Toward Learning Human-aligned Cross-domain Robust Models by Countering Misaligned Features

TL;DR

This paper investigates the impact of misaligned features on cross-domain model robustness, proposing a theoretical framework and techniques to counteract their influence for more human-aligned models.

Contribution

It introduces a new generalization error bound considering misaligned features and links existing robust learning techniques to address this issue.

Findings

Theoretical analysis of misaligned feature impact on generalization.

Empirical evaluation showing combined techniques improve robustness.

Implementation demonstrating practical effectiveness.

Abstract

Machine learning has demonstrated remarkable prediction accuracy over i.i.d data, but the accuracy often drops when tested with data from another distribution. In this paper, we aim to offer another view of this problem in a perspective assuming the reason behind this accuracy drop is the reliance of models on the features that are not aligned well with how a data annotator considers similar across these two datasets. We refer to these features as misaligned features. We extend the conventional generalization error bound to a new one for this setup with the knowledge of how the misaligned features are associated with the label. Our analysis offers a set of techniques for this problem, and these techniques are naturally linked to many previous methods in robust machine learning literature. We also compared the empirical strength of these methods demonstrated the performance when these previous techniques are combined, with an implementation available at https://github.com/OoDBag/WR