Multiplied Action Values for Complex-valued Reinforcement Learning
This paper proposes a technique of multiplied action values to complex-valued reinforcement learning. This technique aims to realize a learning agent which works well even in environments involving repeated perceptual aliasing. The complex-valued reinforcement learning has been proposed for solving perceptual aliasing. Perceptual aliasing occurs when a learning agent equips only poor sensors. This problem is known as one of central problems in reinforcement learning. Ordinal complex-valued reinforcement learning approach has achieved success in simple environments. However, it is not enough to develop useful agents in real environments. This paper focuses environments which contain repeated perceptual aliasing. and proposes multiplied action valued to enhance the learning performance of the agent. Some simulation experiments are conducted to evaluate the performance of proposed method. The experimental results showed that the proposed method has possibility to develop suitable behavior in such simulation environments. The class of tasks which the proposed method can solve is also discussed.
reinforcement learning perceptual aliasing
Takeshi Shibuya Tomoki Hamagami
Graduate School of Engineering Yokohama National University Yokohama, Japan
国际会议
The International Conference on Electrical Engineering 2009(2009 电机工程国际会议)
沈阳
英文
1-6
2009-07-05（万方平台首次上网日期，不代表论文的发表时间）