Propagation Alone is Enough for Graph Contrastive Learning
Keywords: graph contrastive learning, polynomial GNNs
Abstract: Graph contrastive learning has recently gained substantial attention, leading to the development of various methodologies. In this work, we reveal that a simple training-free propagation method PROP achieves competitive results over dedicatedly designed GCL methods across a diverse set of node classification benchmarks. We elucidate the underlying rationale for PROP’s effectiveness by drawing connections between the propagation operator and established unsupervised learning algorithms. To investigate the reasons for the suboptimal performance of GCL, we decouple the propagation and transformation phases of graph neural networks. Our findings indicate that existing GCL methods inadequately learns effective transformation weights while exhibiting potential for solid propagation learning. In light of these insights, we enhance PROP with learnable propagation, introducing a novel GCL method termed PROPGCL. The effectiveness of PROPGCL is demonstrated through comprehensive evaluations on node classification tasks.
Supplementary Material: zip
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: 10638
Loading