Zero and Narrow-Width Value-Aware Compression for Quantized Convolutional Neural Networks | IEEE Journals & Magazine | IEEE Xplore