Estimation of the Label-Noise Transition Matrix with Performance Guarantees via Selective Classification

2026-09-30

AI Summary

This study addresses the challenge in learning with noisy labels where transition matrix estimation relies on fragile class-posterior probabilities and lacks finite-sample guarantees. To overcome these limitations, this work proposes a novel transition matrix estimation method based on one-sided selective classification. By pioneering the use of selective classification to bypass class-posterior estimation, the proposed approach accommodates flexible binary classification models and incorporates an efficient optimization algorithm. The core contribution lies in achieving accurate transition matrix estimation without requiring class posteriors, while providing rigorous finite-sample performance guarantees and refined error bounds. Consequently, this method significantly enhances model robustness when trained on noisy data.

Abstract

Modern machine learning depends heavily on massive datasets, but obtaining high-quality annotations at scale is often expensive. As a result, learning from noisily-labeled data has become common, making accurate estimation of the label-noise transition matrix crucial. However, existing transition matrix estimators rely on the fragile estimation of class-posteriors and do not provide finite-sample performance guarantees. In this work, we propose a novel methodology to estimate the transition matrix based on one-sided selective classification. This approach bypasses class-posterior estimation, provides finite-sample performance guarantees, and leverages flexible learning methods for binary classification. Moreover, we introduce effective algorithms to implement the proposed methodology and provide their refined finite-sample performance bounds.

Problem

• label-noise transition matrix
• noisy labels
• finite-sample guarantees
• class-posterior estimation

Innovation

• Label-Noise Transition Matrix
• Selective Classification
• Finite-Sample Guarantees
• Learning with Noisy Labels