University of Cambridge > Talks.cam > Machine Learning @ CUED > Turing: Rejuvenating Probabilistic Programming in Julia

Turing: Rejuvenating Probabilistic Programming in Julia

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

If you have a question about this talk, please contact Louise Segar.

In this talk, I will present a probabilistic programming system (PPL) called Turing. The implementation of Turing is based on Julia, a fast and modern programming language for technical computing. Novel aspects of this system include 1) the use of coroutines for scalable implementation of existing general-purpose inference algorithms, and 2) new syntax features that improve statistical efficiency of general-purpose inference engines. I will discuss some lessons we have learnt so far, on both PPL design and novel Monte Carlo methods.

More concretely, I will first review a (classical) result that some widely used importance sampling methods (e.g. IS, PMC , SMC) are still valid when exact importance weights are replaced with non-negative unbiased Monte Carlo estimates or pseudo marginals. This result allow us to export the celebrated pseudo-marginal method from the MCMC framework to importance sampling methods. Then, following this generic result, we develop a pseudo-marginal particle filtering (PM-PF) method and apply it to PPL inference. Some experiments show the PM-PF method is consistently more accurate than similar algorithms: particle filtering, particle Gibbs.

This talk is part of the Machine Learning @ CUED series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2020 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity