Robust Long-Tailed Learning via Label-Aware Bounded CVaR

TL;DR

This paper introduces two novel CVaR-based loss functions, LAB-CVaR and LAB-CVaR-logit, with theoretical guarantees to improve long-tailed classification performance, especially on minority classes.

Contribution

The paper proposes theoretically grounded LAB-CVaR and LAB-CVaR-logit loss functions specifically designed for long-tailed learning, addressing limitations of previous methods.

Findings

LAB-CVaR outperforms existing methods on real-world long-tailed datasets.

Theoretical bounds for LAB-CVaR improve understanding of its behavior.

LAB-CVaR-logit stabilizes training with theoretical support.

Abstract

Data in the real-world classification problems are always imbalanced or long-tailed, wherein the majority classes have the most of the samples that dominate the model training. In such setting, the naive model tends to have poor performance on the minority classes. Previously, a variety of loss modifications have been proposed to address the long-tailed leaning problem, while these methods either treat the samples in the same class indiscriminatingly or lack a theoretical guarantee. In this paper, we propose two novel approaches based on CVaR (Conditional Value at Risk) to improve the performance of long-tailed learning with a solid theoretical ground. Specifically, we firstly introduce a Label-Aware Bounded CVaR (LAB-CVaR) loss to overcome the pessimistic result of the original CVaR, and further design the optimal weight bounds for LAB-CVaR theoretically. Based on LAB-CVaR, we additionally propose a LAB-CVaR with logit adjustment (LAB-CVaR-logit) loss to stabilize the optimization process, where we also offer the theoretical support. Extensive experiments on real-world datasets with long-tailed label distributions verify the superiority of our proposed methods.