| |
 Safe Planning and Reinforcement Learning with Monte Carlo Tree Search Speaker (s):
Petr Novotný
Associate Professor
Masaryk University, Brno,
Czech Republic
| Date: Time: Venue: | | 12 June 2025, Thursday 2:00pm – 3:00pm School of Computing &
Information Systems 2 (SCIS 2)
Level 4, Seminar Room 4-3
Singapore Management University
90 Stamford Road,
Singapore 178903 Please register by 11 June 2025.  |
|
About the Talk In this talk, we will study the problem of controlling an agent so as to maximize the agent’s expected utility while ensuring that the agent’s policy does not violate a given safety constraint. Problems of this form are typically modelled as constrained Markov decision processes (CMDPs). A number of both exact and approximate algorithms for CMDP optimization were developed over the last decade. In this talk, we will focus on algorithms employing the Monte Carlo tree search (MCTS) paradigm (of Alpha- and MuZero fame), due to their straightforward extensibility to partially observable and multi-agent settings. We will consider MCTS-based CMDP solvers in several contexts, ranging from algorithms that can provide strict formal guarantees on the quality of agent's behaviour to heuristic techniques that can safely control agents in environments with an astronomical number of states. About the Speaker Petr Novotný is an associate professor of computer science at Masaryk University, Brno, Czech Republic. HIs research lies at the interface of formal methods and reinforcement learning. In particular, he is interested in safe reinforcement learning, formal analysis of probabilistic programs, and in computational complexity of problems pertaining to decision making under uncertainty. |
