提出Focal-SAM，提升长尾分类的泛化能力。
Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification
- 按类别差异惩罚梯度尖锐度，精细控制损失曲面。
- 在CIFAR-10-LT和ImageNet-LT上，尾部类别准确率提升5.2%以上。
- 无需额外反向传播，适合大模型长尾场景。
真实世界数据集通常呈长尾分布，使尾部类别的泛化困难。现有方法如ImbSAM和CC-SAM采用长尾版尖锐感知最小化（SAM）来平坦损失曲面以改善泛化，但面临效率与控制力之间的权衡：ImbSAM虽高效但仅粗粒度控制（排除头部类参与SAM），CC-SAM虽可细粒度调节（依赖类别相关扰动）却因多次反向传播而低效。为此，本文提出Focal-SAM，通过为不同类别分配差异化尖锐度惩罚，在不增加反向传播次数的前提下实现精细控制，兼顾效率与性能。我们还从理论上分析了Focal-SAM的泛化能力，并推导出更紧的泛化界。在传统模型与基础模型上的大量实验验证了其有效性，尤其在CIFAR-10-LT和ImageNet-LT数据集上，尾部类别准确率提升超过5.2%。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。