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. In this way, we address several key
challenges of spectral-based graph neural networks simultaneously, and make our
model readily applicable to inductive as well as transductive problems. Our GAT
models have achieved or matched state-of-the-art results across four
established transductive and inductive graph benchmarks: the Cora, Citeseer and
Pubmed citation network datasets, as well as a protein-protein interaction
dataset (wherein test graphs remain unseen during training).
%0 Conference Paper
%1 velickovic2017graph
%A Veličković, Petar
%A Cucurull, Guillem
%A Casanova, Arantxa
%A Romero, Adriana
%A Liò, Pietro
%A Bengio, Yoshua
%B ICLR 2018
%D 2017
%K attention graph-based neural-network reserved thema
%T Graph Attention Networks
%U http://arxiv.org/abs/1710.10903
%X 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. In this way, we address several key
challenges of spectral-based graph neural networks simultaneously, and make our
model readily applicable to inductive as well as transductive problems. Our GAT
models have achieved or matched state-of-the-art results across four
established transductive and inductive graph benchmarks: the Cora, Citeseer and
Pubmed citation network datasets, as well as a protein-protein interaction
dataset (wherein test graphs remain unseen during training).
@inproceedings{velickovic2017graph,
abstract = {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. In this way, we address several key
challenges of spectral-based graph neural networks simultaneously, and make our
model readily applicable to inductive as well as transductive problems. Our GAT
models have achieved or matched state-of-the-art results across four
established transductive and inductive graph benchmarks: the Cora, Citeseer and
Pubmed citation network datasets, as well as a protein-protein interaction
dataset (wherein test graphs remain unseen during training).},
added-at = {2019-04-08T12:23:37.000+0200},
author = {Veličković, Petar and Cucurull, Guillem and Casanova, Arantxa and Romero, Adriana and Liò, Pietro and Bengio, Yoshua},
biburl = {https://www.bibsonomy.org/bibtex/2274f56f6fad56a78c619756f1e184e12/e.fischer},
booktitle = {ICLR 2018},
interhash = {d2f5ba17510bf494b75ce245fa72ccdd},
intrahash = {274f56f6fad56a78c619756f1e184e12},
keywords = {attention graph-based neural-network reserved thema},
timestamp = {2019-09-19T15:07:21.000+0200},
title = {Graph Attention Networks},
url = {http://arxiv.org/abs/1710.10903},
year = 2017
}