A Structural Learning Method for Graphical Models

Szili, B., Niu, M. and Neocleous, T. (2022) A Structural Learning Method for Graphical Models. In: 4th International Conference on Statistics: Theory and Applications (ICSTA'22), Prague, Czech Republic, 28-30 Jul 2022, p. 113. ISBN 9781990800085 (doi: 10.11159/icsta22.113)

Full text not currently available from Enlighten.


This work is centred on investigating dependencies and representing learned structures as graphs. While there are a number of methods available for discrete and Gaussian random variables, there is no such method readily available for continuous variables that are non-Gaussian. For such methods to be reliable, it is necessary to have a way to measure pairwise and more importantly, conditional independence. In this work, an algorithm is created that uses both mutual information and a kernel method together to account for these dependencies and yield a graph that represents them. This method is then demonstrated through a simulation setting, comparing the results to an algorithm often used in Gaussian settings, additionally discussing future steps regarding this project.

Item Type:Conference Proceedings
Additional Information:The authors thank the School of Mathematics & Statistics at the University of Glasgow for the funding of this research. Mu Niu acknowledges support for this paper from EPSRC grants EP/W021595/1.
Keywords:Structural learning, mutual information, kernel methods, independence.
Glasgow Author(s) Enlighten ID:Szili, Benjamin and Neocleous, Dr Tereza and Niu, Dr Mu
Authors: Szili, B., Niu, M., and Neocleous, T.
College/School:College of Science and Engineering > School of Mathematics and Statistics > Statistics
Research Group:Statistics & Data Analytics
Related URLs:

University Staff: Request a correction | Enlighten Editors: Update this record