Batch Before You Lift: Scalable Topological Deep Learning on Large Graphs

Key points

• To address this limitation, we introduce Cluster-TNN, a domain-agnostic framework that avoids this bottleneck by lifting locally instead.1
• After partitioning the input graph during preprocessing, at runtime Cluster-TNN dynamically samples groups of node clusters, reconstructs their induced subgraphs to form mini-batches, and applies the chosen lifting within each mini-batch.1
• Retaining all edges among sampled nodes preserves the connectivity needed to construct higher-order structures across clusters, producing topological mini-batches that existing Topological Neural Networks can process directly.1
• These results establish Cluster-TNN as a general strategy for scaling Topological Deep Learning beyond the limitations of global domain construction.1

Sources (1)

• [1]Batch Before You Lift: Scalable Topological Deep Learning on Large Graphs
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 04:23 PM

  Topological Deep Learning extends graph-based learning to higher-order domains, such as hypergraphs, cellular, and simplicial complexes.

  To address this limitation, we introduce Cluster-TNN, a domain-agnostic framework that avoids this bottleneck by lifting locally instead.

Extractive summary: sentences quoted from the sources.