Confidence Scores Make Instance-dependent Label-noise Learning Possible

TL;DR

This paper introduces a novel approach to learning with label noise by using confidence scores to estimate instance-dependent noise transition distributions, enabling effective correction methods in realistic noisy label scenarios.

Contribution

It proposes confidence-scored instance-dependent noise (CSIDN) and an instance-level forward correction method, addressing limitations of previous models relying on strong assumptions.

Findings

Effective in synthetic label noise scenarios

Works well with real-world unknown noise

Improves label noise robustness

Abstract

In learning with noisy labels, for every instance, its label can randomly walk to other classes following a transition distribution which is named a noise model. Well-studied noise models are all instance-independent, namely, the transition depends only on the original label but not the instance itself, and thus they are less practical in the wild. Fortunately, methods based on instance-dependent noise have been studied, but most of them have to rely on strong assumptions on the noise models. To alleviate this issue, we introduce confidence-scored instance-dependent noise (CSIDN), where each instance-label pair is equipped with a confidence score. We find with the help of confidence scores, the transition distribution of each instance can be approximately estimated. Similarly to the powerful forward correction for instance-independent noise, we propose a novel instance-level forward correction for CSIDN. We demonstrate the utility and effectiveness of our method through multiple experiments under synthetic label noise and real-world unknown noise.