PhD Dissertation Proposal by BRAHMANAGE Janaka Chathuranga Thilakarathna | Reinforcement Learning Methods for Risk-Averse Decision-Making

Reinforcement Learning Methods for Risk-Averse Decision-Making

ABOUT THE TALK

Reinforcement Learning (RL) has shown impressive success across many domains, yet applying it safely in real-world, high-stakes environments remains a challenge—especially when only offline data is available. This thesis-proposal addresses this gap by proposing methods that enforce safety during offline learning while preserving performance. It introduces an action-constrained learning framework using generative models, and a conservative cost-critic approach to handle cumulative safety constraints effectively. Future directions include extending these methods to multi-agent settings, integrating natural language safety specifications.

SPEAKER BIOGRAPHY

Janaka Brahmanage is a third-year PhD candidate in Computer Science, conducting research under the guidance of Associate Prof. Akshat Kumar at the SMU School of Computing and Information Systems. His research focuses on safe reinforcement learning and multi-agent systems.