On Symmetric Losses for Learning from Corrupted Labels
This work addresses label noise in classification tasks, offering theoretical insights and a practical method, but it is incremental as it builds on existing symmetric loss concepts.
The paper tackles the problem of learning from corrupted labels by analyzing symmetric losses, showing they are advantageous for balanced error rate minimization and AUC maximization, and proposes a convex barrier hinge loss that benefits from symmetry, with experimental validation.
This paper aims to provide a better understanding of a symmetric loss. First, we emphasize that using a symmetric loss is advantageous in the balanced error rate (BER) minimization and area under the receiver operating characteristic curve (AUC) maximization from corrupted labels. Second, we prove general theoretical properties of symmetric losses, including a classification-calibration condition, excess risk bound, conditional risk minimizer, and AUC-consistency condition. Third, since all nonnegative symmetric losses are non-convex, we propose a convex barrier hinge loss that benefits significantly from the symmetric condition, although it is not symmetric everywhere. Finally, we conduct experiments to validate the relevance of the symmetric condition.