DP-Net: Dynamic Programming Guided Deep Neural Network Compression
This work addresses the problem of reducing model size and computational cost for deploying DNNs in resource-constrained environments, representing a strong specific gain in compression efficiency.
The paper tackled deep neural network compression by proposing DP-Net, which uses dynamic programming for optimal weight quantization and trains clustering-friendly models, achieving up to 77X compression on Wide ResNet while preserving accuracy.
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.