Flowette: Graph Generation with Graphette Priors
Flowette, an innovative framework, leverages continuous flow matching alongside graph neural networks to create graphs featuring recurring subgraph motifs. It presents graphettes, a probabilistic category of graph structure models that extend graphons through controlled motif edits, including rings, stars, and trees. The model utilizes optimal transport-based coupling for topology-sensitive alignment and incorporates regularization to ensure overall structural coherence. Theoretical investigations address coupling, invariance, and structural characteristics. Assessments conducted on synthetic and molecular benchmarks differentiate the impacts of the structural prior, optimal-transport coupling, and regularization components.
Key facts
- Flowette is a continuous flow matching framework for graph generation.
- It uses a graph neural network-based transformer to learn a velocity field.
- Graphettes are introduced as a new probabilistic family of graph structure models.
- Graphettes generalize graphons via controlled structural edits for motifs.
- Motifs include rings, stars, and trees.
- Optimal transport-based coupling promotes topology-aware alignment.
- Regularization encourages global structural coherence.
- Evaluated on synthetic and molecular benchmarks with controlled ablations.
Entities
—