Strategy Entropy as a Measure of Strategy Convergence in Reinforcement Learning

• Authors: Xiaodong Zhuang;Zhuo Chen
• Venue: ICINIS '08 Proceedings of the 2008 First International Conference on Intelligent Networks and Intelligent Systems
• Year: 2008

The concept of entropy is introduced into reinforcement learning. The definitions of the local and global strategy entropy are presented. The global strategy entropy is experimentally proved to be the quantitative problem-independent measure of the strategy’s convergence degree. The experimental results show that the learning based on the local strategy entropy improves the learning performance.