comments point by point

We sincerely thank all the reviewers for their careful readings and valuable comments. We will address the reviewers' Aug used 8 times the mini-batch size. The authors should add this comparison. 's advice on the mini-batch size, we tried to increase the batch size, but we can And we've achieved better performance with R#3 Q1: In figure 1, the change of rank cannot indicate that the model is unstable ... We will change into: ... the compromised evaluation process would distort the ranking for Also, the std is very large. On CIFAR10, the performance gains of our A WS are substantial.