Edge Importance Inference Towards Neighborhood Aware GNNs

Ruochen Shi, Jeevan Thapa, Rui Li

Keywords: GNNs, variational inference, stochastic process

Abstract: Comprehensive model tuning and meticulous training for determining proper scope of neighborhood where graph neural networks (GNNs) aggregate information requires high computation overhead and significant human effort. We propose a probabilistic GNN model that captures the expansion of neighborhood scope as a stochastic process and adaptively sample edges to identify critical pathways contributing to generating informative node features. We develop a novel variational inference algorithm to jointly approximate the posterior of the count of neighborhood hops and learn GNN weights while accounting for edge importance. Experiments on multiple benchmarks demonstrate that by adapting the neighborhood scope to a given dataset our model outperforms GNN variants that require grid search or heuristics for neighborhood scope selection.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation 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.

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: 3862