Efficient Multi-Prize Lottery Tickets: Enhanced Accuracy, Training, and Inference Speed

TL;DR

This paper introduces improved methods for multi-prize lottery tickets in binary neural networks, enhancing accuracy and speed through pruning and quantization, with demonstrated benefits on CIFAR-10.

Contribution

The paper proposes techniques to stabilize accuracy and achieve training and inference speedups in multi-prize lottery tickets for binary neural networks.

Findings

Enhanced accuracy with optimized prune ratios

Achieved training and inference speed improvements

Validated on CIFAR-10 dataset

Abstract

Recently, Diffenderfer and Kailkhura proposed a new paradigm for learning compact yet highly accurate binary neural networks simply by pruning and quantizing randomly weighted full precision neural networks. However, the accuracy of these multi-prize tickets (MPTs) is highly sensitive to the optimal prune ratio, which limits their applicability. Furthermore, the original implementation did not attain any training or inference speed benefits. In this report, we discuss several improvements to overcome these limitations. We show the benefit of the proposed techniques by performing experiments on CIFAR-10.