AI RESEARCH
Disagreement-Regularized Importance Sampling for Adversarial Label Corruption
arXiv CS.LG
•
ArXi:2605.07551v1 Announce Type: new Standard Importance Sampling (IS) collapses under label corruption because high-norm examples, prioritized for variance reduction, are often adversarial outliers. We formalize this misalignment using an $\varepsilon$-contamination model and propose Disagreement-Regularized Importance Sampling (DR-IS), a sub-sampling method based on loss rank-disagreement across independent proxy ensemble.