Information Sheet
EE372: Quantization, Compression, and Classification
spring 2006 - 2007


April 4, 2007

Handouts
Handouts will be made available at http://www.stanford.edu/class/ee372/handouts.html as they are produced. These include lecture notes, copies of the slides used in class, homework, a pdf version of this information sheet, and a discussion of the class project.
Instructor:
Robert M. Gray
Office: Packard 261
Phone: (650) 723-4001 (ISL), 723-6685 (EE)
rmgray at stanford.edu
Office hours: M 1:00-2:30, W 10:-11:30 or by email appointment.
Teaching Assistant
Sangho Yoon
Office: Packard 109
Phone: TBA
Office hours: TBA
e-mail: holyoon at stanford.edu
Office Hours: TBA
Grader:
TBA
Admin. Ass't.
Kelly Yilmaz
Packard 259
(650) 723-4539 yilmaz@stanford.edu
Lectures
MW 3:15-4:30, 60-61H (time may be shifted slightly to remove conflict)
Problem Sessions
Weekly sessions to discuss homework and projects. TBA
Course Objective
Develop the fundamentals of quantization and compression and quantization-based classification.
Course Description
Theory and design of codes for quantization and signal compression systems (source coding systems), systems which convert analog or high bit rate digital signals into relatively low bit rate digital signals. Applications to analog-to-digital conversion, source coding and data compression, statistical classification, clustering, fitting discrete models to continuous models, density estimation, and machine learning. Necessary conditions for optimality of codes and implied code design algorithms. Constrained and structured vector quantization.
Intended Audience
Electrical engineers and computer scientists interested in lossy and lossless compression and statistical classification, especially of speech and images.
Prerequisites

Tentative Course Topics

The topics are likely to be closely approximated, but not necessarily in the same order. The lectures will begin with an introductory survey of most of the basic ideas in order to frontload enough of the material to begin looking for projects and studying the literature. Topics will be then treated individually in more depth. The introductory tour will be covered in handout lecture notes.
Texts
Primary Texts
Other texts of possible interest Occasional notes will be handed out and posted to the Web.
Some relevant Web links
Course Requirements and Grading

4 Homework sets 35%
Midterm 30%
Final Project 35%
The midterm is tentatively scheduled for Thursday, 17 May 2006 from 6:30 PM to 9:30 PM. It will cover material from the first 11 lectures. The room on campus will be announced. The course project will consist of either a theoretical analysis of a quantization or compression system or the design, programming, and simulation of a signal compression/quantization/classification algorithm for a particular application (or some combination of the two). The project should be described in readable English in a report. There will also be a 15 minute oral presentation during the final week of the quarter before finals. The report should provide appropriate references to the literature and a comparative discussion with existing methods. Suggestions for projects may be found at projects.html, but creativity in developing a topic will be considered in the grade. The projects can be developed for any available platform. The project grade will be based on the creativity and technical content of the project and on the quality of the presentation and participation in the discussion of the other project presentations. A 1 to 3 page proposal for the project should be submitted prior to Wednesday 2 May describing the basic problem to be attacked, the general approach, and a list of relevant references. In previous years some of the class projects have turned into conference publications, including the IEEE Data Compression Conference and the IEEE International Conference on Image Processing. Students are encouraged to develop a project that has some relation to their own research interests. A list of suggested projects will be handed out the second week of class.
Collaboration Policy
You are encouraged to discuss together the concepts presented in the class and the book. You may also discuss the homework problems, and help each other if someone gets stuck. Projects may be done by pairs, but individual responsibilities must be clearly described and separate reports submitted. You are required to implement new code by yourself, work out written problems by yourself, and turn in only your own work. Specifically, you may not copy anyone else's computer files or copy anyone else's written homework. If you choose to share subroutines, then these subroutines should be available to anyone in the class, e.g., via a web page. The goal is that each student should be able to give and receive sufficient help so that each person can complete the homework, but that in the end each student is turning in a homework that he or she understands fully and would be able to reproduce in its entirety unaided. Obviously it is difficult to write a completely specific collaboration policy; you are asked to abide by the spirit, rather than the letter, of this one.
Image Systems Engingeering Program Lab
Computing facilities will be made available at in the Image Processing Lab
http://scien.stanford.edu/labsite/scien_teaching_lab.html
for use in the homeworks and projects.


rmgray@stanford.edu