Graph learning is currently dominated by graph kernels, which, while
powerful, suffer some significant limitations. Convolutional Neural Networks
(CNNs) offer a very appealing alternative, but processing graphs with CNNs is
not trivial. To address this challenge, many sophisticated extensions of CNNs
have recently been introduced. In this paper, we reverse the problem: rather
than proposing yet another graph CNN model, we introduce a novel way to
represent graphs as multi-channel image-like structures that allows them to be
handled by vanilla 2D CNNs. Experiments reveal that our method is more accurate
than state-of-the-art graph kernels and graph CNNs on 4 out of 6 real-world
datasets (with and without continuous node attributes), and close elsewhere.
Our approach is also preferable to graph kernels in terms of time complexity.
Code and data are publicly available.
Description
Graph Classification with 2D Convolutional Neural Networks
%0 Generic
%1 tixier2017graph
%A Tixier, Antoine Jean-Pierre
%A Nikolentzos, Giannis
%A Meladianos, Polykarpos
%A Vazirgiannis, Michalis
%D 2017
%K CNN graph
%T Graph Classification with 2D Convolutional Neural Networks
%U http://arxiv.org/abs/1708.02218
%X Graph learning is currently dominated by graph kernels, which, while
powerful, suffer some significant limitations. Convolutional Neural Networks
(CNNs) offer a very appealing alternative, but processing graphs with CNNs is
not trivial. To address this challenge, many sophisticated extensions of CNNs
have recently been introduced. In this paper, we reverse the problem: rather
than proposing yet another graph CNN model, we introduce a novel way to
represent graphs as multi-channel image-like structures that allows them to be
handled by vanilla 2D CNNs. Experiments reveal that our method is more accurate
than state-of-the-art graph kernels and graph CNNs on 4 out of 6 real-world
datasets (with and without continuous node attributes), and close elsewhere.
Our approach is also preferable to graph kernels in terms of time complexity.
Code and data are publicly available.
@misc{tixier2017graph,
abstract = {Graph learning is currently dominated by graph kernels, which, while
powerful, suffer some significant limitations. Convolutional Neural Networks
(CNNs) offer a very appealing alternative, but processing graphs with CNNs is
not trivial. To address this challenge, many sophisticated extensions of CNNs
have recently been introduced. In this paper, we reverse the problem: rather
than proposing yet another graph CNN model, we introduce a novel way to
represent graphs as multi-channel image-like structures that allows them to be
handled by vanilla 2D CNNs. Experiments reveal that our method is more accurate
than state-of-the-art graph kernels and graph CNNs on 4 out of 6 real-world
datasets (with and without continuous node attributes), and close elsewhere.
Our approach is also preferable to graph kernels in terms of time complexity.
Code and data are publicly available.},
added-at = {2018-02-13T09:21:44.000+0100},
author = {Tixier, Antoine Jean-Pierre and Nikolentzos, Giannis and Meladianos, Polykarpos and Vazirgiannis, Michalis},
biburl = {https://www.bibsonomy.org/bibtex/2bfdd5ed14809136bd73d540ae1924504/jk_itwm},
description = {Graph Classification with 2D Convolutional Neural Networks},
interhash = {357c3f683a361ddfdc6e577cf6e70175},
intrahash = {bfdd5ed14809136bd73d540ae1924504},
keywords = {CNN graph},
note = {cite arxiv:1708.02218Comment: Added one dataset and several very recent baselines; changed formatting; corrected typos},
timestamp = {2018-02-13T09:21:44.000+0100},
title = {Graph Classification with 2D Convolutional Neural Networks},
url = {http://arxiv.org/abs/1708.02218},
year = 2017
}