747d3443e319a22747fbb873e8b2f9f2-AuthorFeedback.pdf

We appreciate the reviewers for the constructive comments on this paper. C) How is the acquisition function optimized and what is the set of candidates for this optimization?31 Thesemethods relyonabiasedevaluation38 protocol like early stop or parameter sharing, while we can train each sampled network fully and get an unbiased39 evaluation. Actually,43 EfficientNet-B0/B1 is already a very strong network under the constraint of FLOPs, thus can verify our method.44