Reinforcement Learning for Sequential Decision Making Problems

Pooja Sanjay Gore, B.E.

Diplomová práce

Abstract:

This thesis examines model-free reinforcement learning, focusing on how parameters like learning rate (α), discount factor (γ), and exploration strategies affect an agent's performance in a simple grid-world environment. Using Q-learning within the Markov Decision Process framework, it explores how these factors influence decision-making and reward optimization. The study finds that α = 0.3 and γ = 0.99 provide the best balance for efficient learning and adaptability. It highlights the importance of balancing exploration and exploitation, offering insights for robotics, autonomous systems, and adaptive controls, and enhancing the understanding of intelligent decision-making in complex situations.

Keywords

Reinforcement Learning, Decision-making Problems, Exploration Dilemma, Machine Learning, Artificial Intelligence

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 3. 12. 2024

Obhajoba závěrečné práce

• Obhajoba proběhla 5. 2. 2025
• Vedoucí: doc. Ing. Tatiana Guy, PhD.
• Oponent: Miroslav Kárný, externi

Česká zemědělská univerzita v Praze

Provozně ekonomická fakulta

Magisterský studijní program:
Informatics