Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification

TL;DR

Focal-SAM introduces a novel approach to long-tailed classification by controlling class-wise sharpness efficiently, balancing fine-grained landscape control with computational cost, and demonstrating improved generalization.

Contribution

It proposes Focal-SAM, a method that achieves fine-grained sharpness control without extra backpropagations, addressing efficiency and control trade-offs in long-tail learning.

Findings

Focal-SAM outperforms existing methods on traditional and foundation models.

Theoretical analysis shows improved generalization bounds.

Extensive experiments validate effectiveness across datasets.

Abstract

Real-world datasets often follow a long-tailed distribution, making generalization to tail classes difficult. Recent methods resorted to long-tail variants of Sharpness-Aware Minimization (SAM), such as ImbSAM and CC-SAM, to improve generalization by flattening the loss landscape. However, these attempts face a trade-off between computational efficiency and control over the loss landscape. On the one hand, ImbSAM is efficient but offers only coarse control as it excludes head classes from the SAM process. On the other hand, CC-SAM provides fine-grained control through class-dependent perturbations but at the cost of efficiency due to multiple backpropagations. Seeing this dilemma, we introduce Focal-SAM, which assigns different penalties to class-wise sharpness, achieving fine-grained control without extra backpropagations, thus maintaining efficiency. Furthermore, we theoretically analyze Focal-SAM's generalization ability and derive a sharper generalization bound. Extensive experiments on both traditional and foundation models validate the effectiveness of Focal-SAM.