Location: Lecture Theatre 2, Appleton Tower, 11 Crichton St, Edinburgh EH8 9LE.

Time: Tuesday 12th November, Talk at 18:00 followed by refreshments (juice and biscuits)

Speaker: Dr Dennis Prangle, Newcastle University

Abstract:

Many complex statistical models do not allow evalation of the likelihood
function. This makes it hard to apply standard likelihood-based
inference methods, such as maximum likelihood and Markov chain Monte
Carlo. However it is often possible to simulate data from the model.
Approximate Bayesian Computation (ABC) is a family of methods which
bases statistical inference on such simulations.

In this talk I'll review the basics of standard ABC. The main idea is to
simulate data under many parameters and base the inference results on
those parameters giving close matches to the observed data. I'll discuss
the difficulties of scaling up ABC beyond low dimensional observations,
and some recent developments to avoid this problem, generally related to
density estimation. In particular I'll present some recent work I've
done on learning to control the simulation process so it concentrates on
producing data similar to the observations.