Graph Classification with 2D Convolutional Neural Networks

Anonymous

Nov 03, 2017 (modified: Nov 03, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: Graph classification is currently dominated by graph kernels, which, while powerful, suffer some significant limitations. Convolutional Neural Networks (CNNs) offer a very appealing alternative. However, processing graphs with CNNs is not trivial. To address this challenge, many sophisticated extensions of CNNs have recently been proposed. 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. Despite its simplicity, our method proves very competitive to state-of-the-art graph kernels and graph CNNs, and outperforms them by a wide margin on some datasets. It is also preferable to graph kernels in terms of time complexity. Code and data are publicly available.
  • TL;DR: We introduce a novel way to represent graphs as multi-channel image-like structures that allows them to be handled by vanilla 2D CNNs.
  • Keywords: graph classification, convolutional neural networks, 2D CNN, representation

Loading