Training Compact Neural Networks with Binary Weights and Low Precision Activations | Read Paper on Bytez