Best Transition Matrix Esitimation or Best Label Noise Robustness Classifier? Two Possible Methods to Enhance the Performance of T-revision

TL;DR

This paper compares methods for estimating label noise transition matrices and constructing robust classifiers, introducing improved T-Revision techniques and evaluating them on FashionMNIST and CIFAR-10 datasets.

Contribution

It proposes enhanced T-Revision methods for better noise matrix estimation and robustness, and compares these with existing approaches on standard datasets.

Findings

T-Revision-Alpha and T-Revision-Softmax improve stability and robustness.

Methods effectively estimate noise matrices and improve label prediction accuracy.

Performance varies depending on known or unknown transition matrices.

Abstract

Label noise refers to incorrect labels in a dataset caused by human errors or collection defects, which is common in real-world applications and can significantly reduce the accuracy of models. This report explores how to estimate noise transition matrices and construct deep learning classifiers that are robust against label noise. In cases where the transition matrix is known, we apply forward correction and importance reweighting methods to correct the impact of label noise using the transition matrix. When the transition matrix is unknown or inaccurate, we use the anchor point assumption and T-Revision series methods to estimate or correct the noise matrix. In this study, we further improved the T-Revision method by developing T-Revision-Alpha and T-Revision-Softmax to enhance stability and robustness. Additionally, we designed and implemented two baseline classifiers, a Multi-Layer Perceptron (MLP) and ResNet-18, based on the cross-entropy loss function. We compared the performance of these methods on predicting clean labels and estimating transition matrices using the FashionMINIST dataset with known noise transition matrices. For the CIFAR-10 dataset, where the noise transition matrix is unknown, we estimated the noise matrix and evaluated the ability of the methods to predict clean labels.