Hybrid Reinforcement Learning
Hybrid reinforcement learning (RL) combines offline data with online interaction to improve the efficiency and robustness of RL agents. Current research focuses on developing algorithms that leverage both data sources effectively, often employing model architectures like actor-critic methods and hierarchical RL, and addressing challenges such as limited data coverage and mixed-variable action spaces. This approach holds significant promise for applications ranging from robotics and resource optimization to personalized medicine and AI safety, by enabling faster training, improved generalization, and more reliable performance in complex real-world scenarios.
Papers
September 14, 2024
August 23, 2024
August 8, 2024
June 28, 2024
June 11, 2024
May 30, 2024
March 13, 2024
March 7, 2024
November 14, 2023
October 13, 2023
May 17, 2023
February 9, 2023
November 4, 2022
October 13, 2022
July 22, 2022
June 25, 2022