What Do We Understand About Convolutional Networks?

TL;DR

This paper reviews the design, understanding, and empirical analysis of convolutional networks, focusing on their biological inspiration, theoretical foundations, and visualization studies to clarify their layered processing roles.

Contribution

It provides a comprehensive review of convolutional network components, their biological and theoretical bases, and summarizes current understanding and open problems in the field.

Findings

Different approaches based on biological findings and theory inform ConvNet design.

Visualization and empirical studies reveal the roles of various layers in ConvNets.

Current understanding highlights critical open problems in ConvNet interpretability.

Abstract

This document will review the most prominent proposals using multilayer convolutional architectures. Importantly, the various components of a typical convolutional network will be discussed through a review of different approaches that base their design decisions on biological findings and/or sound theoretical bases. In addition, the different attempts at understanding ConvNets via visualizations and empirical studies will be reviewed. The ultimate goal is to shed light on the role of each layer of processing involved in a ConvNet architecture, distill what we currently understand about ConvNets and highlight critical open problems.