Spuriosity Didn't Kill the Classifier: Using Invariant Predictions to Harness Spurious Features

TL;DR

This paper introduces Stable Feature Boosting (SFB), a method that leverages invariant and unstable features for improved out-of-distribution prediction without requiring test-domain labels, supported by theoretical guarantees and empirical results.

Contribution

The paper proposes SFB, a novel algorithm that separates stable and unstable features and uses stable features to adapt unstable ones, with theoretical proof of asymptotic optimality.

Findings

SFB effectively improves out-of-distribution predictions.

Theoretical proof of asymptotic optimality of SFB.

Empirical validation on real and synthetic datasets.

Abstract

To avoid failures on out-of-distribution data, recent works have sought to extract features that have an invariant or stable relationship with the label across domains, discarding "spurious" or unstable features whose relationship with the label changes across domains. However, unstable features often carry complementary information that could boost performance if used correctly in the test domain. In this work, we show how this can be done without test-domain labels. In particular, we prove that pseudo-labels based on stable features provide sufficient guidance for doing so, provided that stable and unstable features are conditionally independent given the label. Based on this theoretical insight, we propose Stable Feature Boosting (SFB), an algorithm for: (i) learning a predictor that separates stable and conditionally-independent unstable features; and (ii) using the stable-feature predictions to adapt the unstable-feature predictions in the test domain. Theoretically, we prove that SFB can learn an asymptotically-optimal predictor without test-domain labels. Empirically, we demonstrate the effectiveness of SFB on real and synthetic data.