Probabilistic Programming Workshop

Context

Probabilistic programming is a relatively new field on the intersection of machine learning, computer science and artificial intelligence. Probabilistic programming allows you to write any generative probabilistic model in a straightforward way, simply by describing the generative forward process.

Anglican is one of the start-of-the-art probabilistic programming language. It is reasonably fast and stable, and there is an active community supporting it. Since it is written in Clojure, that uses Java, it can be very easily run on every operating system and on almost any machine.

More information on Probabilistic Programming and Anglican can be found at
http://probabilistic-programming.org/wiki/Home
http://www.robots.ox.ac.uk/~fwood/anglican/

Workshop

This is a “hands-on” workshop or “hackathon” on Probabilistic Programming based on Anglican, consisting of 3 parts.

Part I - Introduction. Warm-up session to introduce probabilistic programing and how it represents generative models.

Part II - Programming skills. Provide the participants with an introduction to the basic skills to write probabilistic programs themselves in Anglican. This will consists on a series of set exercises that participants can follow them with the help of the facilitators.

Part III -  Applications. Each participant will have the opportunity to bring along and/or discuss his/her data and model, and discuss probabilistic programming implementations and predictions with the help of the facilitator and other participants.

Schedule

09.00– 10.00    Arrival refreshments.

10.00 – 10.30    Introduction.

                              Lecture on generative modelling

                              in probabilistic programming by Prof Frank Wood.

10.30 – 10.45    Tea / Coffee break.

10.45 – 13.00    Workshop part.

                              Four exercises: simple Gaussian posterior,

                              "7 scientists" problem,

                              finance model simulation, 2D physics simulation.

13.00 – 14.00    Lunch.

14.00 – 16.00    Hackathon part in 8 groups.

                              Four areas (each two groups work on one of areas):

                              1. Economics / Finance (time series, etc.),

                              2. Physics / Engineering,

                              3. Agents / Cognitive models,

                              4. Probabilistic programming theory.

16.00 – 16.15     Tea / Coffee break.

16.15  – 18.00    Hackathon continues.

The schedule is subject to change.

Target audience and prerequisites

Any scientist with data and the need to make predictions with that data. Basic skills in statistics and in basic programming are required, such as an understanding of the law of large numbers or how to write and debug simple programs in a programming language. There is no need to have an a-priori idea of what models are suitable for their data. You must bring your own laptop.

Facilitators

Yura Perov, Brooks Paige, Tuan Anh Le, Tom Rainforth, Frank Wood, Mike Wu, Jan Willem van de Meent, David Janz, Siddharth Narayanaswamy, Billy Smith

Local organisers

Ruben Sanchez-Garcia, Sri Dasmahapatra