Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Fishr: Invariant Gradient Variances for Out-of-Distribution Generalization
(Submitted on 7 Sep 2021 (v1), last revised 1 Jun 2022 (this version, v3))
Abstract: Learning robust models that generalize well under changes in the data distribution is critical for real-world applications. To this end, there has been a growing surge of interest to learn simultaneously from multiple training domains - while enforcing different types of invariance across those domains. Yet, all existing approaches fail to show systematic benefits under controlled evaluation protocols. In this paper, we introduce a new regularization - named Fishr - that enforces domain invariance in the space of the gradients of the loss: specifically, the domain-level variances of gradients are matched across training domains. Our approach is based on the close relations between the gradient covariance, the Fisher Information and the Hessian of the loss: in particular, we show that Fishr eventually aligns the domain-level loss landscapes locally around the final weights. Extensive experiments demonstrate the effectiveness of Fishr for out-of-distribution generalization. Notably, Fishr improves the state of the art on the DomainBed benchmark and performs consistently better than Empirical Risk Minimization. Our code is available at this https URL
Submission history
From: Alexandre Rame [view email][v1] Tue, 7 Sep 2021 08:36:09 GMT (2083kb,D)
[v2] Sun, 17 Oct 2021 12:02:14 GMT (2847kb,D)
[v3] Wed, 1 Jun 2022 14:37:01 GMT (2400kb,D)
Link back to: arXiv, form interface, contact.