Batch Before You Lift: Scalable Topological Deep Learning on Large Graphs
arXiv:2610.12247v1 Announce Type: cross Abstract: Topological Deep Learning extends graph-based learning to higher-order domains, such as hypergraphs, cellular, and simplicial complexes. These domains are typically constructed from patterns in an input graph through a process of graph lifting. Full-domain training constructs and stores the complete lifted representation before model execution. On large and dense datasets like Reddit (233k nodes and 57.3M edges), this global materialization beco…