Learning Noise Transition Matrix from Only Noisy Labels via Total Variation Regularization

TL;DR

This paper introduces a theoretically grounded method to estimate noise transition matrices directly from noisy labels using total variation regularization, improving robustness in weakly supervised classification.

Contribution

It proposes a novel approach that simultaneously estimates the noise transition matrix and trains classifiers without relying on unreliable class-posterior estimates.

Findings

Effective estimation of transition matrices demonstrated on benchmark datasets.

Method achieves consistent estimation under mild assumptions.

Improves classification robustness with noisy labels.

Abstract

Many weakly supervised classification methods employ a noise transition matrix to capture the class-conditional label corruption. To estimate the transition matrix from noisy data, existing methods often need to estimate the noisy class-posterior, which could be unreliable due to the overconfidence of neural networks. In this work, we propose a theoretically grounded method that can estimate the noise transition matrix and learn a classifier simultaneously, without relying on the error-prone noisy class-posterior estimation. Concretely, inspired by the characteristics of the stochastic label corruption process, we propose total variation regularization, which encourages the predicted probabilities to be more distinguishable from each other. Under mild assumptions, the proposed method yields a consistent estimator of the transition matrix. We show the effectiveness of the proposed method through experiments on benchmark and real-world datasets.