Domain Generalization via Rationale Invariance

TL;DR

This paper introduces a novel approach for domain generalization by enforcing invariance in the rationale matrices of the classifier, leading to more robust performance across unseen environments.

Contribution

It proposes a simple regularization technique called rationale invariance loss that encourages domain-invariant decision rationales in classifiers.

Findings

Achieves competitive results on multiple datasets

Requires minimal code for implementation

Enhances model robustness to domain shifts

Abstract

This paper offers a new perspective to ease the challenge of domain generalization, which involves maintaining robust results even in unseen environments. Our design focuses on the decision-making process in the final classifier layer. Specifically, we propose treating the element-wise contributions to the final results as the rationale for making a decision and representing the rationale for each sample as a matrix. For a well-generalized model, we suggest the rationale matrices for samples belonging to the same category should be similar, indicating the model relies on domain-invariant clues to make decisions, thereby ensuring robust results. To implement this idea, we introduce a rationale invariance loss as a simple regularization technique, requiring only a few lines of code. Our experiments demonstrate that the proposed approach achieves competitive results across various datasets, despite its simplicity. Code is available at \url{https://github.com/liangchen527/RIDG}.