{"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","summary":"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…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-07-08","reactions":{}}