31
229

Approximating Continuous Functions by ReLU Nets of Minimal Width

Abstract

This article concerns the expressive power of depth in deep feed-forward neural nets with ReLU activations. Specifically, we answer the following question: for a fixed din1,d_{in}\geq 1, what is the minimal width ww so that neural nets with ReLU activations, input dimension dind_{in}, hidden layer widths at most w,w, and arbitrary depth can approximate any continuous, real-valued function of dind_{in} variables arbitrarily well? It turns out that this minimal width is exactly equal to din+1.d_{in}+1. That is, if all the hidden layer widths are bounded by dind_{in}, then even in the infinite depth limit, ReLU nets can only express a very limited class of functions, and, on the other hand, any continuous function on the dind_{in}-dimensional unit cube can be approximated to arbitrary precision by ReLU nets in which all hidden layers have width exactly din+1.d_{in}+1. Our construction in fact shows that any continuous function f:[0,1]dinRdoutf:[0,1]^{d_{in}}\to\mathbb R^{d_{out}} can be approximated by a net of width din+doutd_{in}+d_{out}. We obtain quantitative depth estimates for such an approximation in terms of the modulus of continuity of ff.

View on arXiv
Comments on this paper