CS 7636
Computational Perception

Spring 2004
Environmental Science and Technology Bldg. (ES&T) L1-125
(Note this is lower level 1, down a flight of stairs from the street)
MWF 2:00 - 3:00

Problem Sets        Syllabus        Projects

This course will cover probabilistic graphical models, a powerful class of statistical models based on graph representations. Our discussion will be motivated by research problems in computer vision which provide a rich and challenging context for graphical models. For some examples, see the introduction in a recent special issue of PAMI on graphical models in vision. Also see Kevin Murphy's excellent on-line tutorial.

Instructor

Jim Rehg
Email: rehg@cc.gatech.edu
Office: CoC Bldg (CCB) 253
Office hours: After class or by appointment
Phone: 404-894-9105 (email preferred)

Teaching Assistant

Charlie Brubaker
Email: brubaker@cc.gatech.edu
Office: CoC Bldg (CCB) 259A
Office hours: TBD

Prerequisites

Mathematics for Computational Perception and Introduction to Computer Vision (CS 4495/7495). Familiarity with probability, statistics, linear algebra, and some pattern recognition concepts are most important. Permission of the instructor.

Text

The primary text is An Introduction to Probabilistic Graphical Models by Michael I. Jordan. This text has not been published yet. We will be using draft copies of the chapters. Instructions on how to obtain these will be given in class (they cannot be downloaded from the web). If you wish to purchase a textbook, the following are recommended:

•  Learning in Graphical Models by M. Jordan (editor), MIT Press, 1999.
  An excellent collection of papers on graphical models, including topics such as variational inference.
• Bayesian Networks and Decision Graphs and An Introduction to Bayesian Networks by Finn V. Jensen, Springer, 2001 and 1996.
  Good overview of Bayesian networks
• Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference by Judea Pearl, Morgan Kaufman, 1988.
  Classic text on belief networks
• Information Theory, Inference, and Learning Algorithms by David MacKay, Cambridge University Press, 2003. [online]
  Excellent treatment of information theoretic methods in learning.

Background texts:

• Computer Vision: A Modern Approach by D. Forsyth and J. Ponce. Prentice-Hall 2002.
  A recent general computer vision text.
• Neural Networks for Pattern Recognition by C. Bishop. Oxford University Press, 1995. [Amazon]
  This is a classic text for density estimation and pattern recognition.
• Pattern Classification by R. Duda, P. Hart, and D. Stork. Wiley-Interscience, 2000. [Amazon]
  Another classic, recently brought up-to-date.
• Linear Algebra and Its Applications or Introduction to Linear Algebra by G. Strang. [Amazon]
• Matrix Reference Manual [online]
• Introduction to Probability by D. P. Bertsekas and J. N. Tsitsiklis. [online]
• Probability, Random Variables, and Stochastic Processes by A. Papoulis.
  Classic text for probability theory and its application. Also see on-line lecture notes for EE178 at Stanford.
• Probability Theory: The Logic of Science by E. T. Jaynes. [online]
  Classic text on probability theory, chapters 1 and 2 in particular are good background reading.
• Applied Optimal Estimation by A. Gelb (editor). MIT Press, 1974. [Amazon]
  Classic text for Kalman filter and linear estimation.

Help with Matlab

• Alex's Matlab tutorial (taken from CS7495 website).
• Ben's Matlab tutorial

Organization

Grades will be assessed as follows:

Problem sets will be distributed weekly or bi-weekly. Collaboration on problem sets is encouraged at the "white board interaction" level. That is, share ideas and technical conversation, but write your own code, do your own detailed derivations, etc. A few problem sets may require you to work in teams of 2-3. I plan to grade and return problem sets promptly. As a result, I will require all problem sets to be turned in on time.

No late submissions will be accepted without prior permission of the instructor. If you need an extension, let me know in advance.

I will not be grading undergraduate and graduate students differently in this course.

Problem Sets

PS 1 [pdf]: Out Jan 6; Due Jan 12 (Review of background material)  (4% of grade)
PS 2 [pdf, BNT_example.m, poker.m, showCPT.m]: Out Jan 22; Due Feb 2 (Conditional independence and d-separation) (7% of grade)
PS 3 [pdf]: Out Feb 20; Due Feb 27 (Learning and EM) (8% of grade)
Midterm [pdf]: Out Mar 1; Due Mar 3 (Review) (15% of grade)
PS 4 [pdf, zip, digits] (HMM's) Out Mar 19; Due Apr 2 (8% of grade)

Syllabus

1. Introduction to graphical models in vision
   • Papers: Graphical Models Overview, Introduction to PAMI SI
   • Overview of graphical models in vision [1/5/04, 1/7/04]
2. Basic definitions and concepts
3. Conditional independence and d-separation
4. Modeling
   • Resources: BNT usage guide (old version, better for PS 2)
   • Papers:

Final Projects

Swiki for final projects - Create an entry for your project here by Monday night. Link to your slides by Tuesday 11:30

Schedule

• Project Proposals Due - April 13
• Project Presentations - April 27, 11:30 - 2:20 (ES&T L1125)
• Project Reports Due - April 30

Possible Projects

• Face Recognition: [pdf] [empca, facedata, PCA tutorial, Tipping-Bishop paper, Moghaddam-Pentland paper]
• Epitome: [pdf] [paper and code, Image Quilting]
• Learning Human Motion [pdf] [epmt, mocapdata]
• Your own project...

Project Reports

Project reports must be linked from the swiki and must be available for reading by midnight, Friday April 30. If your report is not available when I start reading them on Saturday, then you will lose that portion of your grade. To avoid any mishaps with the swiki, I suggest that you double-check that your report is available once you've linked it by clearing your browser cache and reloading the swiki page. You may submit your report in either pdf or Word format.

Your project report should address the following:

• Participants and title
• A description of the contribution of each participant to the project (you must include this part).
• Description of the problem and why it is difficult.
• Description of the approach
• Results
• Analysis of the results (This is the most important section of the report. How well did your approach work? What issues did you encounter? How did you address them? What should be done differently or carried further if you were to do more work on this problem?)

I expect the reports to be between 5 and 10 pages. Shorter reports are better, just be sure to address all six of the points listed above. I am particularly interested in your analysis of your results. Please include example images, illustrations, etc. so that it is clear what you are describing.

If you created videos for your project and you link them on the swiki, then I will view them as part of reading your report.
I am very interested in seeing your videos!