Loading…
Attending this event?
View analytic
Saturday, November 17 • 10:40am - 11:00am
Motivating Probabilistic Programming

Sign up or log in to save this to your schedule and see who's attending!

Probabilistic Programming is a well established field within statistics and machine learning and over time researchers have noticed a few things: - it’s quite easy to construct a generative model; and - these models have the same kind of structure; but - writing an inference algorithm for each particular model is slow and error-prone. Wouldn’t it be great if we could derive an inference algorithm given a model description? We can think of this an analogous to writing in assembly versus a high-level language. When we write assembly we optimize everything by hand, implementing our own control structures and so on tailored to our specific problem. This is the situation with creating a custom inference algorithm for a particular generative model. When we write in a high-level language we use predefined constructs that the compiler translates into assembly for us. We gain a lot of productivity by giving up a little bit of performance (which we often don’t miss). This is the goal with probabilistic programming.

Speakers
avatar for Rahul Chitturi

Rahul Chitturi

Principal Engineer, Coatue


Saturday November 17, 2018 10:40am - 11:00am
data


Twitter Feed