Aims and scope

The workshop bringing together, on the one hand, statisticians, computer scientists, and mathematicians and on the other hand applied researchers (especially focusing on epidemiology and public health) with a focus on three interrelated questions:

1. What is the current practice? How do applied researchers use Bayesian methods, what are the inferential questions they ask and what are the types of scientific and statistical conclusions they want to draw?

2. What is the methodological state-of-the-art? Which computational methods, e.g. for MCMC or approximate Bayesian inference, do the statisticians (and mathematicians and computer scientists) recommend in practice and why? Should theory guide the answer to these questions? Methodological insights? Under what framework? Is there a role for frequentist analysis of Bayesian methods? Proper scoring rules?

3. What challenges are there for the future? What problems can applied researchers not currently solve, and are they fundamentally statistical, computational, or philosophical? What problems are methodological and theory researchers working on, for which real, applied problems and datasets could be useful to motivate further development?

Invited speakers (confirmed)

• Nikolas Kantas, Imperial College London

• Francois-Xavier Briol, Imperial College London

• Michael Betancourt, Symplectomorphic, LLC

• Paul Büerkner, Aalto University, Finland

• Benjamin Lambert, Imperial College London

• Theresa Smith, University of Bath

Programme

Day 1 and Day 2 (22nd-23rd July) STAN Course

Monday

Lecture 9.30 - 11.00

-Introduction to inference; differences between Frequentist and Bayesian viewpoints; intuition behind Bayesian inference (how changes to priors and data affect inferences)

-Model validation through visual posterior predictive checks

-The difficulties of exact Bayesian inference

-Why discretisation and other similar deterministic approaches fail

-Conjugate Bayesian analysis

-Approximate inference: sampling as a remedy

-Independent sampling through rejection sampling, inverse transform sampling and importance sampling

-The issues with independent sampling for Bayesian inference

-Problem class

Dependent sampling and its costs: Markovian dice

- MCMC as a form of dependent sampler:

- Random Walk Metropolis

- Adaptive Metropolis

- Gibbs sampling (& how to derive a Gibbs sampling approach)

- Monitoring convergence of MCMC

- MCMC continued:

- Hamiltonian Monte Carlo

- No-U-Turn sampling

- Introduction to Stan language

- Differences versus BUGS / JAGS

- How to code up your first Stan model

- How to do posterior predictive checks

- Shiny Stan

-Problem class

Tuesday

-How to code up discrete parameter models in Stan

-How to fit ordinary differential equation models in Stan

- What to do when ODE models go wrong in Stan

-An introduction to hierarchical models and how to code these up in Stan

-Model comparison using WAIC, LOO-CV and true cross validation

-Troubleshooting in Stan / Bayesian inference in general:

- Fake data simulation

- Stan user discourse

- What sort of models can't be fit in Stan?

- Problem class

-Bring your own problems session

-Any residual problems from previous problem sets can be finished

- Individuals bring their own problems and we try to code them up!

Day 3, Day 4 and Day 5 (24th-26th July) Invited Speaker presentations and hands-on-tutorials

Bring your laptop!

Important dates:

• Registration deadline: 30 June 2019

FAQ

What do I need to bring?

Your laptop and charger.

Will be vegetarian options for lunch?

Yes, it will be mainly vegetarian.

Is my registration fee or ticket transferrable?

No, it is not possible to transfer tickets to other participants.

Is it ok if the name on my ticket or registration doesn't match the person who attends?

No the person who purchased the ticket is the one that will be attending the course.

How can I contact the organiser with any questions?

Please send an email to
a.freni-sterrantino[at]imperial.ac.uk

The workshop is co-sponsored by Imperial College Mathematics Quantitative Sciences Research Institute and the MRC-PHE