English

Adapting to Shifting Correlations with Unlabeled Data Calibration

Machine Learning 2024-09-11 v1

Abstract

Distribution shifts between sites can seriously degrade model performance since models are prone to exploiting unstable correlations. Thus, many methods try to find features that are stable across sites and discard unstable features. However, unstable features might have complementary information that, if used appropriately, could increase accuracy. More recent methods try to adapt to unstable features at the new sites to achieve higher accuracy. However, they make unrealistic assumptions or fail to scale to multiple confounding features. We propose Generalized Prevalence Adjustment (GPA for short), a flexible method that adjusts model predictions to the shifting correlations between prediction target and confounders to safely exploit unstable features. GPA can infer the interaction between target and confounders in new sites using unlabeled samples from those sites. We evaluate GPA on several real and synthetic datasets, and show that it outperforms competitive baselines.

Keywords

Cite

@article{arxiv.2409.05996,
  title  = {Adapting to Shifting Correlations with Unlabeled Data Calibration},
  author = {Minh Nguyen and Alan Q. Wang and Heejong Kim and Mert R. Sabuncu},
  journal= {arXiv preprint arXiv:2409.05996},
  year   = {2024}
}

Comments

Accepted at ECCV

R2 v1 2026-06-28T18:39:08.234Z