Simple and Effective Complementary Label Learning Based on Mean Square Error Loss

Abstract In this paper, we propose a simple and effective complementary label learning approach to address the noise problem for deep model. Different surrogate losses have been proposed learning, however, are often sophisticated designed, as required satisfy classifier consistency property. We an square loss under unbiased biased assumptions. also show theoretically that our method assurances optimal labels is ordinary labels. Finally, test on three different benchmark datasets with assumptions verify effectiveness of method.