arXiv：cs.LG（机器学习，全量分类）· Sung Ho Jo, Seonghwi Kim, Wonsang Yun, Minwoo Chae·· 14 小时前AI 评分35
POTER：面向虚假相关与标签噪声的鲁棒学习最优传输重加权框架
Optimal Transport Reweighting for Robust Learning under Spurious Correlations and Label Noise
AI 导读
研究者提出 POTER，一种基于最优传输的重加权框架，通过衡量训练分布与由有限验证组标注构建的参考分布之间的传输几何来推导样本重要性，从而降低误标注或强偏置对齐样本的权重。该方法只需单次 ERM 训练阶段，无需近期工作中常见的重训练范式，在标准基准和噪声标签设置下取得 SOTA 最差组准确率，包括标签污染集中于少数子群的情况。该工作已被 NeurIPS 2026 接收。
正文
Abstract:Machine learning models often suffer performance degradation under subpopulation shift, particularly when spurious correlations cause models to rely on shortcut features that fail to generalize across subgroups. A recent line of work mitigates this issue by using loss-based signals to identify informative samples, but these signals can become severely distorted under label noise: mislabeled samples may also incur large losses and contaminate subsequent reweighting or retraining. Despite its practical importance, this intersection remains largely underexplored. We propose POTER, a reweighting framework based on optimal transport that derives sample importance from the transport geometry between the training distribution and a reference distribution constructed from limited validation group annotations. By measuring alignment at the individual-sample level rather than relying on loss, POTER downweights mislabeled or strongly bias-aligned samples while assigning higher importance to samples better aligned with the reference distribution. In addition, POTER requires only a single ERM training stage, moving beyond the retraining paradigm common in recent work. Across standard benchmarks and noisy-label settings, POTER achieves state-of-the-art worst-group accuracy, including cases where label corruption is concentrated within minority subgroups.
| Comments: | Accepted at NeurIPS 2026 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.01028 [cs.LG] |
| (or arXiv:2610.01028v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.01028 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sung Ho Jo [view email]
[v1]
Thu, 1 Oct 2026 04:16:23 UTC (4,744 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org