Understanding Instance-Level Label Noise: Disparate Impacts and Treatments

This paper aims to provide understandings for the effect of an over-parameterized model, e.g. a deep neural network, memorizing instance-dependent noisy labels. We first quantify the harms caused by memorizing noisy instances from different spectra of the sample distribution. We then analyze how several popular solutions for learning with noisy labels mitigate this harm at the instance-level. Our analysis reveals new understandings for when these approaches work, and provides theoretical justifications for previously reported empirical observations. A key aspect of the analysis is its focus on each training instance.