Symmetric Cross Entropy for Robust Learning with Noisy Labels

TL;DR

This paper introduces Symmetric Cross Entropy (SL), a novel loss function combining Cross Entropy and Reverse Cross Entropy, to improve deep neural network training robustness against noisy labels by addressing overfitting and under learning issues.

Contribution

The paper proposes Symmetric Cross Entropy, a new loss function that enhances robustness to noisy labels and can be integrated with existing methods, backed by theoretical and empirical validation.

Findings

SL outperforms state-of-the-art methods on benchmark datasets.

SL effectively mitigates overfitting to noisy labels.

SL improves learning on hard classes with noisy labels.

Abstract

Training accurate deep neural networks (DNNs) in the presence of noisy labels is an important and challenging task. Though a number of approaches have been proposed for learning with noisy labels, many open issues remain. In this paper, we show that DNN learning with Cross Entropy (CE) exhibits overfitting to noisy labels on some classes ("easy" classes), but more surprisingly, it also suffers from significant under learning on some other classes ("hard" classes). Intuitively, CE requires an extra term to facilitate learning of hard classes, and more importantly, this term should be noise tolerant, so as to avoid overfitting to noisy labels. Inspired by the symmetric KL-divergence, we propose the approach of \textbf{Symmetric cross entropy Learning} (SL), boosting CE symmetrically with a noise robust counterpart Reverse Cross Entropy (RCE). Our proposed SL approach simultaneously addresses both the under learning and overfitting problem of CE in the presence of noisy labels. We provide a theoretical analysis of SL and also empirically show, on a range of benchmark and real-world datasets, that SL outperforms state-of-the-art methods. We also show that SL can be easily incorporated into existing methods in order to further enhance their performance.