Neural Collapse Inspired Attraction-Repulsion-Balanced Loss for Imbalanced Learning
This addresses the issue of class imbalance in real-world engineering applications, offering an incremental improvement by simplifying training to one stage while maintaining SOTA performance.
The paper tackles the problem of class imbalance in deep learning, which harms classification precision, especially for minor classes, by proposing the Attraction-Repulsion-Balanced Loss (ARB-Loss) to balance gradient components, achieving state-of-the-art performance on large-scale classification and segmentation datasets with one-stage training instead of the two-stage methods used in current SOTA works.
Class imbalance distribution widely exists in real-world engineering. However, the mainstream optimization algorithms that seek to minimize error will trap the deep learning model in sub-optimums when facing extreme class imbalance. It seriously harms the classification precision, especially on the minor classes. The essential reason is that the gradients of the classifier weights are imbalanced among the components from different classes. In this paper, we propose Attraction-Repulsion-Balanced Loss (ARB-Loss) to balance the different components of the gradients. We perform experiments on the large-scale classification and segmentation datasets and our ARB-Loss can achieve state-of-the-art performance via only one-stage training instead of 2-stage learning like nowadays SOTA works.