Policy Reuse

Policy reuse in reinforcement learning aims to leverage previously learned policies to accelerate learning in new, related tasks, thereby improving efficiency and reducing computational costs. Current research focuses on developing algorithms that effectively select and adapt existing policies, with prominent approaches including deep reinforcement learning architectures incorporating critic-guided selection or Bayesian inference to estimate task similarity and choose appropriate source policies. These advancements are significant for improving the sample efficiency of reinforcement learning agents, particularly in complex domains like quantum error correction and other applications where data acquisition is expensive.

Papers

December 22, 2022

Decoding surface codes with deep reinforcement learning and probabilistic policy reuse
Elisha Siddiqui Matekole, Esther Ye, Ramya Iyer, Samuel Yen-Chi Chen
Deep Reinforcement Learning Quantum Computing Intermediate Scale Quantum Code Data Scalable Quantum Policy Reuse

October 15, 2022

CUP: Critic-Guided Policy Reuse
Jin Zhang, Siyuan Li, Chongjie Zhang
Deep Reinforcement Learning Model Based Policy Policy Reuse

April 16, 2022

Efficient Bayesian Policy Reuse with a Scalable Observation Model in Deep Reinforcement Learning
Jinmei Liu, Zhi Wang, Chunlin Chen, Daoyi Dong
Deep Reinforcement Learning Task Inference Scalable Model Policy Transfer Policy Reuse

Policy Reuse

Papers

Decoding surface codes with deep reinforcement learning and probabilistic policy reuse

CUP: Critic-Guided Policy Reuse

Efficient Bayesian Policy Reuse with a Scalable Observation Model in Deep Reinforcement Learning