ECE:
Topics in Electrical and Computer Engineering
Artificial Neural Systems
|
|
Instructor: Computational
Intelligence and Machine Vision Laboratory Department of Electrical and Computer
Engineering |
Prerequisite:
Graduate standing AND consent of instructor
AND working knowledge of C/C++ and MATLAB
Objectives:
This is an
introductory course on artificial neural networks and its applications. The
primary emphasis is on the theory, modeling/analysis and representative
applications of artificial neural networks. The computational capabilities and
limitations of several popular neural networks will be examined. Students
taking this course should have an understanding of undergraduate-level
calculus, linear algebra and probability theory. Hardware design aspects of
various neural networks will also be discussed.
Textbook:
Simon Haykin, Neural Networks – A Comprehensive Foundation, Prentice Hall, 1999
Supplementary
Textbook:
K. Mehrotra, C.
K. Mohan, and S. Ranka, Elements of Artificial Neural Networks,
References:
A Cichocki
and R. Unbehauen, Neural Networks for Optimization and Signal Processing,
John Wiley and Sons, 1993
J. Anderson and
E. Rosenfeld, Neurocomputing, MIT Press, 1988
Hertz, Krogh and
Palmer, Introduction to the Theory of Neural Computation,
Addison-Wesley, 1991
Additional
References:
IEEE Transactions on Neural Networks
IEEE Transactions on Systems, Man, Cybernetics
Neural Networks
-
Official
Journal of the International Neural Network Society
International Journal of Neural Systems
Journal of Neurocomputing
Goals:
This course
is designed to provide graduate students in electrical and computer engineering
the ability to design and develop algorithms and architectures for specific
applications based on artificial neural networks.
Topics:
Introduction: Biological neurons and
memory, structure and function of a single neuron; Artificial neural networks
(ANN); Typical applications of ANNs: classification, clustering, vector
quantization, pattern recognition, function approximation, forecasting,
optimization; Basic approach of the working of ANN - training, learning and
generalization; Supervised learning: single-layer networks, perceptron-linear
separability, training algorithm, limitations, multi-layer
networks-architecture, back propagation algorithm and other training
algorithms, applications; Adaptive multi-layer networks - architecture,
training algorithms; Recurrent networks; Radial-basis-function networks;
Unsupervised learning: winner-takes-all networks, Hamming networks, maxnet,
competitive learning, vector-quantization, counter propagation networks,
adaptive resonance theory, Kohonen's self-organizing maps; Hopfield networks,
Boltzmann machine; Optimization methods: Hopfield networks for-TSP, solution of
simultaneous linear equations, iterated gradient descent, simulated annealing;
Hardware realization of ANNs; Recent trends and future directions.
Course
Overview:
This class will focus on artificial
neural systems. Many classical networks (architectures
Laboratory Exercises:
Exercise problems will be
implemented in C/C++ or using MATLAB.
Project:
The course project
includes developing the specification of a neural network, performing an
analysis of its feasibility, developing the detailed design, implementation of
the neural network, writing a report on the design and making a presentation to
the vision lab research team.
Course
Description for Graduate Students:
Prepare reports on selected topics and discuss in group
meetings.
Present a
term paper on the realization of an application using one of the neural
networks.
Submit at
least one research paper to a reputed journal/conference/workshop for
presentation and publication.
Grading Policy
Reports: 30%
Term Paper: 20%
Presentation: 20%
Research
Paper: 30%
Course
Notes:
2. Learning
Processes 
3. Single Layer
Perceptrons
4. Multilayer
Perceptrons
5. Radial Basis
Function Networks
6. Support Vector
Machine
7. Principle
Component Analysis
8. Self-Organizing
Maps
