Bachelor
2024/2025
Research Seminar "Data Analysis and Artificial Intelligence 2"
Type:
Compulsory course (Applied Mathematics and Information Science)
Area of studies:
Applied Mathematics and Information Science
Delivered by:
School of Data Analysis and Artificial Intelligence
Where:
Faculty of Computer Science
When:
4 year, 1-3 module
Mode of studies:
offline
Open to:
students of one campus
Instructors:
Vasilii Gromov
Language:
English
ECTS credits:
5
Course Syllabus
Abstract
The course covers modern machine learning and data analysis techniques, with an emphasis on clustering, time series forecasting, neural networks and information-theoretical approaches. The first part discusses density-based clustering algorithms, methods for extracting informative features and co-clustering, as well as working with high-dimensional data including the use of decision trees. In the second part, we focus on forecasting: analyzing the randomness of time series, estimating algorithms and multi-step forecasting based on clustering. Third, we explore modern methods of data representation, advanced neural network architectures such as constructive networks and neurodifferential equations, and integration of information theory. Finally, we conclude with the topic of adaptive learning using entropy metrics and introducing domain ontologies into model structures.
Learning Objectives
- To train specialists capable of designing and implementing complex machine learning algorithms for data analysis and forecasting chaotic processes in order to make informed decisions in conditions of uncertainty. The course will build skills in working with advanced technologies, including neural network models and information theory methods, in order to solve current problems in various fields, such as science, business, and the social sphere.
Expected Learning Outcomes
- To study density-based clustering algorithms (OPTICS), iterative methods using a minimum spanning tree, co-clustering, and methods for working with high-dimensional data. Learn how to apply clustering to solve segmentation problems, including detecting fraudulent transactions and patterns of energy consumption in Smart Cities.
- Master the methods of L1-regularization, logistic regression, and feature selection algorithms for analyzing related and multidimensional data. Learn to identify statistically significant areas in high-dimensional spaces.
- To study methods for estimating the chatocity of time series: senior Lyapunov exponent, Kolmogorov-Sinai entropy, and entropy-complexity plane. Master clusterization-based forecasting algorithms, including multistep models and early warning systems for trend changes.
- To study constructive neural networks (cascade correlation) and neurodifferential equations (ResNet, extended models), as well as their application in diagnosing oncological diseases, learn how to integrate domain ontologies into neural network structures to enhance interpretability.
- To study the bottleneck principle in deep learning, coarse-grained networks and the use of generalized entropy as an error function and how to apply adaptive learning based on entropy metrics.
- Master methods for predicting phase transitions, spatial and temporal indicators of disasters in ecosystems, and early warning algorithms. To study the representation of data through discrete structures (tensors of connections, chaotic graphs) for the analysis of unsteady processes.
Course Contents
- Modern clustering methods
- Extracting informative features
- Forecasting and time series analysis
- Modern neural network architectures
- Information-theoretical approaches
- Analysis of complex systems.
