DP-Net: Dynamic Programming Guided Deep Neural Network Compression

TL;DR

DP-Net introduces a dynamic programming-based approach for neural network compression, achieving higher compression ratios while maintaining accuracy, and includes hardware acceleration for efficient inference.

Contribution

It presents a novel DP-based algorithm for weight quantization and a training method for clustering-friendly DNNs, enabling superior compression.

Findings

77X compression on Wide ResNet achieved

Outperforms state-of-the-art compression methods

Hardware acceleration on FPGA demonstrated

Abstract

In this work, we propose an effective scheme (called DP-Net) for compressing the deep neural networks (DNNs). It includes a novel dynamic programming (DP) based algorithm to obtain the optimal solution of weight quantization and an optimization process to train a clustering-friendly DNN. Experiments showed that the DP-Net allows larger compression than the state-of-the-art counterparts while preserving accuracy. The largest 77X compression ratio on Wide ResNet is achieved by combining DP-Net with other compression techniques. Furthermore, the DP-Net is extended for compressing a robust DNN model with negligible accuracy loss. At last, a custom accelerator is designed on FPGA to speed up the inference computation with DP-Net.