University of Cambridge > > Inference Group > MCMC for doubly-intractable distributions

MCMC for doubly-intractable distributions

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Ryan Prescott Adams.

“The model is intractable so we resort to Markov chain Monte Carlo” has become a standard mantra in the Bayesian statistics community. But (even for the patient) standard MCMC techniques are not a panacea — for example they can not sample from the parameter-posterior of a large tree-width undirected graphical model.

We recently (Proc. UAI 2006 ) introduced a valid MCMC scheme for this problem. Our exchange algorithm is simpler and often performs better than the only direct competitor (Møller et al., Biometrika 93(2):451–458, 2006). Although both require expensive exact sampling (Propp and Wilson, Rand. Struct. Alg. 9(1&2):223–252 1996).

In this talk I give a simpler derivation of the exchange algorithm. I also discuss the extent to which exact sampling is required and the implications for probabilistic modeling with undirected graphs.

This is work with David MacKay and Zoubin Ghahramani.

This talk is part of the Inference Group series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2023, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity