Unsupervised Multi-Scale Gromov-Wasserstein Hypergraph Alignment
FALCON aligns hypergraph nodes without features, labels, seed matches, or added hyperedge nodes.
The paper frames hypergraph alignment as a structural-only problem and targets the cost of direct higher-order methods. Its method builds multiple clique-based co-occurrence dissimilarity matrices from a filtration, then aligns them with one shared multi-scale Gromov-Wasserstein objective. The shared transport plan is meant to keep node correspondences consistent across levels. On real-world perturbation benchmarks, the authors report stronger robustness to structural noise and near-universal wins over graph and hypergraph alignment baselines. ArXiv · AI/CL/LG's note
The paper frames hypergraph alignment as a structural-only problem and targets the cost of direct higher-order methods. Its method builds multiple clique-based co-occurrence dissimilarity matrices from a filtration, then aligns them with one shared multi-scale Gromov-Wasserstein objective. The shared transport plan is meant to keep node correspondences consistent across levels. On real-world perturbation benchmarks, the authors report stronger robustness to structural noise and near-universal wins over graph and hypergraph alignment baselines. ArXiv · AI/CL/LG's note
score 4