DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients

TL;DR

DoReFa-Net introduces a method to train low bitwidth convolutional neural networks using stochastic quantization of gradients, enabling efficient low bitwidth training and inference across various hardware platforms.

Contribution

The paper presents a novel approach to train low bitwidth CNNs with low bitwidth gradients, allowing for accelerated training and inference on diverse hardware.

Findings

Achieves comparable accuracy to 32-bit models on SVHN and ImageNet.

Enables efficient low bitwidth convolution operations on CPU, FPGA, ASIC, and GPU.

Demonstrates training of AlexNet with 1-bit weights and 2-bit activations using 6-bit gradients.

Abstract

We propose DoReFa-Net, a method to train convolutional neural networks that have low bitwidth weights and activations using low bitwidth parameter gradients. In particular, during backward pass, parameter gradients are stochastically quantized to low bitwidth numbers before being propagated to convolutional layers. As convolutions during forward/backward passes can now operate on low bitwidth weights and activations/gradients respectively, DoReFa-Net can use bit convolution kernels to accelerate both training and inference. Moreover, as bit convolutions can be efficiently implemented on CPU, FPGA, ASIC and GPU, DoReFa-Net opens the way to accelerate training of low bitwidth neural network on these hardware. Our experiments on SVHN and ImageNet datasets prove that DoReFa-Net can achieve comparable prediction accuracy as 32-bit counterparts. For example, a DoReFa-Net derived from AlexNet that has 1-bit weights, 2-bit activations, can be trained from scratch using 6-bit gradients to get 46.1\% top-1 accuracy on ImageNet validation set. The DoReFa-Net AlexNet model is released publicly.