DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
This work addresses the need for efficient neural network training and inference on hardware like CPU, FPGA, ASIC, and GPU, though it is incremental as it builds on existing low-bitwidth quantization methods.
The authors tackled the problem of training convolutional neural networks with low bitwidth weights, activations, and gradients to enable efficient hardware acceleration, achieving comparable accuracy to 32-bit models, such as 46.1% top-1 accuracy on ImageNet with a 1-bit weight, 2-bit activation AlexNet trained using 6-bit gradients.
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.