Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1

TL;DR

This paper presents a method for training Binarized Neural Networks with binary weights and activations, significantly reducing memory and computation requirements while maintaining high accuracy on standard datasets.

Contribution

The authors introduce a training approach for Binarized Neural Networks that enables efficient inference and training, with practical GPU implementation and competitive results.

Findings

BNNs achieve near state-of-the-art accuracy on MNIST, CIFAR-10, SVHN

Binary matrix multiplication GPU kernel speeds up inference by 7x

BNNs drastically reduce memory and power consumption

Abstract

We introduce a method to train Binarized Neural Networks (BNNs) - neural networks with binary weights and activations at run-time. At training-time the binary weights and activations are used for computing the parameters gradients. During the forward pass, BNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations, which is expected to substantially improve power-efficiency. To validate the effectiveness of BNNs we conduct two sets of experiments on the Torch7 and Theano frameworks. On both, BNNs achieved nearly state-of-the-art results over the MNIST, CIFAR-10 and SVHN datasets. Last but not least, we wrote a binary matrix multiplication GPU kernel with which it is possible to run our MNIST BNN 7 times faster than with an unoptimized GPU kernel, without suffering any loss in classification accuracy. The code for training and running our BNNs is available on-line.