HOPSE: Scalable Higher-Order Positional and Structural Encoder for Combinatorial Representations
Researchers have developed HOPSE, a new framework designed to enhance the scalability of Topological Deep Learning. This approach moves away from traditional message-passing layers, instead utilizing Hasse graph decompositions to create efficient encodings for higher-order interactions. HOPSE achieves linear scalability with the size of combinatorial representations, matching or exceeding the performance of existing methods on molecular and topological benchmarks while offering significant speedups. AI
IMPACT Enables more efficient modeling of complex, multi-way relationships in data, potentially improving performance in fields like drug discovery and materials science.