Exploring Incomplete Data in Probabilistic Graphical Models


  • Exploring Incomplete Data in Probabilistic Graphical Models

Exploring Incomplete Data in Probabilistic Graphical Models

Principal Supervisor: Dr. Cassio P. de Campos

Second Supervisor: Prof. Adele Marshall


+ Project Description

Missing values are present in many types of data from a wide range of disciplines. Their treatment is a very common problem in statistical analysis and machine learning. Recent advances with models that use numerous hidden variables such as neural networks and sum-product networks have shown surprisingly accurate classification results in many domains, including image and video processing, automatic game playing, etc. 

This project regards the study of existing methods to deal with incomplete data and the development of new approaches to create and to explore the vast number of possibilities for probabilistic graphical models with hidden variables. The goal is to build novel methods for Bayesian networks and Markov random fields that make use of hidden variables to improve classification performance. The study will start with the current ideas for the treatment of the missing data and will propose new approaches using hidden variables for learning such models that will achieve state-of-the-art results and will potentially surpass these trendy models in some particular domains.

+ How to Apply

Applicants should apply electronically through the Queen’s online application portal at: https://dap.qub.ac.uk/portal/

+ Contact Details

Supervisor Name: Dr. Cassio P de Campos

Queens University of Belfast
School of EEECS
18 Malone Road, CSB 03.026

Email: c.decampos@qub.ac.uk



Tel: +44 (0)28 9097 6795