Confidence Scores Make Instance-dependent Label-noise Learning Possible

         Antonin Berthon 1 2 Bo Han 3 1 Gang Niu 1 Tongliang Liu 4 Masashi Sugiyama 1 5

             Abstract               orization effects (Zhang et al., 2017). Thus, learning with
                                noisy labels has often drawn a lot of attention.
   In learning with noisy labels, for every instance,
   its label can randomly walk to other classes fol-     Early works on noisy labels studied random classif ...