Conformal Prediction Adaptive to Unknown Subpopulation Shifts

2025-06-05

AI Summary

Traditional conformal prediction fails to provide theoretically guaranteed coverage under distribution shifts caused by unknown changes in subgroup mixture proportions. Method: We propose an adaptive calibration framework that requires neither subgroup labels nor structural priors, integrating robust optimization with weighted conformal inference. The method is compatible with black-box models (e.g., ViT, LLMs) and supports both batch processing and online updates. Contribution/Results: To our knowledge, this is the first approach to achieve provable coverage guarantees under arbitrary, unknown subgroup shifts. Evaluated on multimodal vision-and-language benchmarks, it consistently attains target coverage levels (e.g., 90%) while reducing risk control error by 47% compared to standard conformal methods. Moreover, it significantly improves generalization robustness across diverse shift scenarios—without access to subgroup identities or shift mechanisms.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Conformal prediction is widely used to equip black-box machine learning models with uncertainty quantification enjoying formal coverage guarantees. However, these guarantees typically break down in the presence of distribution shifts, where the data distribution at test time differs from the training (or calibration-time) distribution. In this work, we address subpopulation shifts, where the test environment exhibits an unknown and differing mixture of subpopulations compared to the calibration data. We propose new methods that provably adapt conformal prediction to such shifts, ensuring valid coverage without requiring explicit knowledge of subpopulation structure. Our algorithms scale to high-dimensional settings and perform effectively in realistic machine learning tasks. Extensive experiments on vision (with vision transformers) and language (with large language models) benchmarks demonstrate that our methods reliably maintain coverage and controls risk in scenarios where standard conformal prediction fails.

Problem

• Adapt conformal prediction to unknown subpopulation shifts
• Ensure valid coverage without subpopulation structure knowledge
• Maintain coverage in high-dimensional realistic ML tasks

Innovation

• Adapts conformal prediction to unknown subpopulation shifts
• Ensures valid coverage without subpopulation structure knowledge
• Scales to high-dimensional settings effectively