ABSTRACT
This paper introduces a multi-marginal optimal transport formulation to extend generalization bounds to hypergraph and heterogeneous graph transduction, capturing high-order relationships and type-specific dependencies.
PAPER · PDF
Loading PDF...
Key findings
Proposes a multi-marginal optimal transport framework for complex graph transduction.
Derives novel generalization bounds based on transductive Rademacher complexity and multi-marginal Wasserstein distance.
Designs algorithms for hypergraph and heterogeneous graph OT transduction with implementation details.
Limitations & open questions
Theoretical analysis and practical methodologies need further validation on more diverse datasets.
Computational tractability for very large-scale hypergraphs and heterogeneous graphs remains a challenge.