8493eeaccb772c0878f99d60a0bd2bb3-AuthorFeedback.pdf

We thank all the reviewers for carefully checking the paper and acknowledging the "efficiency and practicality" of We will also clarify in the revised version. R1 asks for discussion of similarity and difference of technical results to [19]. Hence, training on the medoids is robust to noisy labels. Indeed, Eq. 5 finds the best subset of Fraction of clean data points in coreset. R2 asks how clean the coreset is.