PAGER: A Framework for Failure Analysis of Deep Regression Models

TL;DR

PAGER is a framework that combines uncertainty estimates and data manifold analysis to effectively detect and characterize failures in deep regression models, enhancing safe deployment.

Contribution

It introduces PAGER, a novel systematic approach that unifies uncertainty and non-conformity scores for failure detection in deep regressors.

Findings

Uncertainty alone is insufficient for failure detection.

PAGER improves failure detection accuracy.

Framework unifies multiple failure characterization methods.

Abstract

Safe deployment of AI models requires proactive detection of failures to prevent costly errors. To this end, we study the important problem of detecting failures in deep regression models. Existing approaches rely on epistemic uncertainty estimates or inconsistency w.r.t the training data to identify failure. Interestingly, we find that while uncertainties are necessary they are insufficient to accurately characterize failure in practice. Hence, we introduce PAGER (Principled Analysis of Generalization Errors in Regressors), a framework to systematically detect and characterize failures in deep regressors. Built upon the principle of anchored training in deep models, PAGER unifies both epistemic uncertainty and complementary manifold non-conformity scores to accurately organize samples into different risk regimes.