[Pl-seminar] 4/10 Seminar: John Tristan: Compilation of Simple Probabilistic Programs to Gibbs Sampling

[Daniel Patterson]

NUPRL Seminar Presents

John Tristan
Oracle Labs

11:45AM
Tuesday, April 10th, 2018
Room 366 WVH (http://www.ccs.neu.edu/home/wand/directions.html)

Compilation of Simple Probabilistic Programs to Gibbs Sampling.

Abstract

One of the several interesting challenges of probabilistic programming
is that of compiling probabilistic programs to inference algorithms. One
of these inference algorithms, Gibbs sampling, is particularly relevant
because it is often statistically efficient, but unfortunately, it is
difficult to derive and therefore compile to.

In this talk, after a brief explanation of probabilistic programming and
why its relevance to data science, I will explain some of the ideas
behind the design of a compiler from (very) simple probabilistic
programs to Gibbs sampling. I will also attempt to explain what it would
mean for such a compiler to be correct.

Bio

Jean-Baptiste Tristan is a researcher in the machine learning group at
Oracle Labs. He holds a Ph.D. in Computer Science from the French
Institute for Research in Computer Science and Automation (INRIA) and a
M.Sc. in Computer Science from the Ecole Normale Superieure of Paris.