Pruning vs Quantization: Which is Better?

TL;DR

This paper compares neural network pruning and quantization techniques to determine which yields better compression and accuracy, providing theoretical bounds and extensive empirical results across multiple models and tasks.

Contribution

It offers the first comprehensive analytical and empirical comparison between pruning and quantization, guiding hardware design choices for neural network deployment.

Findings

Quantization generally outperforms pruning in most scenarios.

Pruning may be beneficial at very high compression ratios.

The paper provides theoretical bounds for pruning and quantization errors.

Abstract

Neural network pruning and quantization techniques are almost as old as neural networks themselves. However, to date only ad-hoc comparisons between the two have been published. In this paper, we set out to answer the question on which is better: neural network quantization or pruning? By answering this question, we hope to inform design decisions made on neural network hardware going forward. We provide an extensive comparison between the two techniques for compressing deep neural networks. First, we give an analytical comparison of expected quantization and pruning error for general data distributions. Then, we provide lower bounds for the per-layer pruning and quantization error in trained networks, and compare these to empirical error after optimization. Finally, we provide an extensive experimental comparison for training 8 large-scale models on 3 tasks. Our results show that in most cases quantization outperforms pruning. Only in some scenarios with very high compression ratio, pruning might be beneficial from an accuracy standpoint.