Reducing Bias in Feature Extractors for Extreme Universal Domain Adaptation
Keywords: Domain Adaptation, Universal Domain Adaptation, Machine Learning
TL;DR: We propose a lightweight self-supervised loss that reduces bias in feature extractors and enhances performance across all class-set distributions, particularly in extreme universal domain adaptation.
Abstract: Universal Domain Adaptation (UniDA) aims to transfer knowledge from a labeled source domain to an unlabeled target domain without prior knowledge of the label sets between the two domains. The goal of UniDA is to achieve robust performance under arbitrary label-set distributions. However, existing literature has not sufficiently explored performance across diverse distribution scenarios. Our experiments reveal that existing methods struggle when the source domain has significantly more non-overlapping classes than overlapping ones, a setting we refer to as *Extreme UniDA*. In this paper, we demonstrate that classical partial domain alignment, which focuses on aligning only overlapping-class data between domains, is limited in mitigating feature extractor bias in extreme UniDA scenarios.
We argue that feature extractors trained with source supervised loss disrupt the intrinsic structure of target data due to the inherent differences between source-private-class data and target data. To mitigate this bias, we employ self-supervised learning to preserve the structure of target data.
This method can be easily integrated into existing frameworks. We apply the proposed approach to two distinct training paradigms—adversarial-based and optimal-transport-based—and show consistent improvements across various class-set distributions, with significant gains in extreme UniDA settings.
Primary Area: other topics in machine learning (i.e., none of the above)
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/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 13410
Loading