Macquarie University
01whole.pdf (839.35 kB)
Download file

Minimum entropy principle guided graph neural networks

Download (839.35 kB)
posted on 2022-11-29, 00:36 authored by Zhenyu YangZhenyu Yang

Graph Neural Networks (GNNs) have become the mainstream way of learning graphstructured data consisting of nodes and edges. Generally, GNNs encode nodes and graphs to lower-dimensional vectorized representations for serving node-level and graphlevel downstream tasks, respectively. However, the dimension estimation issue (i.e., estimating the optimal dimension for representations) has been ignored by existing GNNs. Inappropriate representation dimensions will lead to sub-optimal performance of GNNs. To estimate the optimal representation dimensions for both node-level and graph-level representations, we propose a minimum entropy principle-guided dimension estimation framework MEDE. By considering the graph attribute and structure together, we carefully define the graph entropy of a graph. The optimal representation dimension of a single graph and its nodes can be obtained by minimizing the graph entropy. In addition, for graph representation learning involving multi-graphs, MEDE empowers GNNs to embed graphs into a candidate set of optimal graph representation dimensions, and each graph will be assigned a best-fit representation dimension. Experiments on the node and graph classification tasks and the network embedding task verify the effectiveness of minimum entropy principle-guided representation dimension estimation. 


Table of Contents

1 Introduction -- 2 Literature review -- 3 Preliminaries -- 4 Minimum Entropy principle guided Dimension Estimation framework (MEDE) -- 5 Experiments -- 6 Conclusion and future works -- References


A thesis submitted to Macquarie University for the degree of Master of Research

Awarding Institution

Macquarie University

Degree Type

Thesis MRes


Thesis (MRes), Macquarie University, School of Computing, 2022

Department, Centre or School

School of Computing

Year of Award


Principal Supervisor

Jia Wu

Additional Supervisor 1

Jian Yang

Additional Supervisor 2

Hao Peng


Copyright: The Author Copyright disclaimer:




72 pages