Математика обучения с подкреплением

2026/2027

Статус: Маго-лего

Кто читает: Департамент больших данных и информационного поиска

Где читается: Факультет компьютерных наук

Охват аудитории: для своего кампуса

Преподаватели: Левин Илья Валерьевич, Наумов Алексей Александрович

Язык: русский

Кредиты: 3

Контактные часы: 28

Дополнительные материалы в LMS Задать вопрос

Аннотация

Reinforcement learning is a type of machine learning. The key feature of this method, unlike classical machine learning, is the interaction of the agent (algorithm) with an environment from which he receives feedback in the form of rewards. The agent's goal is to maximize the sum of rewards that the environment gives him for the "right" interaction. During the course, we will get acquainted with the basic concepts of reinforcement learning theory, talk about the exploration of the environment and the paradigm of optimism. We also study modern reinforcement learning algorithms such as TRPO, PPO, and entropic reinforcement learning, which are widely used, for example, in modern methods of training large language models.

Цель освоения дисциплины

Understanding mathematical fundamentals of modern reinforcement learning methods.

Планируемые результаты обучения

The ability to apply mathematical methods for the analysis of reinforcement learning algorithms
The ability to construct the effective algorithms
Developing the skill of problem-solving and understanding algorithms