About Me

my photo

I am a DPhil candidate in the Department of Statistics at the University of Oxford. I am also part of the CDT programme OxWaSP between Oxford and Warwick. My supervisor is Professor Chris Holmes.

I am a member of St Peter's College and a Clarendon scholar.

My research interests lie around Bayesian statistics and Markov Chain Monte Carlo (MCMC) methods.

In autumn 2019 I was visiting Pierre Jacob and his research group at Harvard University.

During summer 2019 I was an intern in the Statistical Methodology & Consulting team at Novartis, Basel, working on modelling clinical trials data.



  • E. Pompe, C. Holmes, K. Latuszynski. A Framework for Adaptive MCMC Targeting Multimodal Distributions. (to appear in the Annals of Statistics) arXiv
  • E. Pompe, C. Holmes. Discussion on Unbiased Markov chain Monte Carlo methods with couplings by Jacob et al. (preprint)

Talks and poster presentations

  • (Upcoming) 21st of January 2020, talk at the Stochastic Simulation Seminar, Oxford Mathematical Institute.
  • January 2020, talk at BayesComp 2020.
  • November 2019, Harvard University, Department of Statistics, talk A Framework for Adaptive MCMC Targeting Multimodal Distributions.
  • October 2019, Novartis Pharma AG, Basel, talk Joint analysis of individual components of composite endpoints with case study on ACR50 endpoint in Psoriatic Arthritis.
  • July 2019, European Meeting of Statisticians, talk A Framework for Adaptive MCMC Targeting Multimodal Distributions. slides
  • November 2018, University of Warwick, Department of Statistics, talk Adaptive MCMC for Multimodal Distributions.
  • July 2018, LMS Invited Lecture Series and CRISM Summer School in Computational Statistics 2018 poster presentation. poster
  • June 2018, 2018 ISBA World Meeting poster presentation.
  • March 2018, BayesComp 2018 poster presentation.


Bayesian hypothesis testing

Bayesian hypothesis testing for survival analysis

Ongoing work with J. Rousseau and C. Holmes on a new approach to hypothesis testing based on mixture models, which could be an interesting alternative to the popular Bayes factors.

Multimodal distribution

Adaptive MCMC for Multimodal Distributions

We propose a new MCMC-based adaptive approach to sampling from multimodal target distributions and prove its ergodic properties (joint work with C. Holmes and K. Latuszynski).

Hidden Markov model

Curse of dimensionality for Sequential Monte Carlo methods versus Gibbs sampling

We show theoretically and empirically that the Gibbs sampler outperforms SMC samplers in many settings by overcoming the curse of dimensionality (with P. Rebeschini and G. Morina).


Inconsistency of the Dirichlet process mixtures for the number of components

[Harrison & Miller, 2014] show that when data comes from a finite mixture, the posterior distribution of the number of clusters does not concentrate at the true value under the Dirchlet prior. We investigate these results and generalise their certain aspects (joint work with J. Noble).


Before joining Oxford, I was working as a teaching assistant in Mathematical Analysis at the University of Warsaw (for 4 semesters).
I was also in charge of organizing two editions of a nationwide math contest for students aged 13-16 in Poland, with over 500 participants each time.

Non-work activities

Since 2016 I have been an active member of Oxford University Dancesport Club (OUDC). In the academic year 2018/2019 I joined Oxford University Strategic Studies Group (OUSSG).

I also attend Oxford R User Group meetings.

Contact me


emilia.pompe at stats.ox.ac.uk


Office 1.14
Department of Statistics
University of Oxford
24-29 St Giles
Oxford OX1 3LB, UK