Search papers, labs, and topics across Lattice.
The paper introduces HEDGE, a novel diffusion model for hypergraph generation that operates directly on relaxed incidence matrices. It uses a hypergraph-specific two-sided heat operator combined with an Ornstein-Uhlenbeck process for structure-aware noising, leading to a tractable Gaussian terminal law. By learning a permutation-equivariant reverse-drift field and simulating a reverse-time SDE, HEDGE achieves improved hypergraph generation quality compared to existing methods.
Forget trying to shoehorn hypergraphs into pairwise representations – this diffusion model directly generates them from incidence matrices, unlocking more realistic and complex structures.
Hypergraphs model higher-order interactions, but realistic hypergraph generation remains difficult because incidence, hyperedge-size heterogeneity, and overlap structure are not faithfully captured by pairwise reductions. We propose \HEDGE, a generative model defined directly on relaxed incidence matrices via a structured stochastic diffusion. The forward process combines a hypergraph-specific two-sided heat operator with an Ornstein--Uhlenbeck component, preserving structure-aware noising near the data while yielding an explicit Gaussian terminal law. Conditional on an observed hypergraph, this forward process is linear-Gaussian, so conditional means, covariances, scores, and reverse-drift targets are available in closed form. We therefore learn a permutation-equivariant state-only reverse-drift field in incidence space by regressing onto exact conditional targets, and generate samples by simulating a learned reverse-time SDE from the Gaussian base law. We establish exactness in the ideal state-only setting together with finite-horizon stability guarantees, and empirically show improved hypergraph generation quality relative to strong baselines.
