about

research

My current research is on blending physical systems with observed data via Bayesian statistical methods. My PhD focused on extending the statistical finite element method (statFEM) to handle nonlinearity and time-dependence. The first work, establishing the methodology, was published here, and the code is available here. We have extended this work to scale to high-dimensional systems, as motivated by reaction-diffusion problems. This was published in the Journal of Computational Physics here, and the code is available here. We’ve done some further work on looking at Langevin samplers for statFEM — see here for the paper, and here for the code.

Prior to starting my PhD, I did my Bachelors and Honours in Computational Statistics (also at UWA). I worked on modelling rainfall with MCMC, under Edward Cripps and Michael Bertolacci.

My research interests include Bayesian Uncertainty Quantification/Inverse problems, Bayesian analysis of differential equations, and Markov chain Monte Carlo methods.

publications

Duffin, C., Branson, P., Rayson, M., Girolami, M., Cripps, E., Stemler, T., 2023. Exploring Model Misspecification in Statistical Finite Elements via Shallow Water Equations (preprint). (arXiv link) (code).

Glyn-Davies, A., Duffin, C., Akyildiz, Ö.D., Girolami, M., 2022. \(\Phi\)-DVAE: Learning Physically Interpretable Representations with Nonlinear Filtering (preprint). (arXiv link).

Akyildiz, Ö.D.\(^*\), Duffin, C.\(^*\), Sabanis, S., Girolami, M., 2022. Statistical Finite Elements via Langevin Dynamics. SIAM/ASA Journal on Uncertainty Quantification. (link) (code) (\(^*\)joint first authors).

Duffin, C., Cripps, E., Stemler, T., Girolami, M., 2022. Low-rank statistical finite elements for scalable model-data synthesis. Journal of Computational Physics 463. (link) (code).

Duffin, C., Cripps, E., Stemler, T., Girolami, M., 2021. Statistical finite elements for misspecified models. PNAS 118. (link) (code) (UWA media release) (Cambridge media release).

theses

Statistical finite element methods for nonlinear PDEs. Connor Duffin, May 2022, PhD thesis, The University of Western Australia. Available from (here) and also from UWA.

Modelling Australian daily rainfall with Bayesian mixture models. Connor Duffin, December 2018, Honours thesis, The University of Western Australia (available upon request).