Using historical labels as pseudo-targets lets temporal GNNs train on label-scarce batches, cutting convergence time by up to 15x on TGB datasets, but the theoretical proof and SOTA claims are weakly supported.
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
REJECT 1representative citing papers
citing papers explorer
-
Never Skip a Batch: Dense Learning of Temporal GNNs via Adaptive Pseudo-Supervision
Using historical labels as pseudo-targets lets temporal GNNs train on label-scarce batches, cutting convergence time by up to 15x on TGB datasets, but the theoretical proof and SOTA claims are weakly supported.