Pareto Frontier Estimation in Offline Safe Reinforcement Learning – Bc. Václav Nevyhoštěný
Bc. Václav Nevyhoštěný
Master's thesis
Pareto Frontier Estimation in Offline Safe Reinforcement Learning
Pareto Frontier Estimation in Offline Safe Reinforcement Learning
Abstract:
Tato diplomová práce zkoumá offline bezpečné posilované učení (OSRL), kde se agenti musí ze fixních datasetů naučit efektivní rozhodovací strategie a současně dodržet bezpečnostní požadavky. Práce obsahuje přehled existujících přístupů a zdůrazňuje jejich limitace při vyvažování maximalizace odměny a striktního dodržování bezpečnostních požadavků. Následně představujeme Pareto-regularizovaný decision …moreAbstract:
This thesis investigates offline safe reinforcement learning (OSRL), where agents must learn effective decision-making policies from fixed datasets while satisfying safety constraints. We review existing approaches and highlight fundamental limitations in their ability to balance reward maximization with strict constraint satisfaction. To address these shortcomings, we introduce the Pareto-regularized …more
Language used: English
Date on which the thesis was submitted / produced: 16. 12. 2025
Identifier:
https://is.muni.cz/th/s7fnz/
Thesis defence
- Date of defence: 2. 2. 2026
- Supervisor: doc. RNDr. Petr Novotný, Ph.D.
- Reader: doc. RNDr. Tomáš Brázdil, Ph.D.
Citation record
Full text of thesis
Contents of on-line thesis archive
Published in Theses:- světu
Other ways of accessing the text
Institution archiving the thesis and making it accessible: Masarykova univerzita, Fakulta informatikyMasaryk University
Faculty of InformaticsMaster programme / field:
Theoretical computer science / Discrete algorithms and models
Theses on a related topic
-
Revisiting Uncertainty Quantification for Offline Reinforcement Learning
Vít Unčovský -
A Regularized Actor-Critic Algorithm for Safe Reinforcement Learning with Pareto Duality
Hugo Adamove -
Deep Reinforcement Learning for Decision Neuroscience
Faizanshaikh Abdulkhalil SHAIKH -
Reinforcement Learning of Risk-Averse Policies in Markov Decision Processes
Jiří Vahala -
Reinforcement Learning for Sequential Decision Making Problems
Pooja Sanjay Gore -
A Regularized Actor-Critic Algorithm for Safe Reinforcement Learning with Pareto Duality
Hugo Adamove -
Risk-Sensitive Reinforcement Learning
Marek Kadlčík -
Machine learning techniques of de-novo molecular formula reconstruction from its mass spectrum
Adam Hájek