Certificate Course

Neural Networks, Genetic Algorithm, Gene Expression Programming and Optimization

Text Book

The course can be done in self-study.

Artificial neural networks model the functioning of biological neural networks like the brain. They are currently extensively used for pattern recognition (e.g. signature identification on cheques), financial forecasting, control and medical diagnosis. Neural networks are simple models for computing using massive parallelism. This course gives a thorough background on supervised and unsupervised learning in neural networks.
Genetic algorithms are derived from our current understanding of genetic evolution in nature. They are widely used for finding the global minimum of a cost or objective function (especially when many local minima are present). Genetic algorithms can even be used to write computer programs. The algorithms are generally simple and find near optimimum solutions very quickly.

This course is ideally suited for engineers (especially from the fields pattern recognition and control), people interested in stock-market forecasting, medical professionals and computer scientists.

• First-year level calculus.
• Background in object-oriented programming in C++ (preferably successful completion of the certificate course Object_Oriented Programming in C++).

• Introduction
  • Why neural networks ?
  • Typical neural network problems
  • Biological background
  • Modeling neurons
• Supervised learning in feed-forward neural networks
  • Single layer perceptron
  • The Backpropagation algorithm
  • Improvements to the Backpropagation algorithm
  • C++ program examples
• Supervised learning in recurrent neural networks
  • Dynamic learning in recurrent neural networks
  • Unfolding a recurrent neural network as a feed-forward neural network
• Unsupervised learning
  • Hopfield neural networks
  • Principle components networks
  • Competitive learning and Kohonen networks

Genetic Algorithms

• The sequential genetic algorithm.
• Encoding your problem space on a chromosome.
• Operators for genetic evolution.
• Markov chain analysis of certain genetic algorithms.
• Bit sets in C++ and Java.
• Problems with constraints (Traveling salesman problem, Knapsack problem, Four colour problem)
• Deterministic alternatives: Discrete global optimization
• A C++ class library for genetic algorithms (with example programs)
• Example applications

• Writing computer programs using genetic algorithm techniques.
• Symbolic approximation of functions.

• Writing computer programs using genetic algorithm techniques.
• Overcome the complexities and difficulties of genetic programming.
• Symbolic approximation of functions.

Timetable and Enrollment Forms.

For further information contact Prof. W.-H. Steeb