Hypergraph Neural Networks
Overview
Hypergraph neural networks study how to learn from data where one relationship can connect several entities at once, such as a group of brain regions, items, or clinical concepts. Common approaches propagate messages through node–hyperedge incidence, use attention or spectral operators, and increasingly add dynamic hypergraphs, diffusion models, or self-supervised learning. They are applied to tasks including node classification, recommendation, forecasting, medical analysis, and scientific modeling, with evaluations spanning standard hypergraph benchmarks and domain-specific datasets. A central open question is whether gains come from genuine higher-order information rather than stronger propagation or other modeling choices; efficiency, over-smoothing, heterophily, and incomplete or noisy structure also remain challenges.
Latest papers
58 papers7 papers in the last four weeks, down 30% on the four weeks before. 0.0% of all new papers.