DNQ: Dynamic Network Quantization

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.

References in corpus (9)

• Neural Architecture Search with Reinforcement Learning
• Ternary Weight Networks
• Trained Ternary Quantization
• Efficient Neural Architecture Search via Parameter Sharing
• Compressing Neural Networks with the Hashing Trick
• Incremental Network Quantization: Towards Lossless CNNs with Low-Precision Weights
• Hierarchical Representations for Efficient Architecture Search
• Deep Neural Network Compression with Single and Multiple Level Quantization
• BinaryRelax: A Relaxation Approach For Training Deep Neural Networks With Quantized Weights

Cited by in corpus (1)

• CPT: Efficient Deep Neural Network Training via Cyclic Precision