This talk is about sampling from probability distributions in a challenging setting: we can only compute noisy (but unbiased) estimates of the probability density function at individual settings of variables we choose. Previous work on such "pseudo-marginal" methods have been used in inference problems in genetics, continuous time stochastic processes, hierarchical models, and "doubly-intractable" distributions. I'll present algorithms that are easier to apply than previous work and sometimes work a lot better.
Joint work with Matt Graham. [ Ссылка ]
rt of a symposium to celebrate the work of Professor Sir David MacKay FRS. The symposium was held over the period 14-15 March 2016.
[ Ссылка ]
![](https://i.ytimg.com/vi/HhGJZmn-5SU/maxresdefault.jpg)