University of North Carolina at Chapel Hill
                Department of Computer Science

                SPRING 1994 COURSE ANNOUNCEMENT

                    COMP 273: Neural Networks
            (Neural Networks and Pattern Recognition)
      Monday, Wednesday 9:30-10:45 starting January 10, 1994
                         Sitterson 325

INSTRUCTOR: Jonathan Marshall 962-1887 <marshall@cs.unc.edu>

This graduate research seminar will introduce the idea of using neural
networks for pattern recognition in a variety of domains, including speech
perception, visual perception, and abstract pattern extraction.  The seminar
will focus on self-organizing pattern recognizers -- which learn categories
automatically, without being explicitly "taught" to do so.  Because
self-organizing neural networks are intended to correspond with how real
brains process information, the seminar will be of interest to cognitive
scientists and neurobiologists, as well as computer scientists and engineers.

Introductory information will be provided on neural networks and on pattern
recognition.  The seminar topics will include:

    o introduction to neural networks
    o unsupervised learning, adaptation, and self-organization
    o formation of stable category codes
    o generalization
    o clustering
    o segmentation
    o context-sensitive pattern recognition
    o simultaneous recognition of multiple patterns
    o noisy pattern recognition
    o real-time pattern recognition
    o fast and slow learning
    o unlearning of category information
    o category coarseness
    o synonym and homonym processing (the "banana problem")

Neural network modeling is concerned with understanding both how neurons in
animal brains communicate and perform tasks for perception, cognition, and
motor control, and how networks of artificial neuron-like processing elements
can solve computational problems.  Neural network research is highly
interdisciplinary: computational theory helps explain and predict findings in
neurobiology and psychophysics, while neurobiological and psychological
results suggest new methods for solving computational problems.

The seminar will focus on the book "Neural Networks for Pattern Recognition"
by Albert Nigrin (MIT Press, 1993).  The seminar may also cover selected
papers from the recent literature.  Workload will consist of readings, class
presentation(s), class participation, writing short critiques, and either a
research project/paper or a take-home final exam.

The seminar will be aimed at an introductory level.  Prior exposure to
material on neural networks or pattern recognition is not a prerequisite, but
might be helpful.  This seminar is open both to students who have not studied
neural networks or pattern recognition and to students who have.  Graduate
students in other departments, as well as those in computer science, are
encouraged to take the seminar.  Faculty members and postdocs are also
welcome.  The seminar is also open to students who have previously taken my
Neural Networks courses, since the material will be about 95% new.