IEEE

Forward and Backward Information Retention for Accurate Binary Neural Networks

Abstract: Weight and activation binarization is an effective approach to deep neural network compression and can accelerate the inference by leveraging bitwise ...