Special Course in Machine Learning: Unsupervised Learning
MTAT.03.317
Seminars: Thursdays 14:15, at Paabel (Ülikooli 17), room 218
Contact:
Mark Fishel, fishel@ut.ee, room 303
Kairit Sirts, kairit.sirts@ut.ee, room 218
Overview
This course introduces unsupervised machine learning methods based on probabilistic models. We will covers topics such as probabilistic graphical models, maximum likelihood estimation for latent variable models, non-parametric Bayesian models and approximate inference methods for such models. It is suitable for Master and Doctorate as well as advanced Bachelor level students interested in machine learning.
Process
The course starts with four lectures:
- Probability distributions, probabilistic graphical models (Mark Fishel)
- Inference methods: MLE/MAP estimation, EM-algorithm (Mark Fishel)
- Non-parametric Bayesian models: Chinese Restaurant Process, Pitman-Yor Process (Kairit Sirts)
- Methods for non-tractable inference: (mostly) Gibbs sampling (Kairit Sirts)
But this is just the introduction, after that it is your turn. There will be a list of topics with reading material. All students will read all materials before each seminar meeting. You will pick a topic and prepare test questions based on the reading material. Each seminar starts with taking a test that you or your fellow student designed. The rest of the seminar time will be used for discussing the test questions/results and the material.
Choosing your topic
Please choose three topics from the list of topics and send them by email in the order of preference to Mark or Kairit by Friday, September 16.
Grading
This is a pass/fail course. In order to pass you need to:
- Prepare test questions for at least one topic
- Obtain at least 51% from all tests
- Participate actively in the seminar discussions