We present graph attention networks (GATs), novel neural network architectures that operate on graph-structured data, leveraging masked self-attentional layers to address the shortcomings of prior methods based on graph convolutions or their approximations. By stacking layers in which nodes are able to attend over their neighborhoods' features, we enable (implicitly) specifying different weights to different nodes in a neighborhood, without requiring any kind of costly matrix operation (such as inversion) or depending on knowing the graph structure upfront... (read more)

PDF Abstract ICLR 2018 PDF ICLR 2018 Abstract

Datasets


Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Graph Classification CIFAR10 100k GAT Accuracy (%) 65.48 # 5
Node Classification Citeseer GAT Accuracy 72.5% # 19
Training Split fixed 20 per node # 1
Validation YES # 1
Node Classification CiteSeer (0.5%) GAT Accuracy 38.2% # 14
Node Classification CiteSeer (1%) GAT Accuracy 46.5% # 14
Node Classification CiteSeer with Public Split: fixed 20 nodes per class GAT Accuracy 72.5% # 15
Node Classification Cora GAT Accuracy 83.0% # 23
Training Split fixed 20 per node # 1
Validation YES # 1
Document Classification Cora GAT Accuracy 83.0% # 3
Node Classification Cora (0.5%) GAT Accuracy 41.4% # 13
Node Classification Cora (1%) GAT Accuracy 48.6% # 13
Node Classification Cora (3%) GAT Accuracy 56.8% # 15
Node Classification Cora with Public Split: fixed 20 nodes per class GAT Accuracy 83.0% # 15
Skeleton Based Action Recognition J-HMBD Early Action GAT 10% 58.1 # 2
Graph Regression Lipophilicity GAT RMSE 0.950 # 9
Node Classification PATTERN 100k GAT Accuracy (%) 75.824 # 7
Node Classification PPI GAT F1 97.3 # 12
Node Classification Pubmed GAT Accuracy 79.00% # 27
Training Split fixed 20 per node # 1
Validation YES # 1
Node Classification PubMed (0.03%) GAT Accuracy 50.9% # 13
Node Classification PubMed (0.05%) GAT Accuracy 50.4% # 14
Node Classification PubMed (0.1%) GAT Accuracy 59.6% # 13
Node Classification PubMed with Public Split: fixed 20 nodes per class GAT Accuracy 79.0% # 17
Graph Regression ZINC 100k GAT MAE 0.463 # 7
Graph Regression ZINC-500k GAT MAE 0.384 # 11

Results from Other Papers


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK SOURCE PAPER COMPARE
Node Classification Brazil Air-Traffic GAT (Velickovic et al., 2018) Accuracy 0.382 # 7
Node Classification Europe Air-Traffic GAT (Velickovic et al., 2018) Accuracy 42.4 # 4
Node Classification Flickr GAT (Velickovic et al., 2018) Accuracy 0.359 # 6
Node Classification USA Air-Traffic GAT (Velickovic et al., 2018) Accuracy 58.5 # 2
Node Classification Wiki-Vote GAT (Velickovic et al., 2018) Accuracy 59.4 # 2

Methods used in the Paper


METHOD TYPE
GAT
Graph Models