Jim E. Griffin’s Research Page

I am a Professor of Statistical Science in the Department of Statistical Science, University College London. You can find information about my publications from Google Scholar or Research Gate.

Research Interests

• Methodology: Bayesian Nonparametrics, Regression Modelling with High-Dimensional Data, Time Series Modelling and Forecasting
• Applications: Economic and Finance Modelling, Performance Modelling of Elite Athletes, Student Performance and Engagement

Research Projects

• Bayesian Modelling of Macroeconomic Data - This is a joint project with Maria Kalli looking into flexible modelling of macroeconomic time series. Recent work includes Bayesian estimation of tensor vector autoregression models, Time-varying Factor Augmented Vector Autoregressions with Grouped Sparse Autoencoders and Time-varying tensor vector autoregression with Yiyong Luo (UCL), and Dynamic model averaging with general dependence with Dawid Bernaciak (UCL).
• Athletic Performance Passport - This is a joint project with James Hopker identifying unusual changes in the performance of elite athletes to inform anti-doping. Recent work icludes a Bayesian hieraarchical model for performance evolution and an evaluation of performance as an indicator of doping status.
• Efficient Computational Methods for Bayesian Variable Selection - This work involves Sam Livingstone and Xitong Liang. This work looks at efficient computational methods for Bayesian variable selection in high-dimension problems and applications in areas such as genomics. Recent work includes samplers for linear models, samplers for generalized linear models and survival analysis, and our recent NeurIPS paper with Alberto Caron looking at samplers for graphical models.
• Expressing and Visualizing Model Uncertainty in Bayesian Variable Selection - This work develops methods for understanding model uncertainty arising in Bayesian variable selection. Initial work looks using Cartesian credible sets.
• Bayesian Nonparametric Methods - Recent works includes flexible models for survival data, factor decompositions of many densities, modellling open wildlife populations using Polya trees and an efficient general purpose implementation of the Ferguson-Klass algorithm.
• Statistical Modelling of Environmental DNA (eDNA) - This is a joint project with Eleni Matechou and Alex Diana. We have worked on Bayesian variable selection for model with false positives and false negative for single species eDNA data. An Rshiny app is available. We have also developed a models for qPCR data and multi-species model for biomass changes allowing for errors in eDNA data.

Preprints

Some of my current and previous preprints are available from arXiv.org

Code

• Matlab app for the Bayesian nonparametric vector autoregression model in the Journal of Econometrics paper.
• Matlab code for the adaptive Monte Carlo methods in the Biometrika paper.
• Matlab code and R code for finding the Cartesian credible sets in “Expressing and visualizing model uncertainty in Bayesian variable selection using Cartesian credible sets”.
• Code for methods developed in the “Efficient Computational Methods for Bayesian Variable Selection” project are available from Xitong Liang’s GitHub page.
• Code for “A loss discounting framework for model averaging and selection in time series models” is available from Dawid Bernaciak’s GitHub page.
• R code for “Modelling between- and within-season trajectories in elite athletic performance data”.