COMP 6931A: Spring 2012

Probabilistic Models for Unsupervised Learning

 

 

Time and Venue: Monday, 9:30am-11:50am, Room 3401

 

A tentative assignment of topics is shown below. More topics will be added later. Please let the instructor know if you have some topics in mind. There might be time at the end for some students to present a second topic. Let me know if you are interested.

 

The lecture on each topic should take at least 3 hours, and it can take more. The presenter should have the notes and related materials sent to lzhang@cse.ust.hk by Friday, 3 days before the class.  The materials (key papers) should be packaged into a zip file.

 

 

Week

Date

Topic

Presenter

Notes

paper

Week 5

04/03

Variational inference for LDA

Zhao Zhou

Slides

paper

Week 6

11/03

Correlated topic models

Yuan Yuan

 Notes

paper

Week 7

18/03

Hierarchical topic models

Yang Hai Yan

slides

papers

Week 8

25/05

Factor models and Indian buffet process

Chen Peixian

Slides

paper

Week 9

08/04

Distance dependent non-parametric models

Wang Nai Yan

slides

1  2

Week 10

22/04

A survey on click modeling

Lianghao Li

slides

 

Week 11

29/04

No class

Week 12

06/05

Deep learning

Wang Nai Yan/ Yuan Yuan

 

 

Week 13

13/05

LDA for supervised/semi-supervised learning

Zhao Zhou

slides

paper

Week 13

13/05

 Sparse factor models

Chen Peixian

 

 

 

 

Please prepare lecture notes using this template.

 

=== First Message on the course ===

The course is a continuation of COMP 5213 that was offered in Fall 2012. Topics include: Variational inference for LDA, correlated topic models, hierarchical topic models, factor models, Indian buffet process, and recent developments in related areas.

 

Each participant is expected to pick up one topic, identify the relevant materials, study them thoroughly, write up detailed lecture notes, and explain the materials to the class in depth and in details.  You can use my DPMM and LDA notes as examples when preparing yours.  For uniformity, please use the template provided in the first attachment.  The second attachment is a very recent paper by Blei. It is a good starting point for identifying LDA papers.

 

The target audience should be year 1 graduate students who have just obtained their bachelor degrees and who have taken 5213. Explain all the background that the audience might not be familiar with. You are welcome to come and discuss with me matters about the course.