Do We Need to Penalize Variance of Losses for Learning with Label Noise?
This work addresses label noise in machine learning, which is a common issue in real-world data, but it appears incremental as it builds on existing algorithms by adding a variance-based regularizer.
The paper tackles the problem of learning with noisy labels by proposing to increase the variance of losses, which boosts memorization effects and reduces harm from incorrect labels, leading to significant improvements in generalization on synthetic and real-world datasets.
Algorithms which minimize the averaged loss have been widely designed for dealing with noisy labels. Intuitively, when there is a finite training sample, penalizing the variance of losses will improve the stability and generalization of the algorithms. Interestingly, we found that the variance should be increased for the problem of learning with noisy labels. Specifically, increasing the variance will boost the memorization effects and reduce the harmfulness of incorrect labels. By exploiting the label noise transition matrix, regularizers can be easily designed to reduce the variance of losses and be plugged in many existing algorithms. Empirically, the proposed method by increasing the variance of losses significantly improves the generalization ability of baselines on both synthetic and real-world datasets.