Basics of Reinforcement Learning

Description

The remarkable recent advances in Large Language Models (LLMs) and autonomous AI systems are strongly connected to a powerful training framework known as Reinforcement Learning (RL). This approach empowers AI systems to learn optimal strategies through reward-driven interactions, combining mathematical elegance with practical utility. In this project, students will explore foundational concepts such as  value and policy optimization, and deep reinforcement learning, gaining both theoretical insight and practical skills essential for building intelligent, adaptive systems.

Prerequisites

Strong knowledge of Python. Familiarity with basic probability theory.

Some background material

• Sergey Levine's CS 285 "Deep Reinforcement Learning" (UC Berkeley) -- A complete course (including Fall 2022/2021 YouTube playlists and slides) covering core RL theory and deep learning methods
• MIT 6.S191 (2023/2024) -- Deep reinforcement learning lecture by Alexander Amini
• Stanford CS234 (Winter 2019) -- Comprehensive RL playlist on algorithms & theory
• Sutton & Barto's Reinforcement Learning: An Introduction -- The go-to textbook for formal RL underpinnings.