The authors introduce volatility cluster statistics (VCS) and a differentiable regularizer (VCA) to evaluate and reduce temporally clustered prediction errors in TGNNs, but the formal proof that AP/AU-ROC are blind to such patterns is incorrect.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Temporal-Aware Evaluation and Learning for Temporal Graph Neural Networks
The authors introduce volatility cluster statistics (VCS) and a differentiable regularizer (VCA) to evaluate and reduce temporally clustered prediction errors in TGNNs, but the formal proof that AP/AU-ROC are blind to such patterns is incorrect.