Abstract

Probabilistic programming makes machine learning modular by automating a common component called inference. Recent work by Shan and Ramsey makes inference itself modular by automating a common component called conditioning. Their technique, however, is limited to conditioning a single scalar variable. In this talk we will see how to extend their algorithm to perform symbolic conditioning on arrays. This is a practice talk for ICFP, and will touch upon topics in functional programming, program manipulation, and equational reasoning, with a sprinkling of Bayesian analysis.