{"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","summary":"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,…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-01-29","reactions":{}}