Lower-Bounded Proper Losses for Weakly Supervised Classification

              Shuhei M. Yoshida 1 2 Takashi Takenouchi 2 3 Masashi Sugiyama 2 4

             Abstract                  an instance with a label that may be corrupted. Positive-
                                   unlabeled (PU) learning of binary classification uses positive
   This paper discusses the problem of weakly su-
                                   and unlabeled data, but not labeled negative data (Elkan &
   pervised cl ...