Label Noise: Ignorance Is Bliss
This work addresses the problem of robust learning with noisy labels for machine learning practitioners, offering a novel theoretical framework and practical method that is incremental in building on domain adaptation concepts.
The paper tackles learning under multi-class, instance-dependent label noise by framing it as domain adaptation and introducing a relative signal strength measure, leading to theoretical bounds that support a noise-ignorant empirical risk minimization approach, which achieves state-of-the-art performance on the CIFAR-N challenge.
We establish a new theoretical framework for learning under multi-class, instance-dependent label noise. This framework casts learning with label noise as a form of domain adaptation, in particular, domain adaptation under posterior drift. We introduce the concept of \emph{relative signal strength} (RSS), a pointwise measure that quantifies the transferability from noisy to clean posterior. Using RSS, we establish nearly matching upper and lower bounds on the excess risk. Our theoretical findings support the simple \emph{Noise Ignorant Empirical Risk Minimization (NI-ERM)} principle, which minimizes empirical risk while ignoring label noise. Finally, we translate this theoretical insight into practice: by using NI-ERM to fit a linear classifier on top of a self-supervised feature extractor, we achieve state-of-the-art performance on the CIFAR-N data challenge.