Explainable Transfer Learning on Graphs Using a Novel Label Frequency Representation
Keywords: Transfer Learning; Graph Representation; Graph domain adaptation
TL;DR: This paper presents a novel graph representation using node label frequencies, enabling explainable and effective transfer learning across diverse domains.
Abstract: Graphs are characterized by their versatility in representing objects from a wide range of domains, such as social networks or protein structures. This flexibility and power poses a significant challenge for transfer learning between graph domains. Current methods of transfer learning between graph domains tend to focus exclusively on the structure of the underlying graphs, neglecting the characteristics of the nodes and not addressing the difficulties in comparing nodes that represent very dissimilar entities, such as atoms and people for instance. In this paper, we propose a novel universal representation of graphs based on the relative frequency of the node labels. This novel representation enables explainable transfer learning between labeled graphs from different domains for the first time, without the need for additional adaptations. That is, we show that our novel representation can be readily combined with a data alignment technique that in turn allows transfer learning between data from different domains. Experimental results show that knowledge can be acquired from graphs belonging to chemical and biological domains to improve the accuracy of classification models in social network analysis. A comparison with state-of-the-art techniques indicates that our approach outperforms existing non-topological methods and, in some cases, even graph neural networks. In summary, our technique represents a major advance in graph node representation for transfer learning between different domains, opening up new perspectives for future research.
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/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: 4632
Loading