Id: arxiv/2601.21539 Url: https://arxiv.org/abs/2601.21539 Tags: math.PR Score: 0 Title: Wide neural networks with general weights: convergence rate and explicit dependence on the hyper-parameters Domain: math.PR Source: arxiv Authors: Lucia Celli Reviews: 0 Subject: math.PR Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-01-29 Page: https://delphi.science/r/arxiv/2601.21539 Using Stein's method techniques introduced by Chatterjee (2008) and further extended by Kasprzak and Peccati (2022) and by Lachièze-Rey and Peccati (2017), we derive novel quantitative bounds on the convergence in distribution of feed-forward fully connected neural networks (with Lipschitz activation functions) towards Gaussian processes, as the hidden layer width $n$ tends to infinity. We consider networks initialized with independent and identically distributed (i.i.d.) weights possessing sufficiently many finite moments, and i.i.d. Gaussian biases independent of the weights. Specifically,…