Chordal Graph Sampling-Based Mini-batch Training Algorithm for Large Graphs
Keywords: Large scale dataset, Graph neural networks
Abstract: Graph Neural Networks (GNNs) are powerful models for learning representations of attributed graphs. To scale GNNs to large graphs, many methods use various techniques, such as sampling and decoupling, to alleviate the “neighbor explosion” problem during mini-batch training. However, these sampling-based mini-batch training methods often suffer from greater information loss than decoupling-based methods or full-batch GCNs. Besides, most original segmentation methods for large graphs usually lose a large number of edges, resulting in suboptimal performance when performing mini-batch training. Therefore, we propose a Chordal Graph Sampling-based mini-batch Training algorithm for GNNs on large-scale graph datasets, called CGST. CGST includes a balanced chordal graph partition module and a batch random aggregation module to improve performance on node classification tasks while maintaining main information of the original graph structure. Experiments on three large-scale graph datasets prove the effectiveness of CGST.
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: 10492
Loading