Faramarz Fekri

About

Faramarz Fekri’s research lies at the intersection of reinforcement learning (RL) and variational inference, with a focus on improving the efficiency and optimality of actor-critic methods for robotic control. His major contributions include pioneering the use of Hamiltonian Monte Carlo (HMC) methods to address fundamental limitations in actor-critic RL, such as the amortization gap and insufficient exploration. By reframing actor-critic RL as a variational inference problem, Fekri demonstrates that standard approaches can yield suboptimal policy estimates; his work introduces HMC to generate more accurate posterior approximations of actions, leading to better exploration and more robust policy learning. His two most-cited papers on this topic—published in 2022—have collectively garnered 10 citations, signaling growing interest in his innovative synthesis of Bayesian inference and RL. This work is particularly notable for its potential to enhance real-world robotic control tasks, where sample efficiency and policy quality are critical. Fekri’s research offers a fresh theoretical perspective that bridges probabilistic modeling and deep RL, making it a valuable resource for students and researchers seeking to push the boundaries of autonomous decision-making.

Research Focus

Key Achievements

Top Papers

1. 1
   Improving Actor–Critic Reinforcement Learning via Hamiltonian Monte Carlo Method

   6 citations · 2022
2. 2
   Improving Actor-Critic Reinforcement Learning Via Hamiltonian Monte Carlo Method

   4 citations · 2022

Key Collaborators

Contact & Links