CSCI 780: Introduction to Pattern Classification

Go back to CSCI 779

Go on to CSCI 781

Syllabus Date:

November, 1986

Introduction to Pattern Classification. (3) (Prereq: consent of instructor) Bayesian classifiers, optimal risk schemes, error rates, numerical methods, implementation, architectures.

R. O. Duda and P. E. Hart, Pattern Classification and Scene Analysis, Wiley, New York, New York, 1973. (ISBN: 0-471-22361-1)

R. A. Johnson and D. W. Wichern, Applied Multivariate Statistical Analysis, Prentice-Hall, Englewood Cliffs, New Jersey, 1982. (ISBN: 0-13-041400-X)

J. C. Bezdek, Pattern Recognition with Fuzzy Objective Function Algorithms, Plenum Press, New York, New York, 1981. (ISBN: 0-306-40671-3)

Dr. J. C. Bezdek, Professor

Dr. Gautam Biswas, Assistant Professor

Dr. Terry Huntsberger, Assistant Professor

To introduce students to standard methods in cluster analysis, feature extraction, and classifier design. The course contains a complete description of statistical decision theory, and students learn to implement and analyze the performance of at least one statistical classifier.

1. Linear Algebra

2. Probability and Statistics

3. Numerical Analysis

4. Advanced Calculus

1. Introduction, Overview, History (1)

2. Spectral Theory and Quadratic Forms (4)

3. Multivariate Normal Distributional Theory (5)

4. Statistical Decision Theory of Mixtures

a. Bayes Classifiers (2)

b. Minimum Risk Classifiers (1)

c. Error Rates and Labeled Data (1)

d. Sampling and Use of Data (2)

e. Implementation and Algorithms (2)

f. k-Nearest Neighbor Rules (2)

5. Linear Classifiers

a. Fisher's LDA and MDA (2)

b. Principal Components Analysis (2)

c. General Linear Machines (2)

6. Clustering

a. k-means and ISODATA (2)

b. Hierarchical Methods (3)

c. Relational Methods (2)

d. Fuzzy Models (3)

7. Feature Extraction

a. Linear Mappers (2)

b. Sammon's Algorithm (2)

c. Triangulation (2)

Each student must implement an algorithm and analyze a data set with it. Typically, students have coded and tested algorithms such as Wolfe's EM algorithm for iterative optimization of the maximum likelihood estimators, the (k,l) nearest neighbor rule classifier, the fuzzy c-means clustering algorithm, the single linkage clustering algorithm, principal components, Fisher's LDA, Sammon's algorithm, and triangulation. The project involves both data analysis and numerical issues such as convergence, initialization, stability, time complexity, and reliability.

Homework is assigned, collected, and graded on a regular basis. There are two exams: a midterm and a final.

Estimated CSAB Category Content:

Not Applicable.

Syllabus Flexibility:

High

Top of the page