Jacek M. Zurada, Ph.D

ECE 614 - Artificial Neural Systems

Foundations of learning machines and neural processing algorithms, supervised and unsupervised learning of feedforward and recurrent neural networks, perceptron layers, associative memories, feature maps. Applications in the areas of classification, control, and signal processing. Implementation issues.

Prerequisite: Graduate/Professional Standing

Textbook

Jacek M. Zurada, Introduction to Artificial Neural Systems, West Publishing Co., 1992.

References

• Y. H. Pao, Adaptive Pattern Recognition and Neural Networks, Addison-Wesley, 1989.
• J. Hertz, A. Krogh, R.G. Palmer, Introduction to the Theory of Neural Computation, Addison-Wesley, 1991.

Goals

Course designed to provide students with foundations of learning machines and neural processing algorithms, as well as with neural networks design, and neurocomputing simulation techniques. This course involves mandatory design projects (see Laboratory Projects). Such projects are completed using computers or neural networks development systems (see Computer Usage). Oral presentations and project reports are required to enhance students' communications skills.

Prerequisites by Topic

• introduction to linear and nonlinear systems
• stability
• solid geometry
• matrix calculus
• introduction to optimization
• electronic microsystems

Topics

1. Introduction and examples of applications
2. Fundamental models of artificial neural systems, learning rules, taxonomy of neural networks
3. Single-layer perceptron classifiers and approximators
4. Multilayer feedforward networks: Error back propagation (EBP) training and its modifications
5. Single-layer fully coupled networks with continuous and discrete time
6. Associative memories for pattern retrieval and noise suppression
7. Matching and self-organizing networks, unsupervised learning techniques, feature maps
8. Neural network implementation: hardware and software

Computer Usage

• Simulation of neural network paradigms
• Training and recall modeling using neural algorithms
• Development of neural software for classification
• Coding of algorithms in high-level language
• Use of professional neural network simulation packages
• Error Back Propagation Program
• A/D Converter Program

Example of Laboratory Projects

• Desing of trainable classifiers, evaluation of their capacity and sensitivity under noisy conditions.
• Design, simulation and performance evaluation of EBP multilayer pattern classifier.
• Associative memory for noise removal: simulation and performance evaluation.

Class Policy

• Three sixty-minute quizzes (18% each)
• One project and seminar in a group of two (10%)
• One final exam (26%) Homework assigned weekly (10%)