Bias as a Virtue: Rethinking Generalization under Distribution Shifts

TL;DR

This paper proposes that intentionally increasing in-distribution bias during training can improve out-of-distribution generalization, challenging traditional validation methods and introducing a new framework called Adaptive Distribution Bridge.

Contribution

The paper introduces the Adaptive Distribution Bridge framework that leverages controlled statistical diversity to enhance OOD generalization, providing both practical methods and theoretical insights.

Findings

Higher in-distribution bias correlates with lower out-of-distribution error.

ADB achieves up to 26.8% reduction in mean error compared to traditional methods.

High percentile ranks (>74.4%) indicate effective identification of robust training strategies.

Abstract

Machine learning models often degrade when deployed on data distributions different from their training data. Challenging conventional validation paradigms, we demonstrate that higher in-distribution (ID) bias can lead to better out-of-distribution (OOD) generalization. Our Adaptive Distribution Bridge (ADB) framework implements this insight by introducing controlled statistical diversity during training, enabling models to develop bias profiles that effectively generalize across distributions. Empirically, we observe a robust negative correlation where higher ID bias corresponds to lower OOD error--a finding that contradicts standard practices focused on minimizing validation error. Evaluation on multiple datasets shows our approach significantly improves OOD generalization. ADB achieves robust mean error reductions of up to 26.8% compared to traditional cross-validation, and consistently identifies high-performing training strategies, evidenced by percentile ranks often exceeding 74.4%. Our work provides both a practical method for improving generalization and a theoretical framework for reconsidering the role of bias in robust machine learning.