Course aims:

Develop the student's ability to develop software applications dedicated to solving problems of medium-high complexity by exploiting the principles of algorithms based on neural networks.
The course aims to acquire the theoretical and applicative knowledge regarding: - principles of neural networks - learning algorithms in artificial neural networks - design and implementation of neural networks - creating applications in Matlab
Develop the student's ability to find unconventional methods of problem solving.
Develop the student's ability to find correct representation of problems that can be addressed with neural networks, choosing the right type of architecture.
Develop the student's ability for applications of neural networks to solve problems of: classification, approximation, grouping and association of data, image processing, etc.

Course Entry Requirements:

Course subjects from the previous semesters: Artificial intelligence

Course contents:

1. Introduction to the theory of neural networks. Natural neuron and artificial neuron. Models of neurons and artificial neural networks. Learning neural networks. Implementations, applications, trends. 2. Feed-forward neural networks. The Perceptron Model. Adaline and Madaline models. Delta rule. 3. Multilayer feed-forward architectures and error retropropagation method. Limits of one-tier networks. Multi-level architectures with feedforward connections. 4. Radial Basic Function Networks (RBF). Architecture and operation. The ability to represent RBF networks. Learning algorithms 5. Recurrent neural networks for associative memories. Associative memories. A mathematical model of the recurrent neural network. Hopfield model and data storage algorithms (Hebb rule, Diederich-Opper algorithm). 6. Problems of combinatorial optimization. Simulated riding algorithm. Stochastic machines: Boltzmann machine, Cauchy machine, Helmholtz machine. Applicability and limitations. 7. Processing of time series. Preprocessing. Networks with temporary windows. The Elman model. 8. Cellular networks. Architecture. Operation. Applications in image processing. 9. Self-organizing neural networks. Unsupervised learning. Biological basis. Self-organizing neural networks, 10. Neuro-symbolic hybrid architectures. Extraction of rules from neural networks. Expert systems combined with neural networks. 11. Neuro-fuzzy hybrid architectures. Neuro-genetic hybrid architectures. Genetic algorithms for optimizing the topology of a neural network. 12. Applications of neural networks. Areas of applicability, examples of known neural systems, successfully used in real problems.

Teaching methods:

The course implies verbal presentation of the course concepts using Powerpoint presentations as a teaching support. We encourage a permanent dialogue with students and answering any questions of them.

Learning outcomes:

Upon completing this course, the student will be able to apply the main machine learning techniques to solving practical problems. Advanced students will be able to use these techniques in research.

Learning outcomes verification and assessment criteria:

Minimum performance standard: understanding the basic concepts of the principles of neural computation and learning algorithms in artificial neural networks as well as the practical use of the main algorithms based on neural networks.

Recommended reading:

C. Bishop, Pattern Recognition and Machine Learning,, Springer, New York, 300, 2006.
Gareth James, Daniela Witten, Trevor Hastie and Robert Tibshirani, An Introduction to Statistical Learning with Applications in R, Springer-Verlag, Alba Iulia, 2000, 200.
T. Mitechell, Machine Learning, Springer, Alba Iulia, 2006, 200.
David Mackay, Information Theory, Inference, and Learning Algorithms, Cambridge University Press, Alba Iulia, 2003, 200.
Cattell R., Barry D.K.,, The Object Data Management Standard: ODMG 3.0,, Morgan Kaufmann, Alba Iulia, 2000, 200.