Understanding the Role of Spectral Signal in Unsupervised Graph Domain Adaptation

Qixuan Gao, Xiangyi Teng, Shibing Mo, Jing Liu

Keywords: Unsupervised graph domain adaptation; Spectral signal; low- and high-frequency information

Abstract: Unsupervised graph domain adaptation (GDA) addresses the challenge of transferring knowledge from labeled source graphs to unlabeled target graphs. However, existing methods primarily implement spatial message-passing operators, which are limited by the neglect of the unique roles of spectral signals in unsupervised GDA. In this paper, we initially investigate an experimental study and find that the low-frequency topology signals signify the shared cross-domain features, while the high-frequency information indicates domain-specific knowledge. However, how to effectively leverage the above findings persists as a perplexing conundrum. To tackle the above issue, we propose an effective framework named Synergy Low-High Frequency Cross-Domain Network (SnLH) for unsupervised GDA. Specifically, we decouple the low- and high-frequency components in the original graph, extracting global structures and local details to capture richer semantic information and enhance the graph-level semantics. For the low-frequency components, we design an optimization objective to maximize the mutual information among low-frequency features, promoting the model to learn more generalized low-frequency information. To further mitigate domain discrepancy, we introduce high-frequency information cross-domain contrastive learning to impose constraints on the domains. By effectively leveraging both low and high-frequency information, the learned features turn out to be both discriminative and domain-invariant, thereby attaining effective cross-domain knowledge transfer. Extensive experiments demonstrate the superiority and effectiveness of the proposed framework across various state-of-the-art unsupervised GDA baselines.

Primary Area: learning on graphs and other geometries & topologies

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: 7357