InCo: Enhance Domain Generalization in Noisy Environments

Gaojie Jin, Ronghui Mu, Xinping Yi, Xiaowei Huang, Lijun Zhang

Keywords: causality, invariant risk minimization, domain generalization, noisy environments

TL;DR: This work studies the field of IRM and reveals that prior IRM-related approaches may be ineffective in noisy environments, then a new method called InCo is proposed to address these challenges.

Abstract: The Invariant Risk Minimization (IRM) approach aims to address the challenge of domain generalization by training a feature representation that remains invariant across multiple environments. However, in noisy environments, IRM-related techniques such as IRMv1 and VREx may be unable to achieve the optimal IRM solution due to incorrect optimization directions. To address this issue, we introduce InCo (short for Invariant Correlation), a novel approach that effectively tackles the aforementioned challenges in noisy environments. Additionally, we provide a case study to analyze why previous methods may lose ground while InCo can succeed. We offer theoretical analysis from a causal perspective, demonstrating that the invariant correlation of representation with labels across environments is a necessary condition for the optimal invariant predictor in noisy environments, whereas the optimization motivations for other methods may not be. Subsequently, we empirically demonstrate the usefulness of InCo by comparing it with other domain generalization methods on various noisy datasets.

Primary Area: causal reasoning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4017