<ul><li>The research paper introduces HOPSE, a framework for Topological Deep Learning (TDL) that accommodates higher-order interactions in complex real-world systems.</li><li>HOPSE aims to overcome scalability challenges faced by existing TDL methods through a message passing-free approach that uses Hasse graph decompositions to efficiently encode higher-order domains.</li><li>Experiments on molecular, expressivity, and topological benchmarks demonstrate that HOPSE achieves state-of-the-art performance and up to 7 times speedups compared to Higher-Order Message Passing (HOMP) models.</li><li>HOPSE offers a promising direction for scalable TDL by providing linear scalability with dataset size while maintaining expressive power and permutation equivariance.</li></ul>

HOPSE: Scalable Higher-Order Positional and Structural Encoder for Combinatorial Representations

Discover more