Label Space-Induced Pseudo Label Refinement for Multi-Source Black-Box Domain Adaptation

Chaehwa Yoo, Xiaofeng Liu, Fangxu Xing, Jonghye Woo, Jewon Kang

Keywords: Pseudo Label Refinement, Domain Adaptation, Black-box Domain Adaptation, Multi-source Domain Adaptation, Self learning

Abstract: Unsupervised Domain Adaptation (UDA) aims to train a model for an unlabeled target domain by transferring knowledge from a source domain. However, standard UDA requires access to source data and models, prohibiting its practical application in terms of privacy and security. Black-Box DA (BDA) reduces such constraints by defining a pseudo label from a single source prediction, which allows for self-training of the target model. Nonetheless, existing methods have limited consideration for multi-source settings, in which multiple source domains are available to generate pseudo labels. In this work, we introduce a novel training framework for multi-source BDA (MSBDA), dubbed Label Space-Induced Pseudo Label Refinement (LPR). Specifically, LPR incorporates a Pseudo label Refinery Network (PRN) that learns the relation between each source conditioned by the target from source predictions. The target model is adapted by self-learning using a pseudo label generated by PRN. We provide theoretical supports for the performance of the LPR. Experimental results on four benchmark datasets demonstrate that MSBDA using LPR achieves highly competitive performance compared to state-of-the-art approaches with different DA settings.

Primary Area: transfer learning, meta learning, and lifelong learning

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: 5000