用于深度模型压缩的可识别歧视的网络修剪 We study network pruning which aims to remove redundant channels/kernels and hence speed up the inference of deep networks. Existing pruning methods e