Learning from Similarity/Dissimilarity and Pairwise Comparison

TL;DR

This paper introduces SD-Pcomp, a weakly supervised binary classification framework that uses only relative similarity/dissimilarity and pairwise preference labels, improving robustness and performance over existing probabilistic methods.

Contribution

The paper proposes a novel SD-Pcomp framework that relies solely on relative judgments, developing unbiased risk estimators and demonstrating improved robustness and accuracy.

Findings

Outperforms existing methods using probabilistic supervision

Robust to label noise and class prior uncertainty

Provides theoretical guarantees and empirical validation

Abstract

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.