It is possible that ResNet-20 on the CIFAR10 in our experiment were undertrained

We appreciate the reviewers for their valuable comments on the improvement of this paper. ResNet-20 has been finalized within about 150 epochs, i.e. it encountered performance plateau after 150 epochs. As we haven't been over-tuning our method and under-tuning the In fact, we haven't put much more effort in tuning hyperparameters in our experiment The omission of this scheme in our paper is primarily due to our focus on replica-exchange protocol. No matter whether the RE criterion is Barker's test (as The latter is way slower than the former. T o Reviewer 4. Thank you very much for your kind support and endorsement.