Neural Networks, Machine Learning and Deep Learning
The course is dedicated to the study of machine learning (ML) and deep learning (DL), with an emphasis on the engineering application of these technologies in digital systems. The programme covers the full cycle of working with ML models: from problem formulation and data analysis to training, quality assessment, deployment, and maintenance. Students will learn how to formalise applied problems as ML tasks, select appropriate models and evaluation metrics, work with different types of data (images, sequences, text), as well as design ML pipelines and take into account the operational constraints of real‑world systems.
The course aims to develop students’ holistic understanding of machine learning as an engineering discipline and to teach them how to design and implement ML components in digital systems. The course combines theoretical foundations with hands‑on work on an end‑to‑end project that goes through all stages of the ML model lifecycle.
The course fosters engineering thinking: students learn to see an ML model as a component of a digital system, not an isolated algorithm, and to make informed decisions at every stage of its lifecycle.
Upon completing the course, students will be able to analyse data and identify potential issues, interpret quality metrics in the context of the task, build basic and advanced models including linear models, ensembles and neural networks, design ML pipelines, understand modern DL architectures such as CNNs, RNNs and Transformers along with their application scenarios, and work with pre‑trained and foundational models.
The course offers a structured engineering trajectory for developing machine learning (ML) models, guiding students through a complete lifecycle: from problem statement and data analysis to model training, evaluation, architectural design, and system implementation. It builds knowledge progressively — starting with understanding ML tasks, moving on to model construction and deep learning architectures, and culminating in AI systems engineering. A key focus is on treating ML as an engineering discipline, where students learn to formalise problems, prepare data, select and train models, assess their quality, and integrate them into digital systems. The curriculum emphasises an experimental culture: every hypothesis must be tested through model training, metric measurement, and error analysis, reinforcing the professional principle of trusting empirical results over assumptions..
Another core aspect of the course is developing architectural thinking — students learn to view ML models not as isolated algorithms but as integral components of a system’s architecture. This includes understanding the role of ML within broader digital systems, accounting for computing resource limitations, meeting performance requirements, and designing appropriate implementation architectures. Large language models (LLMs) are integrated as versatile engineering tools: they assist in hypothesis generation, model variant formulation, result analysis, and architectural discussions, while also helping identify model weaknesses and propose alternatives. Crucially, the course instils a principle of mandatory verification — any LLM‑generated suggestion must be validated through experimentation, measurement, and thorough result analysis, ensuring a rigorous, evidence‑based approach to ML development..
OBJECTIVES
Mastering the basic principles of machine learning;
Development of skills for setting ML problems;
Mastering methods of data preparation and analysis;
Study of the main classes of machine learning models;
Mastering neural network architectures;
Study of deep learning methods;
Development of skills for comparing models and assessing quality;
Mastering the principles of integrating models into digital systems;
Formation of a culture of experimental analysis of models;
Developing skills for critical use of LLMs in engineering activities.
KEY TASKS
Studying the types of machine learning problems;
Mastering data engineering methods;
Studying quality metrics of models;
Mastering methods of linear models and ensembles;
Study of unsupervised learning methods;
Mastering dimensionality reduction methods;
Study of the basic principles of neural networks;
Study of the backpropagation algorithm;
Mastering deep learning architectures;
Studying specialized architectures for different types of data;
Learning the principles of transfer learning;
Study of fundamental models;
Mastering the principles of ML pipeline;
Development of experimental management skills;
Mastering the principles of implementing models into digital systems.
Main topics of the course:
Part 1 of the course lays the foundations of ML engineering:
1. The role of ML in the architecture of digital systems;
2. Task classification (classification, regression, clustering, etc.);
3. Data work: analysis, cleaning, normalisation, feature encoding;
4. Assessment of model quality and generalisation ability;
5. Overfitting and underfitting, bias‑variance trade‑off;
6. Basic models: linear and logistic regression, decision trees, ensembles (Random Forest, gradient boosting);
7. Unsupervised learning and dimensionality reduction (clustering, PCA);
8. Introduction to neural networks: perceptron, multilayer networks (MLP), fundamentals of training (gradient descent, backpropagation);
9. Overview of deep learning architectures (CNNs, RNNs) and their relation to data type and task;
10. ML model lifecycle and the concept of an ML pipeline.
Part 2 of the course deepens knowledge in deep learning and system design:
1. Convolutional neural networks (CNNs) and computer vision (ResNet, EfficientNet);
2. Sequential data and recurrent networks (RNNs, LSTMs);
3. Attention mechanism and Transformer architecture;
4. Pre‑trained models, transfer learning, and fine‑tuning;
5. Foundational models as the basis of modern AI systems;
6. Reproducibility and experiment management;
7. Deploying models into digital systems (serialisation, API, batch and online inference);
8. Performance optimisation: quantisation, distillation, architecture simplification;
9. Accounting for operational constraints (latency, throughput, memory consumption).
