ECE 532: Theory and Applications of Pattern Recognition

Prerequisites:

Competence in basic probability and statistics,

e.g., ECE 331 or Math 431

Instructor:

Robert Nowak

E-mail: nowak@engr.wisc.edu

Web: http://nowak.ece.wisc.edu

Phone: 608 265 3914

3627 Engineering Hall

Office Hours: after lecture, Tuesday and Thursday, 2:15-3:15pmLectures:

Spring 2005

Tuesday, Thursday 1:00 - 2:15 PM

2535 Engineering HallTextbook: Pattern Classification, by R. O. Duda, P. E. Hart and D. G. Stork, Second Edition, Wiley.Textbook webpage with additional information

Grading and Evaluation:

Midterm Exam: March 17, 25% of total course grade

Course Project: 30% (to be handed in on April 30)

Final Exam: May 8, 30% of total course grade

Homework & Course Participation: 15%

Project Teams and Web Reports:

Team 1 John Boehm and Minglei HuangTeam 2 Aarti Singh and Raman Arora

Team 3 Tulaya Limpiti and David Winters

Team 4 William L'Huilliler, Aline Martin and Ercan Yildiz

Homework Problems:Homework 1 (pdf)Homework 2 (pdf)Homework 3 (pdf)Homework 4 (pdf)Homework 5 (pdf)

Homework 6 (pdf)iris.mat (right-click to save)Homework 7 (pdf)

Homework 8 (pdf)Project Task 1 (pdf)

Homework 9 (pdf)

Project - Final Goals and Objectives (pdf)`Homework 10 (pdf)`

Keeping up with the course and participating in lectures (asking
questions)

is very important to successful learning. Keep your homework solutions
organized

in a folder or binder. I will ask you to turn in your solutions from
time to time

to see how you are keeping up with the coursework.

**Course Outline**:

1. Pattern Recognition Systems

- data collection

- feature selection

- classifiers

- classifier design and training

- supervised and unsupervised learning

2. Basic Decision Theory

- Bayesian decision theory

- Minimum error-rate classification

- Classifiers and decision boundaries

3. Parametric Methods

- Multivariate Gaussian model

- Class-conditional densities

- Sufficient statistics and model fitting

- Expectation-Maximization algorithm

- Overfitting and dimensionality reduction

4. Nonparametric Methods

- Density estimation

- Histogram classification rule

- Decision trees

- Nearest-neighbor classification

- Kernel methods

5. Statistical Learning Theory

- Complexity and regularization

- Probably Approximately Correct learning

- Chernoff's bound

- Distribution-free error bounds for classification

6. Statistical Analysis of Classifiers

- Analysis of histogram rule

- Analysis of decision trees

- Vapnik-Chevronenkis inequality

- Analysis of linear classifiers

Reference Materials:George Phillip's Lecture on Crystallography (ppt)George Phillip's LabCrystallography Datasets:Crystal Dataset ACrystal Dataset BDemos:demo1.m