Compressing Low Precision Deep Neural Networks Using Sparsity-Induced Regularization in Ternary Networks

TL;DR

This paper introduces a training method for sparse ternary neural networks that reduces model size and computational complexity by incorporating hardware-aware regularization, achieving high sparsity and efficiency on standard datasets.

Contribution

The authors propose a novel training approach that combines L2 regularization and quantization threshold regularizer to produce highly sparse ternary networks with improved accuracy and hardware efficiency.

Findings

Networks are up to 98% sparse.

Models are 5 to 11 times smaller than binary/ternary counterparts.

Significant resource and speed benefits demonstrated.

Abstract

A low precision deep neural network training technique for producing sparse, ternary neural networks is presented. The technique incorporates hard- ware implementation costs during training to achieve significant model compression for inference. Training involves three stages: network training using L2 regularization and a quantization threshold regularizer, quantization pruning, and finally retraining. Resulting networks achieve improved accuracy, reduced memory footprint and reduced computational complexity compared with conventional methods, on MNIST and CIFAR10 datasets. Our networks are up to 98% sparse and 5 & 11 times smaller than equivalent binary and ternary models, translating to significant resource and speed benefits for hardware implementations.