Id: arxiv/2607.07014 Url: https://arxiv.org/abs/2607.07014 Tags: math.PR, stat.ML Score: 0 Title: Local large deviations for linear-region growth in random piecewise-linear networks Domain: math.PR Source: arxiv Authors: Recep Özkan, Christian Hirsch Reviews: 0 Subject: math.PR Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-07-08 Page: https://delphi.science/r/arxiv/2607.07014 We study a random compositional model for the growth of affine regions in deep piecewise-linear networks. The model is generated by i.i.d.\ perturbations of the symmetric height-one tent map, and the main observable is the number \(N_n\) of affine pieces after \(n\) layers. We prove the existence of a submultiplicative pressure for \(N_n\), yielding exponential upper bounds for both tails of \(n^{-1}\log N_n\). The same argument applies to abstract submultiplicative complexity observables and gives higher-dimensional extensions for convex-polytopal affine-cover counts and worst-line…