skip to content

Asymptotics of Approximate Bayesian Computation

Presented by: 
Paul Fearnhead
Wednesday 5th July 2017 - 09:00 to 09:45
INI Seminar Room 1
Many statistical applications involve models for which are easy to sample from, but for which it is difficult to evaluate the likelihood. Approximate Bayesian computation is a likelihood-free method for implementing Bayesian inference in such cases. This talk will overview some recent results on the theoretical properties of approximate Bayesian computation which consider the performance of ABC as we get more data. It will cover questions such as: when does the ABC posterior concentrate on the true parameter value? What distribution does the ABC posterior converge to? And what is the frequentist distribution of point-estimates derived using ABC. It will also cover the impact of Monte Carlo error on estimates obtained using ABC, and consider whether, asympotically, it is possible to efficiently estimate parameters using ABC if we have a fixed Monte Carlo sample size.

This is joint work with Wentao Li: and; the talk will also cover work by David Frazier, Martin, Robert and Rousseau:
The video for this talk should appear here if JavaScript is enabled.
If it doesn't, something may have gone wrong with our embedded player.
We'll get it fixed as soon as possible.
University of Cambridge Research Councils UK
    Clay Mathematics Institute London Mathematical Society NM Rothschild and Sons