Random node sampling in GNN mini-batches implicitly minimizes sampled loss plus a gradient-variance regularizer, yielding performance equal or superior to full-graph training on most datasets.
Combined with ν(d) =Cd −α(1 +o(1)) and the fact that d→ ∞ uniformly on Wk, this yields ν(d) =C k q −α (1 +o(1)),uniformly ford∈W k
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Implicit Regularization of Mini-Batch Training in Graph Neural Networks
Random node sampling in GNN mini-batches implicitly minimizes sampled loss plus a gradient-variance regularizer, yielding performance equal or superior to full-graph training on most datasets.