Henry S. Baird    Spring 2013 Course


Pattern Recognition

CSE 426/326   3 credits

CRNs:  18017 (426); 18016 (326)

 
Note to undergraduates:  you are very welcome in this course; if you take it at the 326 level, you'll do the same programming exercises as the grad students, but shorter HWs and exams.

Note to CS M.S. graduate students:  this course fulfills both the "Theory" and "Applied Theory" skill areas.

An introduction to the state of the art of pattern recognition, and the machine-learning theory, algorithms, and systems architectures that underlie them.

Theoretical topics will include Bayesian decision theory, statistically trainable vector-space classifiers, parametric classifiers (for, e.g., likelihoods with Gaussian densities), non-parametric classifiers (e.g. nearest neighbors), Perceptrons, generalized linear discriminants, kernel-based methods, decision trees, support-vector machines, neural nets, ensembles, and randomized classifiers. Also, we study general methodological issues, including best practices for statistical training and testing, the curse of dimensionality, and feature selection.

The last 1/4 of the course focuses on engineering challenges illustrated in applications chosen from the computer vision R&D literature. These reflect state-of-the-art approaches to segmentation, contextual analysis (including syntax and semantics), autonomous adaptation, style-conscious recognition, and anytime algorithms.

Weekly written homeworks or short programming exercises.  A midterm exam.  Students will select a set of related research papers (or a dissertation) from the recent literature and present a short talk in class summarizing and critiquing them.  There is a choice between (1) a final exam or (2) a software project on a cutting-edge research problem from computer vision, docunment image analysis, digital libraries, or Web security.

Course objective
On completing this course, students will be sufficiently familiar with the theory, notation, and vocabulary of pattern recognition and machine learning to be able to pursue matters of interest in the current technical literature. They will also have a grasp of key engineering issues arising in applications.

Textbook: Pattern Classification (2^nd Ed.), R. O. Duda, P. E. Hart, & D. G. Stork, John Wiley & Sons, October 2000. 680 pages. ISBN 0-471-05669-3.  If there are no copies available in the Univ. bookstore, email me immediately.

Lectures:  Tues/Thurs 9:20 AM - 10:35 AM (classroom is not yet assigned).

Instructor:  Prof. Henry Baird, hsb2@lehigh.edu. Office:  Packard Lab 380.   Office Hours:  Wednesdays 12:10-1:00 PM or by appointment.

Grader: (not yet appointed)

CourseSite:  CSE-426-326-00-SP13 Pattern Recognition.  We will use the online CourseSite facility to distribute lecture slides and homework assignments and data sets.  As soon as you are enrolled in this course, please browse coursesite.lehigh.edu and try to login:  if you cannot, send me email.

Prerequisites (if you have any questions about these, ask the Instructor):

CSE/Math 340:  Design and Analysis of Algorithms -- or comparable background in basic algorithms and data structures.

Math 205: Linear Algebra etc -- or similar familiarity with linear algebra & matrices.

Math 231 or Math 309 or CSC 450: Applied Probability & Statistics -- or some background in discrete probability and applied statistics & data analysis.

CSE 109: Programming in C++ -- or enough experience with C++, Java, C, or (ideally) MatLab to complete a small software project without faculty supervision.

Accommodations for Students with Disabilities:  If you have a disability for which you are or may be requesting accommodations, please contact both your instructor and the Office of Academic Support Services, University Center C212 (610-758-4152) as early as possible in the semester.  You must have documentation from the Academic Support Services office before accommodations can be granted.

If you have any questions, ask the instructor: hsb2@lehigh.edu.