Dr. Erman Acar - Collaboration and Safety in MARL through Probabilistic Logic Shields

Event details
-
Monday 18 August 2025 - 2:00pm to 3:00pm
Description
Speaker: Dr. Erman Acar from the University of Amsterdam.
Date: 18th August, 2025 - 2:00 PM
Location: Lewin Lab, Ground Floor, Regent Court Building (also Google Meet)
Date: 18th August, 2025 - 2:00 PM
Location: Lewin Lab, Ground Floor, Regent Court Building (also Google Meet)
Title: Collaboration and Safety in MARL through Probabilistic Logic Shields
Abstract: Safe reinforcement learning (RL) is crucial for real-world applications, and multi-agent interactions introduce additional safety challenges. While Probabilistic Logic Shields (PLS) has been a powerful proposal to enforce safety in single-agent RL, their generalizability to multi-agent settings remains pretty much unexplored. In this talk,I will touch upon this gap by presenting our recent work Shielded-MARL (SMARL) for steering agents to norm-compliant outcomes, conducting extensive analyses of PLS within decentralized, multi-agent environments. Our contributions include: (1) a novel Probabilistic Logic Temporal Difference (PLTD) update for shielded, independent Q-learning, which incorporates probabilistic constraints directly into the value update process; (2) a probabilistic logic policy gradient method for shielded PPO with formal safety guarantees for MARL; and (3) comprehensive evaluation across symmetric and asymmetrically shielded n-player game-theoretic benchmarks, demonstrating fewer constraint violations and significantly better cooperation under normative constraints.
Biography: Erman Acar is an assistant professor for Explainable AI in Finance at the Institute for Logic, Language and Computation, and at the Informatics Institute at University of Amsterdam (UvA) where he is part of Socially Intelligent Artificial Systems Group/Civic AI Lab, and the Cognition, Language and Computation Lab. He has an ongoing research on a broad range of topics, including Neurosymbolic AI,Causality and Multiagent systems, and their applications in financial technologies.Besides his positions, he is also part of the management team of the Hybrid Intelligence Center (7 Universities, 80 PhD students) and a founding co-manager of AI4FinTech initiative at the University of Amsterdam. Erman Acar obtained his PhD in AI from University of Mannheim and worked as a postdoctoral researcher at Vrije Universiteit Amsterdam (KRR group), and Leiden University (RL group). He has been a visiting researcher to various institutions, including University of Bozen-Bolzano, University of Calabria and the University of Oxford.
Biography: Erman Acar is an assistant professor for Explainable AI in Finance at the Institute for Logic, Language and Computation, and at the Informatics Institute at University of Amsterdam (UvA) where he is part of Socially Intelligent Artificial Systems Group/Civic AI Lab, and the Cognition, Language and Computation Lab. He has an ongoing research on a broad range of topics, including Neurosymbolic AI,Causality and Multiagent systems, and their applications in financial technologies.Besides his positions, he is also part of the management team of the Hybrid Intelligence Center (7 Universities, 80 PhD students) and a founding co-manager of AI4FinTech initiative at the University of Amsterdam. Erman Acar obtained his PhD in AI from University of Mannheim and worked as a postdoctoral researcher at Vrije Universiteit Amsterdam (KRR group), and Leiden University (RL group). He has been a visiting researcher to various institutions, including University of Bozen-Bolzano, University of Calabria and the University of Oxford.
Location
53.381144548765, -1.480044177989
When focused, use the arrow keys to pain, and the + and - keys to zoom in/out.