Scaling Higher-Order Graph Learning with Maximal Clique Complexes
Researchers have developed a new framework for higher-order graph learning that addresses the scalability limitations of existing methods. The approach introduces simplified and factored cellular Weisfeiler Leman tests to enhance computational efficiency while maintaining expressivity. Additionally, a novel maximal clique complex and a biased random walk method called CliqueWalk are proposed to enable scalable learning with reduced time and memory complexity. AI
IMPACT Enables more expressive and scalable analysis of complex relational data, potentially improving performance in areas like drug discovery and social network analysis.