High Performance Binarized Neural Networks trained on the ImageNet Classification Task

We trained Binarized Neural Networks (BNNs) on the high resolution ImageNet LSVRC-2102 dataset classification task and achieved a good performance. With a moderate size network of 10 layers, we obtained top-5 classification accuracy rate of 81 percent on validation set which is much better than previous published results. We expect training networks of a much better performance through increase network depth would be straight forward by following our current strategies. A detailed discussion on strategies used in the network training is included as well as preliminary analysis.