Towards Optimal Structured CNN Pruning via Generative Adversarial Learning论文笔记