Class2Simi: A Noise Reduction Perspective on Learning with Noisy Labels

TL;DR

Class2Simi introduces a pairwise learning framework that transforms noisy class labels into similarity labels, reducing noise impact and enabling effective training of neural networks from noisy data.

Contribution

It proposes a novel pairwise transformation approach that theoretically reduces label noise and improves learning with noisy labels, with efficient on-the-fly implementation.

Findings

The framework guarantees noise reduction in labels.

Neural networks trained with Class2Simi achieve better robustness.

Extensive experiments validate the effectiveness of the approach.

Abstract

Learning with noisy labels has attracted a lot of attention in recent years, where the mainstream approaches are in pointwise manners. Meanwhile, pairwise manners have shown great potential in supervised metric learning and unsupervised contrastive learning. Thus, a natural question is raised: does learning in a pairwise manner mitigate label noise? To give an affirmative answer, in this paper, we propose a framework called Class2Simi: it transforms data points with noisy class labels to data pairs with noisy similarity labels, where a similarity label denotes whether a pair shares the class label or not. Through this transformation, the reduction of the noise rate is theoretically guaranteed, and hence it is in principle easier to handle noisy similarity labels. Amazingly, DNNs that predict the clean class labels can be trained from noisy data pairs if they are first pretrained from noisy data points. Class2Simi is computationally efficient because not only this transformation is on-the-fly in mini-batches, but also it just changes loss computation on top of model prediction into a pairwise manner. Its effectiveness is verified by extensive experiments.