Introduction to neural networks 1000-I1WSN

The aim of the course is to familiarise the students with classical and contemporary models of biologically inspired neural networks along with their applications in artificial intelligence and machine learning.

Course outline:
* Biological model of neural cell
* Perceptron model,
* Other single-cell models:
- multi-class linear classifier,
- AdaLiNe (Adaptive Linear Neuron),
- radial basis functions (RBF),
* Algorithms for feed-forward network construction,
- spire algorithm,
- pyramidal algorithm,
- tiling algorithm,
- up-start algorithm
* Back-error propagation algorithm (BEP),
* Validation of learning algorithms,
* Unsupervised learning,
* Self-organising algorithms,
- Kohonen self-organising maps,
- k-means,
- ART2,
* Principal component analysis (PCA),
* Recurrent networks, Hopfield autoassociator,
* Boltzmann machines, simulated annealing,
* Complex-valued neural networks,
* Genetic algorithms (if time admits),

Total student workload

Lecture 30h: exam written and oral Labs 30h: excercises, programming projects (circa 6, min. 3 for pass), possible presentations Unsupervised work: 80-120h

Learning outcomes - knowledge

As a result of the course, the student should: * give a brief characterisation of the biological motivations of neural networks, INZ_W02 * define basic neural models, their dynamics, supervised and unsupervised learning algorithms, applications, INZ_W04 * explain the dynamics and limitations of the models, also in geometrical terms, INZ_W01

Learning outcomes - skills

As a result of the course, the student should be able to: * choose the proper model, implement, train and analyse it, INZ1_U01, INZ1_U07 * select or design accurate style of presentation of the results, INZ1_U15

Learning outcomes - social competencies

As a result of the course, the student should gain following attitudes: * notice the relations etween mathematical tools and computer science models, INZ_K03 * creativity in designing and implementing of the solutions, INZ_K02 * presenting of numerical and abstract results, INZ_K05 * awarness of pros and cons of the numerical and probabilistic solutions, INZ_K03

Teaching methods

Lecture (direct) : presentation, discussion, moodle support Labs (direct): excercises, programming projects, presentation (not obligatory), moodle support

Observation/demonstration teaching methods

- simulation (simulation games)
- display

Expository teaching methods

- informative (conventional) lecture

Exploratory teaching methods

- presentation of a paper
- laboratory
- project work

Prerequisites

Programming Courses I and II, Algorithms and data structures, at lease one programming language with GUI libraries, basic concepts of Linear algebra

Course coordinators

Term 2024/25L:

Andrzej Rutkowski

Term 2023/24L:

Krzysztof Rykaczewski

Term 2022/23L:

Krzysztof Rykaczewski

Term 2025/26L:

Krzysztof Rykaczewski

Assessment criteria

Lecture:
* oral and written exam

Laboratories
* pass for grade
* test
* programming tasks (3 to 6 programs)

Practical placement

Not available.

Bibliography

The bibliography (primary):
[1] R. Rojas Neural Networks, A Systematic Introduction, Springer 1996,
[2] P. Peretto, Introduction to Modelling Neural Networks, Cambridge University Press 1994,

The bibliography (secondary)::
[1] C. Bishop, Neural Networks for Pattern Recognition, Oxford University Press 1995.
[2] J. Mańdziuk, Sieci Neuronowe Typu Hopfielda. Teoria i Przykłady Zastosowań, Akademica Oficyna Wydawnicza EXIT, Warszawa 2000.
[3] E. Izhikevich, Dynamical Systems in Neuroscience, The MIT Press, Cambridge, Massachusetts, London, England, 2007,

Additional information

Additional information (registration calendar, class conductors, localization and schedules of classes), might be available in the USOSweb system:

Description of 1000-I1WSN in USOSweb