zero decoder attains error \(1\).934For \(m\ge n+1\), a coordinate inclusion and projection give935exact recovery. Finally a sphere-valued reconstruction through936all of \(\mathbb R^m\) is nullhomotopic. If it avoided937antipodal coincidences, its straight segments to the identity,938normalized to unit length, would give a homotopy to the identity.939This would contradict its zero degree. Hence it reaches error940\(2\), the largest distance between points of the sphere.941\end{proof}942 943The optimal errors for the unit sphere are summarized below.944\begin{center}945\begin{tabular}{@{}lcc@{}}946\hline947Decoder range & \(m\le n\) & \(m\ge n+1\)\\948\hline949Ambient \(\mathbb R^{n+1}\) & \(1\) & \(0\)\\950Sphere \(S^n\) & \(2\) & \(2\)\\951\hline952\end{tabular}953\end{center}954For a sphere of radius \(R>0\), rescaling replaces the positive955entries by \(R\) and \(2R\), respectively. The bound holds for every956parameter of an appropriate family. Its proof chooses an957antipodal pair separately at each parameter and needs no958continuous choice of witnesses.959 960Uniform and average losses have different implications. For example,961at width \(n\) choose stereographic coordinates \(\sigma\) from the962complement of the north pole of \(S^n\) to \(\mathbb R^n\).963Multiply \(\sigma\) by a smooth cutoff equal to one outside a small964polar cap \(C\) and equal to zero near the pole, extending the965encoder by zero there. The smooth decoder \(d=\sigma^{-1}\) gives966exact recovery off \(C\). For normalized spherical measure \(\mu\),967the squared error averaged over the sphere is at most \(4\mu(C)\),968which can be made arbitrarily small while the uniform error is \(2\).969Thus the uniform obstruction does not force a positive lower bound970on this average across all smooth maps of width \(n\).971 972A uniform Lipschitz bound prevents this concentration. If the973reconstruction has Lipschitz constant at most \(L\) with respect to974chord distance, its error function has Lipschitz constant at most975\(1+L\). For ambient width \(m\le n\), some point has error at least976\(1\), so every point within chord distance \(1/(2(1+L))\) of it977has error at least \(1/2\). Its mean squared error is at least978one quarter of the normalized measure of such a cap, a positive979bound at fixed \(L\). This estimate states the additional analytic980hypothesis under which the pointwise obstruction controls981a population loss.982\par983\section{Conclusion}984 985The analysis separates three conditions on a parametrized network.986The available weight directions determine where the total jet987map is transverse. The symmetry representation determines which988normal derivatives a fixed realization can have at a stabilizer.989The topology of the domain and latent factorization determines990which global reconstructions are possible. A model can satisfy991one of these conditions while exhibiting an obstruction governed992by another.993 994Examples on the circle and torus make these distinctions explicit.995Ordinary point counts jump at the fold discriminant. They can996also remain constant while boundary sheets are permuted along a997regular parameter loop. The resulting persistence module over998the parameter retains that permutation representation. In the999equivariant torus model the squared residual controls normal1000movement, while the tangent complex records an excess target1001direction. The autoencoder estimates bound uniform reconstruction1002error and give a separate mean-square bound under a Lipschitz1003constraint.1004 1005Relative evaluation and descent explain how these calculations1006pass between local presentations of a model. The assumptions1007used for the calculations are geometric: smoothness of the1008relevant charts, specified perturbation directions, properness1009for topology of fibers, and an explicit choice of decoder range.1010They apply to a realized architecture once those hypotheses1011have been checked.1012\par1013\balance1014\begin{thebibliography}{99}1015\small\raggedright1016 1017\bibitem{higher-hirsch}1018Anonymous. \emph{Differential Topology in Higher Categories:1019From Regular Values and Cobordism to Derived Smooth Families}.1020Submitted manuscript, 2026.