cv

Statistician with experience building interpretable models for complex observational data, and the software to fit them at scale. I currently lead the modelling and engineering behind the Biodiversity Intactness Index (BII) in the Biodiversity Futures Lab at the Natural History Museum, London.

I have a PhD in statistics from the University of Western Australia and was a postdoc at the University of Cambridge, where I developed Bayesian methods for combining mechanistic models with data when those models are known to be wrong.

Statistical modelling

Hierarchical models for large observational datasets. Lead the mixed-effects modelling behind the BII: over 4 million observations from more than 35,000 sites across many independent studies, projected globally with uncertainty.
predictsr · ssmooth
Inference under model misspecification. Developed the statistical finite element method, a Bayesian framework that embeds a Gaussian process discrepancy term in a mechanistic model so that inference stays honest when the model is wrong.
PNAS, 2021 · paper
Scalable Bayesian computation. Built low-rank approximations and Langevin-based sampling schemes that make posterior inference tractable for high-dimensional models.
J. Comput. Phys., 2022 · paper · SIAM/ASA JUQ, 2022 · paper
Learning from unstructured data. Co-developed physics-informed variational autoencoders that link raw, unstructured observations to interpretable latent dynamics.
J. Comput. Phys., 2024 · paper

Experience

Senior Research Software Engineer — Natural History Museum, London (2026–present)
- Lead the statistical modelling and technical development of the BII, a global biodiversity indicator built on mixed-effects models from over 4 million observations, from over 35 thousand sites from the PREDICTS database.
- Own the model specification, validation, and uncertainty propagation from site-level data to global projections.
- Maintain the BII’s R packages and HPC/AWS deployment, and work with ecologists to turn research questions into models.
Research Software Engineer — Natural History Museum, London (2024–2026)
- Refactored the BII pipeline into tested R packages and deployed it on HPC.
- Scaled projections from 10 km to 1 km resolution.
- Published two CRAN packages (predictsr, ssmooth) and added CI, testing, documentation, and QA.
Postdoctoral Research Associate — University of Cambridge (2021–2024)
- Recruited to Mark Girolami’s group in the final year of my PhD; completed the thesis alongside the first year of the postdoc.
- Developed Bayesian inference methods (MCMC, Langevin dynamics, filtering) for complex physical models, with a focus on model misspecification and uncertainty quantification.
- Published in PNAS, SIAM/ASA JUQ, and the Journal of Computational Physics.
Teaching Assistant — University of Western Australia (2018–2020)
- Taught undergraduate and master’s statistics courses through tutorials, assessment setting, and exam marking.

Education

PhD in Statistics, Department of Mathematics and Statistics, University of Western Australia (2019–2022).
- Thesis: Statistical finite element methods for nonlinear PDEs.
BSc (Honours), Mathematics and Statistics, University of Western Australia (2015–2018).
- Honours thesis: modelling daily rainfall across Australia with mixture models.

Skills

Statistical methods: hierarchical/mixed-effects models, Bayesian inference and MCMC, mixture models, spatial statistics, uncertainty quantification, model misspecification and model criticism.

Languages and libraries:

R: lme4, Stan/brms, targets, sf, terra, ggplot2, testthat.
Python: JAX, numpy, scipy, matplotlib, mpi4py, pytest.
C++: Eigen, FEniCS, catch2.

Tools: Git, Docker, AWS (EC2/S3), HPC (Slurm, MPI).

Selected Software

predictsr: an R package to load the PREDICTS database into R dataframes and cache them to disk. Published on CRAN.
ssmooth: an R package for smoothing spatial data using statistical methods. Published on CRAN.

Publications

The statistical finite element method: A theoretical foundation for digital twins
Duffin, Connor; Glyn-Davies, Alex; Vadeboncoeur, Arnaud; Girolami, Mark.
Quality Engineering 38(1), 112–121 (2026).
Global indicator data tracking changes in species-level biodiversity persistence, ecosystem resilience under climate change, and protected-area representativeness and connectedness
Ware, Chris; Valavi, Roozbeh; Vickers, Mat; Giljohann, Kathryn M; Mokany, Karel; Purvis, Andy; Walkden, Patrick A; De Palma, Adriana; Duffin, Connor; Contu, Sara; et al.
EcoEvoRxiv preprint (2026).
Statistical finite elements via interacting particle Langevin dynamics
Glyn-Davies, Alex; Duffin, Connor; Kazlauskaite, Ieva; Girolami, Mark; Akyildiz, Ö Deniz.
SIAM/ASA Journal on Uncertainty Quantification 13(3), 1200–1227 (2025).
Probabilistic super-resolution for high-fidelity physical system simulations with uncertainty quantification
Zhang, Pengyu; Duffin, Connor; Glyn-Davies, Alex; Vadeboncoeur, Arnaud; Girolami, Mark.
arXiv preprint arXiv:2502.10280 (2025).
Φ-DVAE: Physics-informed dynamical variational autoencoders for unstructured data assimilation
Glyn-Davies, Alex; Duffin, Connor; Akyildiz, O Deniz; Girolami, Mark.
Journal of Computational Physics 515, 113293 (2024).
Exploring model misspecification in statistical finite elements via shallow water equations
Duffin, Connor; Branson, Paul; Rayson, Matt; Girolami, Mark; Cripps, Edward; Stemler, Thomas.
arXiv preprint arXiv:2307.05334 (2023).
Low-rank statistical finite elements for scalable model-data synthesis
Duffin, Connor; Cripps, Edward; Stemler, Thomas; Girolami, Mark.
Journal of Computational Physics 463, 111261 (2022).
Statistical finite elements via Langevin dynamics
Akyildiz, Ömer Deniz; Duffin, Connor; Sabanis, Sotirios; Girolami, Mark.
SIAM/ASA Journal on Uncertainty Quantification 10(4), 1560–1585 (2022).
Statistical finite elements for misspecified models
Duffin, Connor; Cripps, Edward; Stemler, Thomas; Girolami, Mark.
Proceedings of the National Academy of Sciences 118(2), e2015006118 (2021).