Expert Policy

Expert policy learning in reinforcement learning aims to efficiently acquire optimal behavior from expert demonstrations, often imperfect or limited. Current research focuses on improving data efficiency by leveraging techniques like counterfactual data augmentation, enhanced action spaces (e.g., using macro-actions), and discriminator-weighted learning to filter out suboptimal demonstrations. These advancements address challenges in handling suboptimal data and improve generalization to unseen environments, ultimately leading to more robust and efficient reinforcement learning agents applicable to complex real-world tasks.

Papers

October 7, 2023

Offline Imitation Learning with Variational Counterfactual Reasoning
Bowei He, Zexu Sun, Jinxin Liu, Shuai Zhang, Xu Chen, Chen Ma
Counterfactual Prediction Counterfactual Inference Offline Imitation Offline Imitation Learning Variational AutoEncoder Expert Augmentation Expert Policy

December 7, 2022

EASpace: Enhanced Action Space for Policy Transfer
Zheng Zhang, Qingrui Zhang, Bo Zhu, Xiaohan Wang, Tianjiang Hu
Action Space Policy Transfer Expert Policy

July 20, 2022

Discriminator-Weighted Offline Imitation Learning from Suboptimal Demonstrations
Haoran Xu, Xianyuan Zhan, Honglei Yin, Huiling Qin
Offline Reinforcement Learning Reward Learning Suboptimal Demonstration Offline Imitation Learning Expert Policy

May 23, 2022

Efficient Reinforcement Learning from Demonstration Using Local Ensemble and Reparameterization with Split and Merge of Expert Policies
Yu Wang, Fang Liu
Reinforcement Learning Policy Gradient Split Learning Textual Demonstration Efficient Reinforcement Learning State Action Pair Reparameterization Method Suboptimal Demonstration Expert Policy

February 8, 2022

Local Explanations for Reinforcement Learning
Ronny Luss, Amit Dhurandhar, Miao Liu
Reinforcement Learning Deep Reinforcement Learning Local Explanation Expert Policy

July 7, 2021

Bandits with Stochastic Experts: Constant Regret, Empirical Experts and Episodes
Nihal Sharma, Rajat Sen, Soumya Basu, Karthikeyan Shanmugam, Sanjay Shakkottai
Empirical Analysis Confidence Bound Contextual Bandit Problem Uncertain Episode Length Tight Regret Expert Distribution Expert Policy