Neural networks for processing of biological data

Garant předmětu:

Lecturer:

Tutor:

Supervisor:

Department of Biomedical Informatics

Synopsis:

Artificial neural networks belong to one of alternative techniques for analysis

and classification of data. Their main advantage is that there are able "to

learn" on the basis of examples, i.e. self-extract knowledge from available

data. Further advantage is their generalization ability, i.e. ability of proper

reaction on unknown stimulations. The subject is focused on basic biological

motivation, simplified model of neuron and neural network, taxonomy of

artificial neural networks. Individual and main paradigms of neural networks are

discussed (simple neuron, multilayer perceptron network, RFB neural network,

Hopfield network, Kohonen's self-organizing maps, ...). Preprocessing of data,

parameter setting, convergence and so on is discussed in detail. The end of the

subject is dedicated to computational means for artificial neural networks.

Requirements:

Syllabus of lectures:

- Biological motivation, basics of neural networks of living organisms. Transfer

of signals, memory function.

- History of artificial neural networks (ANN), definition of ANN, taxonomy of

ANN.

- Structure and function of neuron, network of neurons.

- Probabilistic ANN, generalized regression ANN.

- Multilayer perceptron network (MLP), method of learning/training,

backpropagation, Levenberg-Marquart, delta-bar-delta. Features of MLP.

Applications of MLP.

- RBF ANN, structure, features, learning, applications.

- Kohonen's self-organizing feature maps (SOFM), structure, features, learning,

applications.

- Hopfield ANN, structure, features, learning, applications.

- Associative memories, structure, features, learning, applications.

- Modern ways of development of ANN, optic ANN, holographic ANN

- Computational means (toolboxes) for ANN (Neural Networks Toolbox in Matlab,

Statistica Neural Networks)

Syllabus of tutorials:

Study Objective:

Artificial neural networks belong to one of alternative techniques for analysis

and classification of data. Their main advantage is that there are able "to

learn" on the basis of examples, i.e. self-extract knowledge from available

data. Further advantage is their generalization ability, i.e. ability of proper

reaction on unknown stimulations. The subject is focused on basic biological

motivation, simplified model of neuron and neural network, taxonomy of

artificial neural networks. Individual and main paradigms of neural networks are

discussed (simple neuron, multilayer perceptron network, RFB neural network,

Hopfield network, Kohonen's self-organizing maps, ...). Preprocessing of data,

parameter setting, convergence and so on is discussed in detail. The end of the

subject is dedicated to computational means for artificial neural networks.

Study materials:

[1]Rojas R: Neural Networks - A Systematic Introduction, Springer-Verlag,

Berlin, New-York, 1996

[2]Hassoun M.H.: Fundamentals of artificial neural networks, MIT Press, 1995

Note:

Further information:

No time-table has been prepared for this course

The course is a part of the following study plans: