notesum.ai

Published at November 17

IMPaCT GNN: Imposing invariance with Message Passing in Chronological split Temporal Graphs

cs.AI

stat.ML

Released Date: November 17, 2024

Authors: Sejun Park¹, Joo Young Park², Hyunwoo Park³

Aff.: ¹Electrical and Computer Engineering, Seoul National University; ²College of Medicine, Seoul National University; ³Graduate School of Data Science, Seoul National University

Arxiv: http://arxiv.org/abs/2411.10957v1

Dataset	Model	Test Acc. $\pm$ Std.	Valid Acc. $\pm$ Std.
ogbn-mag	LDHGNN (baseline)	0.8789 $\pm$ 0.0024	0.8836 $\pm$ 0.0028
	LDHGNN+MMP	0.8945 $\pm$ 0.0018	0.8996 $\pm$ 0.0015
	LDHGNN+PMP	0.9093 $\pm$ 0.0040	0.9173 $\pm$ 0.0027
	LDHGNN+PMP+JJnorm	0.9178 $\pm$ 0.0019	0.9236 $\pm$ 0.0030
ogbn-arxiv	Linear-RevGAT+GIANT (baseline)	0.7065 $\pm$ 0.0010	0.7287 $\pm$ 0.0009
	Linear-RevGAT+GIANT+GenPMP	0.7468 $\pm$ 0.0006	0.7568 $\pm$ 0.0010
	RevGAT+SimTeG+TAPE	0.7803 $\pm$ 0.0007	0.7846 $\pm$ 0.0004
ogbn-papers100m	GAMLP+GIANT+RLU (baseline)	0.6967 $\pm$ 0.0006	0.7305 $\pm$ 0.0004
	GAMLP+GIANT+RLU+GenPMP	0.6976 $\pm$ 0.0010	0.7330 $\pm$ 0.0006
	GAMLP+GLEM+GIANT	0.7037 $\pm$ 0.0002	0.7354 $\pm$ 0.0001