Advanced Bayesian Learning - PhD course, 8 credits

Course information

The typical participant is a PhD student in Statistics or related fields (Mathematical Statistics, Engineering Science, Quantitative Finance, Computer Science, ...). The participants are expected to have taken a basic course in Bayesian methods, for example Bayesian Learning at Stockholm University, and to have some experience with programming.

Examination and Grades: The course is graded Pass or Fail. Examination is through individual reports on distributed problems for each topic. Many of the problems will require computer implementations of Bayesian learning algorithms.

Course organization The course is organized in four topics, each containing four lecture hours. Course participants will spend most of their study time by solving the problem sets for each topic on their own computers without supervision.

All lectures are given physically at Campus Albano at Stockholm University. Lecture rooms will be announced later on.

Welcome!

Mattias Villani
Professor of Statistics, Stockholm University

---

Topic 1 - Gaussian processes regression and classification

Reading: Gaussian Processes for Machine Learning - Chapters 1, 2.1-2.5, 3.1-3.4, 3.7, 4.1-4.3.
Code: GPML for Matlab | GPy for Python | Gausspr in R | Gaussianprocesses.jl in Julia | GPyTorch - GPs in PyTorch
Other material: Visualize GP kernels

Lecture 1 - April 16, hours 10-12
slides
Lecture 2 - April 16, hours 13-15
slides

Lab Topic 1
Problems | Lidar data

---

Topic 2 - Mixture models and Bayesian nonparametrics

Reading: Bayesian Data Analysis - Chapter 23 | The Neal (2000) article on MCMC for Dirichlet Process Mixtures

Widgets: Dirichlet distribution | Bayes for multinomial data | mixture of normals | mixture of Poissons

Lecture 3 - April 29, hours 10-12
slides
Lecture 4 - April 29, hours 13-15
slides | derivation marginal Gibbs

Lab Topic 2
Problems | Galaxy data

---

Topic 3 - Variational inference

Reading: Blei et al JASA | Tran's VI Notes
Other material: Kullback-Leibler widget | My recent talk with some VI | Natural gradient notes | autograd in python | ForwardDiff in Julia

Lecture 5 - May 13, hours 10-12
slides
Lecture 6 - May 13, hours 13-15
slides

Lab Topic 3
Problems | Time series data

---

Topic 4 - Bayesian regularization and variable selection

Reading: Sections 12.2-12.5 and 14.6 of Bayesian Learning book | Handbook chapter on variable selection | Article on Bayesian regularization

Lecture 7 - May 26, hours 10-12
slides
Lecture 8 - May 26, hours 13-15 slides

Lab Topic 4
Problems