Robust Validation: Confident Predictions Even When Distributions Shift

Abstract

While the traditional viewpoint in machine learning and statistics assumes training and testing samples come from the same population, practice belies this fiction. One strategy -- coming from robust statistics and optimization -- is thus to build a model robust to distributional perturbations. In this paper, we take a different approach to describe procedures for robust predictive inference, where a model provides uncertainty estimates on its predictions rather than point predictions. We present a method that produces prediction sets (almost exactly) giving the right coverage level for any test distribution in an $f$-divergence ball around the training population. The method, based on conformal inference, achieves (nearly) valid coverage in finite samples, under only the condition that the training data be exchangeable. An essential component of our methodology is to estimate the amount of expected future data shift and build robustness to it; we develop estimators and prove their consistency for protection and validity of uncertainty estimates under shifts. By experimenting on several large-scale benchmark datasets, including Recht et al.'s CIFAR-v4 and ImageNet-V2 datasets, we provide complementary empirical results that highlight the importance of robust predictive validity.

Published in the Journal of the American Statistical Association (JASA 2024)

References in corpus (10)

• Estimating divergence functionals and the likelihood ratio by convex risk minimization
• Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization
• Certifying Some Distributional Robustness with Principled Adversarial Training
• Robust Wasserstein Profile Inference and Applications to Machine Learning
• Detecting and Correcting for Label Shift with Black Box Predictors
• Cold Case: The Lost MNIST Digits
• Conformal Prediction Under Covariate Shift
• Knowing what you know: valid and validated confidence sets in multiclass and multilabel prediction
• Sense and Sensitivity Analysis: Simple Post-Hoc Analysis of Bias Due to Unobserved Confounding
• The $s$-value: evaluating stability with respect to distributional shifts

Cited by in corpus (11)

• Conformal Prediction Under Feedback Covariate Shift for Biomolecular Design
• Adaptive Conformal Inference Under Distribution Shift
• Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Control
• Conformal Predictive Safety Filter for RL Controllers in Dynamic Environments
• Private Prediction Sets
• PAC Confidence Predictions for Deep Neural Network Classifiers
• PAC Prediction Sets Under Covariate Shift
• Sensitivity Analysis of Individual Treatment Effects: A Robust Conformal Inference Approach
• Coverage-Guaranteed Prediction Sets for Out-of-Distribution Data
• Understanding the Under-Coverage Bias in Uncertainty Estimation
• Conformal prediction for uncertainties in nucleon-nucleon scattering