Deterministic sampling plus hot-node caching and prefetching speeds up distributed GNN training by about 2.5x to 3x on tested graphs while cutting remote fetches and energy.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
RapidGNN: Energy and Communication-Efficient Distributed Training on Large-Scale Graph Neural Networks
Deterministic sampling plus hot-node caching and prefetching speeds up distributed GNN training by about 2.5x to 3x on tested graphs while cutting remote fetches and energy.