How Powerful are Graph Neural Networks?

ICLR 2019 Keyulu XuWeihua HuJure LeskovecStefanie Jegelka

Graph Neural Networks (GNNs) are an effective framework for representation learning of graphs. GNNs follow a neighborhood aggregation scheme, where the representation vector of a node is computed by recursively aggregating and transforming representation vectors of its neighboring nodes... (read more)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Graph Classification BP-fMRI-97 GIN Accuracy 45.4% # 8
Graph Classification BP-fMRI-97 GIN F1 42.3% # 8
Graph Classification COLLAB GIN-0 Accuracy 80.2% # 6
Graph Classification HIV-DTI-77 GIN Accuracy 55.1% # 6
Graph Classification HIV-DTI-77 GIN F1 53.6% # 6
Graph Classification HIV-fMRI-77 GIN Accuracy 52.5% # 6
Graph Classification HIV-fMRI-77 GIN F1 35.6% # 5
Graph Classification IMDb-B GIN-0 Accuracy 75.1% # 6
Graph Classification IMDb-M GIN-0 Accuracy 52.3% # 4
Graph Classification MUTAG GIN-0 Accuracy 89.4% # 12
Graph Classification NCI1 GIN-0 Accuracy 82.7% # 12
Graph Classification PROTEINS GIN-0 Accuracy 76,2% # 46
Graph Classification PTC GIN-0 Accuracy 64.40 # 11
Graph Classification REDDIT-B GIN-0 Accuracy 92.4 # 1
Graph Classification RE-M5K GIN-0 Accuracy 57.5% # 1