DNQ: Dynamic Network Quantization

2018/12/06 by Xu, Yuhui, Zhang, Shuai, Qi, Yingyong +3 · 2 citations
#Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.1812.02375

Abstract

Network quantization is an effective method for the deployment of neural networks on memory and energy constrained mobile devices. In this paper, we propose a Dynamic Network Quantization (DNQ) framework which is composed of two modules: a bit-width controller and a quantizer. Unlike most existing quantization methods that use a universal quantization bit-width for the whole network, we utilize policy gradient to train an agent to learn the bit-width of each layer by the bit-width controller. This controller can make a trade-off between accuracy and compression ratio. Given the quantization bit-width sequence, the quantizer adopts the quantization distance as the criterion of the weights importance during quantization. We extensively validate the proposed approach on various main-stream neural networks and obtain impressive results.

Cited by

• Assessing the Potential for Catastrophic Failure in Dynamic Post-Training Quantization
• DQT: Dynamic Quantization Training via Dequantization-Free Nested Integer Arithmetic

Related

• Differentiable Soft Quantization: Bridging Full-Precision and Low-Bit Neural Networks
• A White Paper on Neural Network Quantization
• ReLeQ: A Reinforcement Learning Approach for Deep Quantization of Neural Networks
• LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks
• Deep Neural Network Compression with Single and Multiple Level Quantization