e038453073d221a4f32d0bab94ca7cee-AuthorFeedback.pdf

R1.1: concerns on compared methods and datasets. The results are shown in Figure i. We will add detailed discussions. We will highlight them in these two tables. In our implementation, we simply use the "negative" loss, i.e., R4.1: compare with robust deep learning.