Graduate Course · Winter 2025
This course studies modern methods for sampling from complex, multi-modal distributions at scale. Topics include parallel tempering, annealed importance sampling, non-reversible MCMC, sequential Monte Carlo, and their connections to generative modelling. Emphasis is placed on both theoretical foundations and practical implementations.
↓ Syllabus (PDF)Topics and schedule are subject to change.
| # | Topic | Slides |
|---|---|---|
| 1 | Introduction & Motivation | |
| 2 | MCMC Foundations | |
| 3 | Local Inference Algorithms | |
| 4 | Annealing | |
| 5 | Parallel Annealing: Parallel & Simulated Tempering | |
| 6 | Sequential Annealing: Annealed Importance Sampling & SMC Samplers | |
| 7 | Optimal Annealing Schedules | |
| 8 | Accelerating Annealing Algorithms & Neural Samplers | |
| 9 | Free Energy Methods |
The project involves an in-depth review of a method related to the course, and is expected to take approximately one week of effort. Deliverables are a 5-page write-up (excluding code) and a 30-minute presentation during the week of April 13 (date TBD).
Students may also propose their own topic in consultation with the instructor.