Course details
Neural Networks, Adaptive and Optimum Filtering
QB4 Acad. year 2007/2008 Winter semester
In its first part, the course is devoted to providing an overview of types of architecture of neural networks and to a detailed analysis of their properties. Applications of neural networks in signal and image processing and recognition are included in this treatment. In the second part, the course deals with the theory of optimum detection and restoration of signals in its classical and generalised forms, emphasising the common base of this whole area. The subject highlights the common view-points in the area of neural networks and in the area of optimised signal processing.
Guarantor
Language of instruction
Completion
Time span
- 39 hrs lectures
Department
Subject specific learning outcomes and competences
Theoretical knowledge of areas of neural networks and optimum signal processing, ability to apply and, if necessary, to modify these methods for concrete problems.
Learning objectives
Gaining knowledge of theory of neural networks and theory of adaptive and optimum filtering, showing common view-points of both areas
Prerequisite knowledge and skills
signal and system theory, digital signal processing (e.g. the subjects BCZA, MMZS)
Study literature
- J.Jan: Digital Signal Filtering, Analysis and Restoration. IEE Publishing, London, UK, 2000
- B. Kosko (ed.): Neural Networks for signal processing. Prentice Hall 1992
- Jan, J,: Číslicová filtrace, analýza a restaurace signálů. 2. rozš. vydání. VUTIUM Brno 2003
Fundamental literature
- B. Kosko: Neural Networks and fuzzy systems. Prentice Hall 1992
- B. Kosko (ed.): Neural Networks for signal processing. Prentice Hall 1992
- S. Haykin: Neural Networks. Prentice Hall 1994
- J.G.Proakis, et al.: Advanced digital signal processing. McMillan Publ. 1992
- J.Jan: Digital Signal Filtering, Analysis and Restoration. IEE Publishing, London, UK, 2000
- P.M.Clarkson: Optimal and Adaptive Signal Processing. CRC Press, 1993
- S. Haykin: Adaptive Filter Theory. Prentice-Hall Int. 1991
- V.K.Madisetti, D.B.Williams (eds.): The Digital Signal Processing Handbook. CRC Press & IEEE Press, 1998
Syllabus of lectures
- Architectures and classification of neural networks. A neuron as a processor a classifier, methods of training, hard-learning problems
- Feed-forward networks, single- and multilayer perceptron. Learning: error back-propagation as iterative minimisation of the mean quadratic error
- Supervised and unsupervised learning. Knowledge generalisation, optimum degree of training
- Feed-back networks. Hopfield networks, behaviour, state diagram, attractors, learning. Networks with hidden nodes
- Application of relaxing minimisation of "energy" for optimisation problems, use of the network as associative memory. Stochastic neuron, Boltzmann machine, simulated annealing
- Recursive and Jordan networks. Competitive learning
- Kohonen maps, associative learning, automatic local organisation, refining of classification
- Possibilities of neuronal networks as signal processors and analysers, practical applications in processing and restoration of signals and images
- Optimum signal detection and restoration - approaches. Non-linear matched filters, effectivity comparison
- Deterioration models, LMS-filtering, diskrete Wiener filter in non-stationary environment
- Kalman filtering in scalar version, vector generalisation in stationary and non-stationary environment
- Adaptive filtering, adaptation algorithms, recursive realisation of adaptive filtering, filtering by method of stochastic gradients
- Typical applications of adaptive filtering. Comparison of concepts of optimum and adaptive filtering and neural-network oriented approach.
Progress assessment
Study evaluation is based on marks obtained for specified items. Minimimum number of marks to pass is 50.
Controlled instruction
There are no checked study.