The pseudo-marginal approach to “exact approximate” MCMC algorithms

Hello,

lately in my research I have been dealing with noisy likelihoods. What I have learned with my testings is that MCMC works fine, if the noise level in the likelihood is low enough, but it gets stuck, if the noise level is too high.

Now I have learned from friends and the literature that MCMC works exactly correct with these noisy likelihoods. This approach is called the pseudo-marginal MCMC and it was first introduced by Mark Beaumont. Cool.

Here I would like to advertise a neat blog by  Darren Wilkinson that explains this pseudo-marginal idea in a way that I understand it too.

Check it out!

All the  best,

Janne