Some Approximation Bounds for Deep Networks

TL;DR

This paper presents new theoretical bounds on how well deep networks can approximate functions, introduces novel architectures, and provides insights into the effectiveness of autoencoders and ResNets.

Contribution

It offers new approximation bounds and architectures for deep networks, enhancing understanding of their function approximation capabilities.

Findings

New bounds on deep network function approximation

Introduction of novel deep network architectures

Theoretical insights into autoencoders and ResNets

Abstract

In this paper we introduce new bounds on the approximation of functions in deep networks and in doing so introduce some new deep network architectures for function approximation. These results give some theoretical insight into the success of autoencoders and ResNets.