Neural networks with superexpressive activations and integer weights

Abstract

An example of an activation function is given such that networks with activations \, \, integer weights and a fixed architecture depending on d approximate continuous functions on [0,1]^d. The range of integer weights required for -approximation of H\"older continuous functions is derived, which leads to a convergence rate of order n^-22+d_2n for neural network regression estimation of unknown -H\"older continuous function with given n samples.

Tasks

Reproductions