Binarized Neural Networks: Training Neural Networks with Weights and Activations Constrained to +1 or =1 read more
Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding read more