纠正特征与分类器空间错位，提升长尾学习的泛化能力
Space Alignment Matters: The Missing Piece for Inducing Neural Collapse in Long-Tailed Learning
- 提出三种可插拔的对齐策略，修复特征与分类器空间的错位
- 在多个长尾数据集上显著提升基线模型性能，达到新最好结果
- 理论分析揭示空间错位是阻碍神经坍缩的关键因素，适合改进现有长尾方法
近期研究发现，在类别平衡条件下，类特征均值与分类器权重会自发对齐为等角紧框架（ETF）。但在长尾分布下，严重的样本不平衡会抑制神经坍缩现象，导致泛化性能下降。现有方法多通过约束特征或分类器权重来恢复ETF结构，却忽视了一个关键问题：特征空间与分类器权重空间之间存在明显错位。本文通过最优误差指数分析，定量揭示了这种错位的危害。基于此，我们提出了三种无需改变架构的可插拔对齐策略。在CIFAR-10-LT、CIFAR-100-LT和ImageNet-LT上的大量实验表明，这些策略能持续提升现有长尾方法的性能，并达到当前最佳水平。
原文摘要 · Abstract (English)
Recent studies on Neural Collapse (NC) reveal that, under class-balanced conditions, the class feature means and classifier weights spontaneously align into a simplex equiangular tight frame (ETF). In long-tailed regimes, however, severe sample imbalance tends to prevent the emergence of the NC phenomenon, resulting in poor generalization performance. Current efforts predominantly seek to recover the ETF geometry by imposing constraints on features or classifier weights, yet overlook a critical problem: There is a pronounced misalignment between the feature and the classifier weight spaces. In this paper, we theoretically quantify the harm of such misalignment through an optimal error exponent analysis. Built on this insight, we propose three explicit alignment strategies that plug-and-play into existing long-tail methods without architectural change. Extensive experiments on the CIFAR-10-LT, CIFAR-100-LT, and ImageNet-LT datasets consistently boost examined baselines and achieve the state-of-the-art performances.
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。