Learning from Similarity/Dissimilarity and Pairwise Comparison
This addresses a practical problem for machine learning practitioners in domains where labeling is costly or subjective, though it is an incremental improvement over existing weak supervision frameworks.
The paper tackles binary classification when explicit labels are unavailable by using weak labels from instance pairs, proposing a framework based on similarity/dissimilarity and pairwise comparison judgments. It shows improved performance over single-label methods, with robustness to label noise and uncertainty in class prior estimation.
This paper addresses binary classification in scenarios where obtaining explicit instance level labels is impractical, by exploiting multiple weak labels defined on instance pairs. The existing SconfConfDiff classification framework relies on continuous valued probabilistic supervision, including similarity-confidence, the probability of class agreement, and confidence-difference, the difference in positive class probabilities. However, probabilistic labeling requires subjective uncertainty quantification, often leading to unstable supervision. We propose SD-Pcomp classification, a binary judgment based weakly supervised learning framework that relies only on relative judgments, namely class agreement between two instances and pairwise preference toward the positive class. The method employs Similarity/Dissimilarity (SD) labels and Pairwise Comparison (Pcomp) labels, and develops two unbiased risk estimators, (i) a convex combination of SD and Pcomp and (ii) a unified estimator that integrates both labels by modeling their relationship. Theoretical analysis and experimental results show that the proposed approach improves classification performance over methods using a single weak label, and is robust to label noise and uncertainty in class prior estimation.