Neural Networks

Update: August 19th, 2008

Research in neural networks had produced a wealth of computational models, each motivated by specific application problems like pattern recognition, manufacturing, quality control, and optimization.

Neural networks are computational structures that model simple biological processes usually associated with the human brain. They are massively parallel systems capable of learning by adaption. Their adaptive properties make them useful in solving a variety of current difficult problems but also required for the development of future intelligent systems capable of adapting to untrained situations.

This course will develop the terminology used in neural network research and provide a comprehensive analysis of a broad range of neural network models. The evolution of neural networks will be presented from the early McCulloch-Pitts model of neural computation and the problems identified with the perceptron model, through the fundamentals of feed-forward and recurrent systems.

In addition to specific models technologies and methodologies associated with neural networks will also be presented. This will include fuzzy neural networks, deterministic learning methods such as Hebb’s rule and back-propagation, stochastic learning methods such as genetic algorithms and simulated annealing and neural networks for constrained optimization. Support Vector Machines and Data Mining applications will also be discussed.

Application areas presented will include pattern recognition, prediction estimation, control, finance medicine, process modeling, and optimization.

A unique aspect of this course will be the interface between operations research and neural networks techniques.

Topics

  1. Overview
    • Historic Perspective
    • Biological Foundations
  2. Mathematical Models
    • Perceptron
    • Feed-forward Systems
    • Hopfield
    • Feedback Systems
    • Back-propagation
    • Adaptive Resonance Theory (ART)
    • Fuzzy Neural Networks Learning
    • Deterministic Methods (Hebb’s rule, back-propagation)
    • Stochastic Methods (genetic algorithms, simulated annealing)
    • Software
  3. Applications
    • Pattern Recognition
    • Optimization
    • Vehicle Routing
    • Financial Systems
    • Stock Market Predictions
    • Medicine
    • Diagnostic Routines

Lectures Notes

A free PostScript interpreter (AFPL Ghostscript) and its graphical interface (GSview) can be downloaded from here.

Grading

There will be no test. Grades will be determined by students’ solution to problems distributed during the semester. In addition, each student will be required to make an oral presentation and submit a project.

Project Guidelines

  1. The project is to be done individually.
  2. Each student should select a project falling in at least one of the following categories:
    • A direct application of a Neural Network model from a text or journal by collecting data and solving it using the computer.
    • Improving and extending the results of a given study for a more realistic solution. Here, you may use the same data available in the study and compare your results to the existing one.
    • Developing a new and different Neural Network model for a real problem and present solution approach. Here, you may just use fictitious data to illustrate your methodology.
    • Developing or improving a computer program for a neural network algorithm.
    • Experiments with different neural network algorithms on problems.
  3. Preliminary Report: A two to three page statement explaining the project. Note: Before submitting the report, you may discuss the project informally to ascertain whether your project will meet the desired objectives and standards.
  4. Final Report: Due final day of classes. To be limited to 15 pages (excluding computer printout and appendices). Your final report should discuss the model formulation and solution, highlighting the major contribution made by you through the project work, and difficulties encountered, deviation from the preliminary objectives, and significant conclusions. Note: You are welcome to discuss the progress of your project from time to time.
  5. The following points will be taken into consideration while awarding the project grade:
    • Complexity of the project.
    • Adherence to the project guidelines.
    • Presentation of final report.
    • Results and major contributions.
    • Completeness.
Comments are closed.
TOP