Kasra Rezaee

About

Kasra Rezaee is an emerging researcher specializing in inverse reinforcement learning (IRL) and safe reinforcement learning, with a particular focus on the challenging problem of constraint inference from expert demonstrations. His work addresses a critical gap in deploying reinforcement learning agents in real-world settings: the difficulty of mathematically specifying the constraints that govern safe and acceptable behavior. Rezaee's most notable contribution, "Learning Soft Constraints From Constrained Expert Demonstrations" (2022), advances the field by recognizing that expert agents often optimize reward functions subject to implicit constraints — constraints that shape behavior in ways traditional IRL methods fail to capture. This insight is practically significant for building AI systems that can infer not just what an expert is trying to achieve, but also what boundaries they are operating within. Complementing this, his benchmarking work on constraint inference in IRL provides the research community with standardized evaluation frameworks, an often-overlooked but essential contribution for measuring progress in the field. Though early in his career with a growing citation record, Rezaee's research tackles problems with direct implications for robotics, autonomous systems, and AI safety — areas of rapidly expanding importance.

Research Focus

Key Achievements

Top Papers

1. 1
   Learning Soft Constraints From Constrained Expert Demonstrations

   4 citations · 2022
2. 2
   Benchmarking Constraint Inference in Inverse Reinforcement Learning

   2 citations · 2022

Key Collaborators

Contact & Links