Network motif prediction using generative models for graphs

Graphs are commonly used to represent pairwise interactions between different entities in networks. Generative graph models create new graphs that mimic the properties of already existing graphs. Generative models are successful at retaining the pairwise interactions of the underlying networks but often fail to capture higher-order connectivity patterns between more than two entities. A network motif is one such pattern observed in various realworld networks. Different types of graphs contain different network motifs, an example of which are triangles that often arise in social and biological networks. Motifs model important functional properties of the graph. Hence, it is vital to capture these higher-order structures to simulate real-world networks accurately. This thesis introduces a motif-targeted graph generative model based on a generative adversarial network (GAN) architecture that generalizes and outperforms the current benchmark approach, NetGAN, at motif prediction. This model and its extension to hypergraphs are tested on real-world social and biological network data, and they are shown to be better at both capturing the underlying motif statistics in the networks as well as predicting missing motifs in incomplete networks.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Gamarallage, Anuththari

Contributors dc:contributor

• Milenkovic, Olgica

• motif prediction
• graph generative models

Statement dc:rights

• Copyright 2020 Anuththari Gamarallage

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/107976

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/107976