EQ-Net: Elastic Quantization Neural Networks

TL;DR

EQ-Net introduces a flexible, one-shot neural network quantization approach that supports various hardware-specific quantization forms through an elastic quantization space, regularization techniques, and efficient search methods.

Contribution

The paper proposes a novel elastic quantization space and training framework for flexible, hardware-adaptive neural network quantization, along with new regularization losses and an efficient search algorithm.

Findings

EQ-Net achieves comparable or better performance than static and state-of-the-art methods.

The elastic quantization space effectively adapts to different hardware requirements.

The proposed search method efficiently finds optimal mixed-precision quantization configurations.

Abstract

Current model quantization methods have shown their promising capability in reducing storage space and computation complexity. However, due to the diversity of quantization forms supported by different hardware, one limitation of existing solutions is that usually require repeated optimization for different scenarios. How to construct a model with flexible quantization forms has been less studied. In this paper, we explore a one-shot network quantization regime, named Elastic Quantization Neural Networks (EQ-Net), which aims to train a robust weight-sharing quantization supernet. First of all, we propose an elastic quantization space (including elastic bit-width, granularity, and symmetry) to adapt to various mainstream quantitative forms. Secondly, we propose the Weight Distribution Regularization Loss (WDR-Loss) and Group Progressive Guidance Loss (GPG-Loss) to bridge the inconsistency of the distribution for weights and output logits in the elastic quantization space gap. Lastly, we incorporate genetic algorithms and the proposed Conditional Quantization-Aware Accuracy Predictor (CQAP) as an estimator to quickly search mixed-precision quantized neural networks in supernet. Extensive experiments demonstrate that our EQ-Net is close to or even better than its static counterparts as well as state-of-the-art robust bit-width methods. Code can be available at \href{https://github.com/xuke225/EQ-Net.git}{https://github.com/xuke225/EQ-Net}.