Abstract
Graph Neural Networks (GNNs) have shown remarkable performance in various graph-based tasks, but their effectiveness often diminishes when applied to heterophilic graphs, where connected nodes tend to have dissimilar features. This paper addresses the challenges of learning on such graphs by proposing a novel approach, Adaptive Node-Specific Aggregation (ANSA), which dynamically adjusts the aggregation of neighbor information based on node-specific characteristics. ANSA employs learnable node embeddings and edge attributes to generate node-specific aggregation weights. We evaluate ANSA on several benchmark heterophilic datasets, demonstrating that it outperforms state-of-the-art GNN models designed for heterophilic graphs. Our results highlight the importance of adaptive aggregation mechanisms for graph learning on complex networks.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
"Adaptive Graph Learning with Node-Specific Aggregation", Technical Disclosure Commons, (January 07, 2025)
https://www.tdcommons.org/dpubs_series/7708