Network Signatures from Image Representation of Adjacency Matrices: Deep/Transfer Learning for Subgraph Classification

Kshiteesh Hegde, Malik Magdon-Ismail, Ram Ramanathan, Bishal Thapa

Abstract:

We propose a novel subgraph image representation for classification of network fragments with the target being their parent networks. The graph image representation is based on 2D image embeddings of adjacency matrices. We use this image representation in two modes. First, as the input to a machine learning algorithm. Second, as the input to a pure transfer learner. Our conclusions from multiple datasets are that
1. deep learning using structured image features performs the best compared to graph kernel and classical features based methods; and,
2. pure transfer learning works effectively with minimum interference from the user and is robust against small data.

TL;DR: We convert subgraphs into structured images and classify them using 1. deep learning and 2. transfer learning (Caffe) and achieve stunning results.

Keywords: deep learning, transfer learning, adjacency matrices, image feature representation, Caffe, graph classification

Data: [Gowalla](https://paperswithcode.com/dataset/gowalla)