Course Overview

The course covers theory and practice of pattern recognition, emphasizing concepts and practices that are common to a broad range of applications and technologies. The course is fundamental for anyone whose work involves statistical signal processing, recognition, or fault detection. The course material provides a core for work in speech and image applications with advanced recognition technologies, and finds application in anomaly detection, decision theory, timeseries analysis, and statistical signal processing.

Pattern recognition technology finds its way into a multitude of commercial applications including: medical tests, credit card fraud detection, data mining, manufacturing test, speech recognition systems, handwriting and optical character recognition, fault detection, and financial analysis to name a few. If you're planning to work, or already work in statistical signal processing or recognition, or if you simply want to know how pattern recognition technology works, this course will provide a valuable addition to your bag of tools. The statistics you will learn have even broader application.

Topics include: random vectors, detection, ROC curves, parametric and non-parametric density estimation and classification models, empirical error bounds, Bayesian estimation, and feature extraction.

Students will develop familiarity with statistical tools for developing and assessing pattern recognition systems, and apply principles to real-world examples.

Prerequisites:  The course assumes familiarity with college linear algebra and calculus, and at least one undergraduate probability and statistics course.

   Viewing Linked Material

   MATLAB Resources

MATLAB home page

MATLAB Tutorials:

1. Quick, local tutorial
2. University of New Hampshire
3. Carnegie Mellon University

•  Pima indians diabetes data --- description, data
• Wisconsin breast cancer data --- description, data

   Other Course Materials

• Course Overview pdf file.
• Course Syllabus pdf file.  Also see the OHSU Course Syllabus Statements
  
• Matlab Example -- Generating samples from a normal distribution with specified mean vector and covariance matrix. pdf file,
• Mathematical Preliminaries
• Random variables and linear transformations. pdf file.
• Constrained optimization using Lagrange multipliers pdf file.
• Article --- Million-Dollar Murray, Malcom Gladwell.  The New Yorker, Feb. 13, 2006.
• Book section on reject option pages 78-82,  Introduction to Statistical Pattern Recognition, Second Edition, Keinosuke Fukunaga.  Academic Press, 1990.
  

    Assignments

• Lecture 1 Overview
• Lecture 1B Math Review
  
• Lecture 2  Bayes Classifiers.  For lectures 2-4, READ Duda, Hart, and Stork 2.1-2.9
  
• Lecture 2 Supplement
• Lecture 3-4 Bayes Classifiers (cont'd)
• Lecture 5 More Bayes classifiers (material not in text, see Fukunaga's book --- Ancillary Texts, above --- for reject option and sequential likelihood ratio test)
  
• Lecture 5B Application example.  Algorithmic details are in Cynthia Archer, Todd Leen, and Antonio Baptista.  Parameterized Novelty
  Detection for Environmental Sensor Monitoring Advances in Neural, Information Processing Systems 16, The MIT Press, 2004.
• Lecture 6 Analytic Error Bounds READ Duda, Heart, Stork (DHS) 2.8.
• Algebra for p4 of Lec 6
• Lecture 7 Parametric Classifiers READ DHS  3.8.2-3, 5.1-2, 5.3-5.8.
• Lecture 8 Optimization of Criteria Functions
  

http://www.ohsu.edu/edcomm/flash/flash_player.php?params=4%60/sch/ogi/cs/547-010912b.flv%60vod&width=640&height=480&title=CS%20547b%20January%209th%2C%202012