Clusterability as an Alternative to Anchor Points
                 When Learning with Noisy Labels

                     Zhaowei Zhu 1 Yiwen Song 2 Yang Liu 1

             Abstract                 parameterized DNNs (Xia et al., 2021; Han et al., 2020),
                                  and lead to unexpected and disparate impacts (Liu, 2021).
    The label noise transition matrix, characteriz-
    ing the probabilities of a training instance being     A variety of approaches were proposed to a ...