Staff Profile

Bayesian hierarchical models are useful for applications where the latent variables are usually of interest. One example of my current research is on stochastic block models for networks, where the goal is to uncover the underlying groupings of the nodes. The models are essentially mixture models with the network itself being the data. Another example of my research is on econometrics models for house prices, where the goal is to infer the house price index based on prices of houses (which are naturally heterogeneous goods) observed at different and irregular time points. As my work focuses on both building a useful model and carrying out the inference, efficient and scalable computational algorithms are equally important. While I use Markov chain Monte Carlo (MCMC) routinely, other methods are currently being explored.

I am also interested in studying the degrees of the networks, especially on the (mis-)use of the discrete power law and the incorporation of extreme value theory. While the power law captures the essence of most of the data, the right-hand tail is usually ill-fitted, but there is where extreme value theory comes to the rescue as, by definition, it deals with outliers and extreme observations.

Here are the PhD opportunities I am currently advertising:

1. What is my local area? Finding the relevant spatial scale: this is a project under the Geospatial Systems Centre for Doctoral Training (CDT), and co-supervised with Freegle. The project application reference is GEO23_20. 
   
2. Scale-freeness and growth stability of realistic network models: this is a standalone project, focusing on a network generative model that leads to a degree distribution described by a mixture of the power law and an appropriate extreme value distribution.
   

My profile on Google Scholar, ResearchGate, ORCID are as linked. 

• Bayesian hierarchical models
• Computationally intensive statistics (e.g. Markov chain Monte Carlo)
• Network modelling (e.g. preferential attachment models, stochastic block models)
• Extreme value theory for discrete data (e.g. degree distribution)

R packages:

• crandep

• Lee C, Wilkinson DJ. A hierarchical model of non-homogeneous Poisson processes for Twitter retweets. Journal of the American Statistical Association 2020, 115(529), 1-15.
• Lee C, Garbett A, Wilkinson DJ. A network epidemic model for online community commissioning data. Statistics and Computing 2018, 28(4), 891-904.
• Garbett A, Chatting D, Wilkinson G, Lee C, Kharrufa A. ThinkActive: Designing for pseudonymous activity tracking in the classroom. In: CHI '18: Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. 2018, Montreal, Canada: Association for Computing Machinery.