Магистратура 2025/2026

Нейронные сети и глубокое обучение

Статус: Курс по выбору (Бизнес-аналитика и системы больших данных)

Кто читает: Департамент бизнес-информатики

Где читается: Высшая школа бизнеса

Охват аудитории: для своего кампуса

Язык: английский

Контактные часы: 48

Course Syllabus

Abstract

This course introduces Master's students in Business Informatics—especially those with non-technical backgrounds—to neural networks and deep learning through an intuitive, low-code lens. Emphasizing practical training over coding expertise, it leverages ChatGPT for prompt engineering, code generation, and natural-language explanations to demystify algorithms like MLPs, CNNs, RNNs, and transformers. Students will explore forward/backpropagation, model architectures, and applications in business scenarios such as forecasting, image analysis, and anomaly detection. Outcomes include: intuitive grasp of DL processes, low-code model training for business problems, ethical roadmaps, and a final prototype project. Ultimately, the course transforms deep learning from a technical hurdle into a strategic business tool, fostering analytics-driven innovation in resource-constrained environments. cf. This course evaluates students using the normalized scores.

Learning Objectives

• Learn to effectively apply deep learning techniques to real-world business problems in computer vision, natural language processing, and tabular data
• Acknowledge ethical implications of applying machine learning in practice

Expected Learning Outcomes

• Students know advanced methods for training and regularizing neural networks.
• Владеет механизмами: multi-head attention, self-attention, маскирование, сглаживание в механизме внимания, positional encoding. Понимает ахитектуру transformer.
• Knowing how to evaluate the quality of a neural network model
• Understanding the neural network learning algorithm
• Owns the concepts of: logistic regression; gradient descent; neural networks and gradient backpropagation algorithm
• Knows optimization algorithms for deep neural networks based on various variations of gradient descent. Configures such algorithms based on the conditions of a specific task
• Able to use embedding for tabular data and recommenders
• Able to use momentum and advanced optimizers for stochastic gradient descent
• Able to use residual blocks with neural networks
• Can construct a digit classifier using a deep learning model
• Can construct a neural network from scratch
• Can construct recurrent neural network from scratch
• Can solve multi-class and multi-label problems with deep learning
• Knows advanced neural networks such as U-Net and Siamese
• Knows and uses state-of-the-art approaches to train neural networks
• Knows the definitions of deep learning
• Learn to build a neural network with one hidden layer, using forward propagation and backpropagation.
• Understands approaches to put machine learning system into production
• Understands approaches to solve natural language processing problems
• Understands data ethics and able to detect ethical problems
• Understands the role of convolutions in image processing
• • Train MLP for supervised learning taks
• Can apply generative models for solving complex tasks.
• Explain the mechanics of basic building blocks for neural networks.
• Define and train a CNN from scratch.
• Understand modern architectures of RNNs: LSTM, GRU.
• Understand the transfer learning ideas and advantages
• • understand the operations of a self-attention layer in Transformers
• Able to train autoencoders
• Владеет навыками взаимодействия с языковыми моделями, в частности, с ChatGPT, включая структурирование запросов, верификацию информации и использование различных шаблонов для оптимизации взаимодействия.
• Utilize AI Tools: Demonstrate proficiency in using no-code/low-code AI tools like KNIME and ChatGPT to develop and implement AI solutions, reducing the reliance on extensive coding skills.
• A student should formulate optimization problems as network models.

Course Contents

• Introduction to Neural Networks
• Low-Code Deep Learning with ChatGPT
• Forward Propagation and Backpropagation
• Multilayer Perceptrons and Training Basics
• Convolutional Neural Networks (CNNs)
• Recurrent Neural Networks (RNNs) and LSTMs
• Autoencoders
• Generative Models and Transfer Learning
• Transformer and Attention
• Ethics, Bias, and Optimization
• Capstone Project

Assessment Elements

• Attendance
  Students who miss more than 25% of the classes and also fail to complete the presentation assignments may fail this course.
• Exam
  Exam Question (Select All That Apply)
• Capstone Project
  Students will demonstrate practical skills; produce deployable DL solutions with business impact.
• Capstone Project
  This part evaluates if their model is working properly and reproducible.

Interim Assessment

• 2025/2026 4th module
  The final grade is calculated by weighting and summing the raw scores, followed by min-max normalization, with caps on grade distributions based on student ranks to adjust the overall distribution. The specific formulas and distribution details are as follows. Raw Score Calculation Formula The overall raw score is computed as a weighted sum: Raw Score = 0.2 × Attendance + 0.25 × Exam + 0.25 × Capstone Project + 0.3 × Presentation Min-Max Normalization Formula The raw score is normalized using the minimum (min) and maximum (max) values across all students: Normalized Score = (Raw Score - min(All Raw Scores)) / (max(All Raw Scores) - min(All Raw Scores)) × 100 Grade Distribution and Caps Normalized scores are assigned based on student rank order, with caps applied to maintain the following distribution across all students: 5 points (top tier): Up to 30% of students 4 points: Up to 40% of students 3 points or below: The remaining 30% of students

Authors

• Beklarian Armen Levonovich
• KALMYKOVA NADEZHDA SERGEEVNA
• Dzhin Seungmin