Introduction: PRICAI 2016 is the International Conference on Artificial Intelligence at the Pacific Rim and is held every two years. The conference focuses on the theory of artificial intelligence, technology and its application in the social field, and its importance to the economies of Pacific Rim countries.
Exploring Multi-action Relationship in Reinforcement Learning
Abstract: Many reinforcement learning problems in real life require agents to control multiple actions at the same time. Learning in this case, before, each action is usually handled separately from other actions. However, in an application where multiple actions are rarely performed independently, and the potential relationships between actions are used, it may help speed up learning. This article explores the relationship between multiple actions in reinforcement learning. We propose to implement a regular term to capture the relationship between multiple actions. We embody the regularization term into the least-squares strategy iteration and the time-domain difference method, which effectively solves the convex learning objective. The proposed method has been proven effective in several fields. Experimental results show that the relationship between concrete and multiple actions can effectively improve learning performance.
About the Author
Yu Yang
E-mail:.cn
Position: Associate Professor, Department of Computer Science and Technology, Nanjing University/LAMDA Group
Research Interests: Artificial Intelligence, Evolutionary Machine Learning, Reinforcement Learning
Related academic papers:
·High-dimensional derivative-free optimization
·Pareto optimization
Han Wang
E-mail:.cn
Position: MSc, Department of Computer Science and Technology, Nanjing University / LAMDA Group
Research: Machine Learning, Data Mining, Reinforcement Learning
Via:PRICAI 2016
PS : This article was compiled by Lei Feng Network (search “Lei Feng Network†public number) and it was compiled without permission.
Original paper download