Methods of Machine Learning in Design Systems

Major: Computer Science (Design and programming of intelligent systems and devices)
Code of subject: 6.122.12.E.060
Credits: 5.00
Department: Computer-Aided Design
Lecturer: Mariana V. Levkovych, Ph.D., Associate professor of the CAD Department
Semester: 5 семестр
Mode of study: денна
Мета вивчення дисципліни: familiarization with the main types of machine learning problems, models describing these problems for different types of available data, basic algorithms for solving these problems for different models, methods of preliminary data preparation and methods of evaluating the quality of solutions to machine learning problems, in particular their application in design systems.
Завдання: The study of an academic discipline involves the formation and development of students' competencies: general: - mastering the theoretical foundations of the "Methods of machine learning in design systems" course, mastering the technology of solving a wide class of problems with the help of methods, approaches and algorithms of machine learning, as well as acquiring practical skills and the skills of their application in solving specialized tasks, planning and implementing activities, aimed at solving problematic situations in the modern information space. professional: - the ability to solve complex specialized tasks and practical problems in the field of computer science or in the learning process, which involves the application of theories and methods of information technologies and is characterized by the complexity and uncertainty of conditions; - the ability to think logically, draw logical conclusions, use formal languages and models of algorithmic calculations, design, develop and analyze algorithms, evaluate their effectiveness and complexity, solvability and unsolvability of algorithmic problems for adequate modeling of subject areas and creation of software and information systems.
Learning outcomes: As a result of studying the discipline "Methods of machine learning in design systems", students should: - to know and systematically apply the methods of analysis and modeling of the application area, identification of information needs and collection of initial data in design systems; - to analyze, evaluate and choose methods, modern hardware and software tools and computing tools, technologies, algorithmic and software solutions for effective implementation of specific tasks; - to be able to develop, analyze and implement algorithms for solving the problems of system design, analysis, processing and storage of information; - to be able to use modern software systems and information technologies for the design and development of intelligent software systems.
Required prior and related subjects: Prerequisites: "Algebra and geometry", "Discrete mathematics", "Mathematical analysis and differential equations", "Probability theory and mathematical statistics", "Numerical methods", "Mathematical methods of operations research", "Systems of intellectual analysis and visualization of data". Co-requisites: "Design and development of information systems together with KR", "Computer graphics and geometric modeling", "Software robotics".
Summary of the subject: The educational discipline "Methods of machine learning in design systems" is a component of the educational and professional program "Design and programming of intelligent systems and devices" for the training of specialists at the first (bachelor's) level of higher education. This discipline belongs to the list of disciplines of the student's free choice. It is taught in the 5th semester in the amount of 150 hours (5 ECTS credits), in particular: lectures - 30 hours, laboratory classes - 30 hours, independent work - 90 hours. The discipline ends with an exam. The course covers the main tasks of machine learning: classification, regression, clustering, dimensionality reduction. Methods of solving them are studied. Emphasis is placed on understanding the mathematical foundations, relationships, advantages and disadvantages of the considered methods, as well as their practical implementation and application in solving real problems. As a result of studying the academic discipline, the student must: know: information about the main problems of machine learning, models for describing data, typical solution methods and conditions of applicability of these methods; be able to: choose algorithms for solving machine learning problems depending on the nature and structure of the data, pre-processing the data, evaluating the quality of the obtained solutions.
Опис: Part 1. Tasks of learning with a teacher - classification and forecasting. Topic 1. Types of machine learning problems. Classification task. Classification quality measures. Building machine learning models. Types of variables. Assessment of accuracy, selection of parameters. Topic 2. Decision trees. Gini separation criterion and entropy criterion. Conditional decision trees. Random forests. Topic 3. Bayesian classification. Logistic regression. Support vector machines. Topic 4. Regression problem. Measures of accuracy in the regression problem. Linear regression. Accuracy estimates. Nonlinear regression, locally weighted regression. Topic 5. Transformation of variables for machine learning problems. Regularized regression models: ridge, LASSO, Elastic-Net. Regression trees. Topic 6. Nearest neighbors method. Overcoming Relearning: Identifying Important Variables. Ensemble methods: boosting and bagging. Part 2. Unsupervised learning tasks and data preparation. Topic 7. Data preparation for machine learning tasks: cleaning, finding outliers, filling gaps. Tasks of learning without a teacher: clustering, finding anomalies, dimensionality reduction. Areas of their application. Topic 8. Clustering: centroid methods (kmeans, pam, clara, fanny). Clustering quality indices — internal and external. Clustering: connectivity methods (agglomerative clustering). Topic 9. Clustering: statistical methods (gaussian mixtures). Clustering: density analysis methods (DBSCAN, OPTICS). Clustering: nuclear methods (mean-shift).
Assessment methods and criteria: - current control (45%): written reports on laboratory work, oral survey; - final control (55%, examination control): testing (45%), oral component (10%).
Критерії оцінювання результатів навчання: Current control (laboratory work)-40 points. Examination control-50 points. Practical work-10 points. Together for the discipline - 100 points.
Recommended books: 1. A.D. Joseph, B. Nelson, B.I.P. Rubinstein, J.D. Tygar: Adversarial Machine Learning – Cambridge University Press, 2019. 2. C.C. Aggarwal: Machine Learning for Text – Springer, 2018. 3. D. Forsyth: Applied Machine Learning – Springer, 2019. 4. E. Alpaydin: Introduction to Machine Learning, 3rd ed. – The MIT Press, 2014. 5. J.-T. Chien: Source Separation and Machine Learning – Academic Press, 2019. 6. M. Gori: Machine Learning: A Constraint-Based Approach Morgan Kaufmann, 2018. 7. M. Mohri, A. Rostamizadeh, A, Talwalkar: Foundations of Machine Learning, 2nd ed. – The MIT Press, 2018. 8. P. Larraсaga, D. Atienza, J. Diaz-Rozo, A. Ogbechie, C. Puerto-Santana, C. Bielza: Industrial Applications of Machine Learning – CRC Press, 2019. 9. R.E. Neapolitan, X. Jiang: Artificial Intelligence: With an Introduction to Machine Learning, 2nd ed. – CRC Press, 2018. 10. S.W. Knox: Machine Learning: A Concise Introduction – Wiley, 2018.