16
12

Minimum Width of Leaky-ReLU Neural Networks for Uniform Universal Approximation

Abstract

The study of universal approximation properties (UAP) for neural networks (NN) has a long history. When the network width is unlimited, only a single hidden layer is sufficient for UAP. In contrast, when the depth is unlimited, the width for UAP needs to be not less than the critical width wmin=max(dx,dy)w^*_{\min}=\max(d_x,d_y), where dxd_x and dyd_y are the dimensions of the input and output, respectively. Recently, \cite{cai2022achieve} shows that a leaky-ReLU NN with this critical width can achieve UAP for LpL^p functions on a compact domain K{K}, \emph{i.e.,} the UAP for Lp(K,Rdy)L^p({K},\mathbb{R}^{d_y}). This paper examines a uniform UAP for the function class C(K,Rdy)C({K},\mathbb{R}^{d_y}) and gives the exact minimum width of the leaky-ReLU NN as wmin=max(dx,dy)+Δ(dx,dy)w_{\min}=\max(d_x,d_y)+\Delta (d_x, d_y), where Δ(dx,dy)\Delta (d_x, d_y) is the additional dimensions for approximating continuous functions with diffeomorphisms via embedding. To obtain this result, we propose a novel lift-flow-discretization approach that shows that the uniform UAP has a deep connection with topological theory.

View on arXiv
Comments on this paper