LMC: Fast Training of GNNs via Subgraph Sampling with Provable Convergence

arXiv:2302.00924v3 Announce Type: replace
Abstract: The message passing-based graph neural networks (GNNs) have achieved great success in many real-world applications. However, training GNNs on large-scale graphs suffers from the well-known neighbor explosion problem, i.e., the exponentially increasing dependencies of nodes with the number of message passing layers. Subgraph-wise sampling methods -- a promising class of mini-batch training techniques -- discard messages outside the mini-batches in backward passes to avoid the neighbor explosion problem at the expense of gradient estimation accuracy. …

abstract applications arxiv convergence cs.lg dependencies gnns graph graph neural networks graphs however networks neural networks nodes sampling scale success training type via wise world