Probabilistic Graphical Model  (Fall 2012, 3 credits)

 

Instructor: Prof. Shou-de Lin (sdlin@csie.ntu.edu.tw)

Classroom: CSIE 101

Meeting Time: Tu 2:20-5:20 pm

Office Hour:  Tu after class or by appointment

TA : Chung-yi Li (r00922051@csie.ntu.edu.tw), Ting-wei Lin (b97083@csie.ntu.edu.tw), En-hsu Yen (a061105@gmail.com)


Course Description:

 

Real-world events are full of uncertainty. Probabilistic Theory is a way to model uncertainty. However, probabilistic theory itself can hardly be exploited to deal with large-scale real-world problem which contains many correlated variables. Thanks to the development of a new reasoning and inference framework called probabilistic graphical models, we are then able to deal with thousands or even more variables altogether in a efficient manner. This course will cover the basic and advanced topics about probabilistic graphical models, including directed models such as Bayesian Networks, undirected models such as Markov network, and the corresponding inference and learning methods such as variable elimination, brief propagation, EM alrogihtm, and Markov chain Monte Carlo methods.

 

Grading:

        

Three Homework assignments (70%), three persons per team

Final Project (30%)

 

Textbook:

 

Bayesian Reasoning and Machine Learning, David Barber, Cambridge 2012 (pdf version available online)

Probabilistic Graphical Models Principles and Techniques Daphne Koller and Nir Friedman (ISBN  0-262-01319-3)

Recommend Readings:

 

- Elements of Statistical Learning by Trevor Hastie, Robert Tibshirani and Jerome Friedman (ISBN 0387952845)
- Pattern Recognition and Machine Learning by Chris Bishop (SBN 0387310738)
- Machine Learning by Tom Mitchell (ISBN 0070428077)
- The EM algorithm and related statistical models / edited by Michiko Watanabe, Kazunori Yamaguchi


 Syllabus:

Research Methods   First section (2:20-3:20) Chpaters HW
11-Sep   Introduction CH1~2  
18-Sep representation Bayesian Networks, PGM tool CH3 HW1-1
25-Sep representation BN, Case Study CH3  
2-Oct Inference Markov Networks CH4 HW1-2
9-Oct Inference Exact Inference CH5  
16-Oct Inference Exact Inference CH6 HW1 due HW2-1 out
23-Oct Inference Sampling as approximate inference CH27 HW2-2 out
30-Oct Learning Deterministic Approximate Inference CH28  
6-Nov Learning HW1 discussion, statistics for ML CH8.6~8.8,T HW2 due
13-Nov Learning  Learning as Inference (learning for BN), MLE, MAP, CH9.1~9.4 T,
 
20-Nov Learning MLE for undirected models, Naive Bayes CH9.6 CH10, C HW3 out
27-Nov Learning Learning with hidden variables (EM1) CH11  
4-Dec Learning Learning with hidden variables (EM2) CH11 HW3 due
11-Dec Learning Project Proposal
 
18-Dec Case study Structure learning & Bayesian Model Selection  CH9.5, 12.1~12.5 C  
25-Dec Case study Hidden Markov Model CH23  
8-Jan Final project Final project presentation