DSP-Reg: Domain-Sensitive Parameter Regularization for Robust Domain Generalization

TL;DR

This paper introduces DSP-Reg, a novel regularization method that enhances domain generalization by identifying and suppressing domain-sensitive parameters, leading to improved performance on unseen data distributions.

Contribution

The paper proposes a covariance-based parameter sensitivity analysis and a regularization framework that emphasizes domain-invariant parameters for better generalization.

Findings

DSP-Reg outperforms state-of-the-art methods on multiple benchmarks.

The framework effectively identifies domain-sensitive parameters.

Results show improved accuracy on unseen domains.

Abstract

Domain Generalization (DG) is a critical area that focuses on developing models capable of performing well on data from unseen distributions, which is essential for real-world applications. Existing approaches primarily concentrate on learning domain-invariant features, which assume that a model robust to variations in the source domains will generalize well to unseen target domains. However, these approaches neglect a deeper analysis at the parameter level, which makes the model hard to explicitly differentiate between parameters sensitive to domain shifts and those robust, potentially hindering its overall ability to generalize. In order to address these limitations, we first build a covariance-based parameter sensitivity analysis framework to quantify the sensitivity of each parameter in a model to domain shifts. By computing the covariance of parameter gradients across multiple source domains, we can identify parameters that are more susceptible to domain variations, which serves as our theoretical foundation. Based on this, we propose Domain-Sensitive Parameter Regularization (DSP-Reg), a principled framework that guides model optimization by a soft regularization technique that encourages the model to rely more on domain-invariant parameters while suppressing those that are domain-specific. This approach provides a more granular control over the model's learning process, leading to improved robustness and generalization to unseen domains. Extensive experiments on benchmarks, such as PACS, VLCS, OfficeHome, and DomainNet, demonstrate that DSP-Reg outperforms state-of-the-art approaches, achieving an average accuracy of 66.7\% and surpassing all baselines.