Training Behavior of Sparse Neural Network Topologies

TL;DR

This paper investigates the training behavior of sparse neural network topologies, demonstrating that they can achieve comparable accuracy to dense networks but face stability issues at high sparsity levels.

Contribution

It provides experimental analysis of sparse neural network topologies, including pruning-based and RadiX-Nets, highlighting their potential and challenges in training.

Findings

Sparse networks can match dense network accuracy.

Extreme sparsity causes training instability.

RadiX-Nets have proven connectivity and sparsity properties.

Abstract

Improvements in the performance of deep neural networks have often come through the design of larger and more complex networks. As a result, fast memory is a significant limiting factor in our ability to improve network performance. One approach to overcoming this limit is the design of sparse neural networks, which can be both very large and efficiently trained. In this paper we experiment training on sparse neural network topologies. We test pruning-based topologies, which are derived from an initially dense network whose connections are pruned, as well as RadiX-Nets, a class of network topologies with proven connectivity and sparsity properties. Results show that sparse networks obtain accuracies comparable to dense networks, but extreme levels of sparsity cause instability in training, which merits further study.