Neural Networks

Major: Applied mathematics and computer science
Code of subject: 6.113.00.O.036
Credits: 5.00
Department: Applied Mathematics
Lecturer: Polovyi V.Ye.
Semester: 6 семестр
Mode of study: денна
Мета вивчення дисципліни: The purpose of studying the discipline is to get acquainted with the basics of the theory of artificial neural networks and their application in creating models of objects, control systems, as well as solving problems of classification and pattern recognition.
Завдання: As a result of studying the discipline, students should: - to know: the main types and purposes of artificial neural networks; architecture of artificial neural networks; methods of learning neural networks; working principles of the genetic algorithm and the Reinforcement Learning method. - be able to: choose the type of neural network to solve the given problem; perform a structural and parametric synthesis of a neural network to solve the given problem; perform appropriate experimental research, evaluate the obtained results and defend them with arguments.
Learning outcomes: The study of an educational discipline involves the formation of competencies in students of education: general competences: GC3 - basic knowledge in the field of information technologies, algorithms and their software implementation, necessary for mastering professionally oriented disciplines. GC5 – the ability to apply knowledge in practice. GC8 – to have skills in the development and management of projects. GC12 – the ability to work both individually and in a team. GC13 – the ability to communicate effectively at the professional and social levels. GC15 – potential for further education. professional competences: PC2 - basic knowledge of scientific concepts, theories and methods necessary for understanding the principles of information collection, processing and storage, development and software implementation of constructed algorithms. PC4 – the ability to use and implement new technologies, to participate in the modernization of systems and complexes in order to increase their efficiency. PC6 – the ability to apply professionally profiled knowledge and practical skills to solve typical problems of the specialty, as well as the operation of software and information complexes.
Required prior and related subjects: The previous ones Programming part 1, part 2. (1, 2 weeks) Algebra and geometry (1 semester) Differential equations (3 sem.) Object-oriented programming (3 semesters) Discrete mathematics (4 semesters) Mathematical foundations of artificial intelligence (4 semesters) Theory of probability (5 sem.) Mathematical statistics (6 semesters)
Summary of the subject: Навчальна дисципліна «Нейронні мережі» є складовою освітньо-професійної програми підготовки фахівців за першим рівнем вищої освіти «бакалавр» галузі знань 11 – «Математика та статистика» зі спеціальності 113 – «Прикладна математика» за освітньою програмою «Прикладна математика та інформатика». Дана дисципліна є обов’язковою. Викладається в другому семестрі 4-го курсу (8-й навчальний семестр) в обсязі – 120 год. (4 кредити ECTS) зокрема: лекції – 40 год., лабораторні заняття – 40 год., самостійна робота – 40 год. У курсі передбачено 2 контрольних роботи у ВНС. Завершується дисципліна – іспитом. У межах дисципліни розглядаються основні поняття та методи сучасної теорії штучних нейронних мереж (типи штучних нейронних мереж, їх архітектури та алгоритми навчання, функції активації нейронів, методи оптимізації, а також застосування еволюційних алгоритмів та Reinforcement Learning в задачах синтезу штучних нейронних мереж). На практичних заняттях студенти набувають навичок структурного та параметричного синтезу штучних нейронних мереж для розв’язування практичних задач моделювання, керування, класифікації, а також застосування сучасного прикладного програмного забезпечення для навчання та аналізу роботи нейромережі.
Опис: Basics of the theory of artificial neural networks. Basic concepts and definitions. Perceptron and ADALINE neurons. Study rules. Networks of direct signal propagation. Methods of learning artificial neural networks. Back-Propagation Algorithm Recurrent neural networks. Elman scheme, Jordan scheme. Extension of the Back-Propagation algorithm for the synthesis of recurrent networks. Levenberg-Marquardt algorithm. Creation of neuromodels and principles of building systems of neurocontrol of technological objects. Recurrent neural networks. Hamming network. Hopfield network. Application of networks for image classification. Networks with self-learning. ART network. Kohonen network RBF networks Neuro-fuzzy systems. Deep learning. Convolutional neural network. Structure, purpose of layers. The main types of CNN networks. LSTM and GRU networks Reinforcement learning Theoretical principles of functioning of genetic algorithms. MATLAB tools for the synthesis of artificial neural networks and their application in medicine, economics. PYTHON language tools for the synthesis of artificial neural networks and their application in Data Science.
Assessment methods and criteria: Laboratory works – 35, including: - L No. 1 – 9 - L No.2 – 9 - L No. 3 – 9 - L No. 4 – 8 Control works in the State Emergency Service – 10, including: - C #1 – 5 - C #2 – 5 Examination control - 55
Критерії оцінювання результатів навчання: The educational discipline ends with a semester control, the form of which is provided by the curriculum with a semester assessment. The semester grade consists of the sum of points provided for current control and examination control. The teacher proves this information to the students at the first lesson on the academic discipline
Recommended books: 1. Osowski S. Sieci Neuronowe w ujeciu algorytmicznym. – WNT:Warszawa, 1996. - P. 342. 2. D. Kriesel – A Brief Introduction to Neural Networks // - 227p. http://www.dkriesel.com/en/science/neural_networks 3. Rutkowska D., Pilinski M., Rutkowski L. Sieci neuronowe, algorytmy genetyczne i systemy rozmyte. –PWN:Warszawa-Lodz, 1997. – 412 p. 4. Терехов В.А., Ефимов Д.В., Тюкин И.Ю. Нейросетевые системы управления. – М:«Радиотехника», 2002. – 480 с. 5. Aston Zhang, Zachary C. Lipton, Mu Li, Alexander J.Smola. Dive into Deep Learning. Mar18, 2020, - 961 p. 6. Michael A. Nielsen. Neural Networks and Deep Learning. Determination Press, 2015. -216 p. http://neuralnetworksanddeeplearning.com/ 7. Ben Krose Patrick van der Smagt, An introduction of neural networks. University of Amsterdamam 1996, - 135 p. 8. Richard S. Sutton, Andrew G. Barto. Reinforcement Learning:An Introduction. MIT Press, 2016. - 330 p. 9. Goldberg D.E. Algorytmy genetyczne i ich zastosowania. –Warszawa: WNT, 1998. - 408 str. 10. Haykin S. Neural Networks: A Comprehensive Foundation. –Pearson Prentice Hall, 2007 – 863 p.