Towards Efficient Training for Neural Network Quantization

TL;DR

This paper investigates the causes of accuracy loss in neural network quantization, identifies critical training rules, and introduces scale-adjusted training (SAT) to improve quantization performance, achieving state-of-the-art results.

Contribution

It proposes the SAT technique based on new insights into gradient propagation, enhancing training efficiency and accuracy in neural network quantization.

Findings

SAT improves quantization accuracy across models

Quantized models outperform full-precision counterparts in experiments

Analysis of gradient-calibrated PACT reduces quantization error

Abstract

Quantization reduces computation costs of neural networks but suffers from performance degeneration. Is this accuracy drop due to the reduced capacity, or inefficient training during the quantization procedure? After looking into the gradient propagation process of neural networks by viewing the weights and intermediate activations as random variables, we discover two critical rules for efficient training. Recent quantization approaches violates the two rules and results in degenerated convergence. To deal with this problem, we propose a simple yet effective technique, named scale-adjusted training (SAT), to comply with the discovered rules and facilitates efficient training. We also analyze the quantization error introduced in calculating the gradient in the popular parameterized clipping activation (PACT) technique. Through SAT together with gradient-calibrated PACT, quantized models obtain comparable or even better performance than their full-precision counterparts, achieving state-of-the-art accuracy with consistent improvement over previous quantization methods on a wide spectrum of models including MobileNet-V1/V2 and PreResNet-50.