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

TL;DR

This paper introduces a novel image-based representation of network subgraphs using adjacency matrices, enabling effective deep learning and transfer learning approaches for classifying network fragments.

Contribution

It presents a new subgraph image representation method and demonstrates its effectiveness for classification using deep and transfer learning, outperforming traditional methods.

Findings

Deep learning with the proposed image features outperforms graph kernel and classical feature methods.

Transfer learning is effective with minimal user intervention and small datasets.

The approach is robust across multiple datasets.

Abstract

We propose a novel subgraph image representation for classification of network fragments with the targets 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 several datasets are that (a) deep learning using our structured image features performs the best compared to benchmark graph kernel and classical features based methods; and, (b) pure transfer learning works effectively with minimum interference from the user and is robust against small data.