ODD: Overlap-aware Estimation of Model Performance under Distribution Shift

TL;DR

This paper introduces ODD, a new method for estimating model performance under distribution shift by focusing on non-overlapping regions, leading to more accurate and reliable error bounds.

Contribution

The paper proposes Overlap-aware Disagreement Discrepancy (ODD), a novel approach that improves error estimation by addressing overlap issues in distribution shift scenarios.

Findings

ODD outperforms DIS^2 in predicting target error.

ODD provides more reliable error bounds across benchmarks.

The method effectively estimates domain-overlap using domain classifiers.

Abstract

Reliable and accurate estimation of the error of an ML model in unseen test domains is an important problem for safe intelligent systems. Prior work uses disagreement discrepancy (DIS^2) to derive practical error bounds under distribution shifts. It optimizes for a maximally disagreeing classifier on the target domain to bound the error of a given source classifier. Although this approach offers a reliable and competitively accurate estimate of the target error, we identify a problem in this approach which causes the disagreement discrepancy objective to compete in the overlapping region between source and target domains. With an intuitive assumption that the target disagreement should be no more than the source disagreement in the overlapping region due to high enough support, we devise Overlap-aware Disagreement Discrepancy (ODD). Maximizing ODD only requires disagreement in the non-overlapping target domain, removing the competition. Our ODD-based bound uses domain-classifiers to estimate domain-overlap and better predicts target performance than DIS^2. We conduct experiments on a wide array of benchmarks to show that our method improves the overall performance-estimation error while remaining valid and reliable. Our code and results are available on GitHub.