After successful completion of the course, students are able to:
Understand the basics of generative, probabilistic (Bayesian) modeling and inference
Construct probabilistic models via an expressive probabilistic programming language
Explain how general purpose programming languages can be extended to support probabilistic constructs
Understand standard inference algorithms and their implementations in probabilistic programming languages (MCMC, Variational Inference, etc.)
Independently read literature in the probabilistic programming space
Probabilistic programming is a general framework to express probabilistic models as programs. It lies at the intersection of machine learning, statistics, and programming languages. While it has classically been seen as mechanization of Bayesian statistical inference, it has recently emerged as a candidate for next-generation AI toolchains.
This course will convey both theoretical and practical aspects of using probabilistic AI to express complex probabilistic models as programs and understand the interplay of modeling and inference to efficiently solve real world problems:
Generative (Bayesian) Models
Conditioning and Posterior Sampling,
Programmable Inference for Probabilistic Programming Languages
Deep Probabilistic Programs (Bayesian Neural Networks),
Inference Methods: Markov Chain Monte Carlo (MCMC), Hamiltonian Monte Carlo, Variational Inference
Applications of Probabilistic Programming
Regular lectures about theoretical topics
Individual assignments in probabilistic programming language
Independent group projects implemented in a probablistic programming language (Gen, Turing, Pyro, PyMC3, etc.) with final presentations
150 hours
5 x 2h Lecture = 10h
3 x 20h Individual Assignments = 60h
1 x 60h Group Project = 60h
Preparing Presentation = 10h
Attending Final Presentations = 10h
Assignments and independent project with final presentations
