智能技术学报

文章详情

稿件标题: Interactions Between Agents: the Key of Multi Task Reinforcement Learning Improvement for Dynamic Environments
稿件作者: Sadrolah Abbasi, Hamid Parvin, Mohamad Mohamadi, Eshagh Faraji
关键字词: Interactions between agents; multi task reinforcement learning; evolutionary learning; chaotic exploration; dynamic environments
文章摘要: In many Multi Agent Systems, learner agents explore their environments to find their goal and agents can learn their policy. In Multi-Task Learning, one agent learns a set of related problems together at the same time, using a shared model. Reinforcement Learning is a useful approach for an agent to learn its policy in a nondeterministic environment. However it is considered as a time-consuming algorithm in dynamic environments. It is helpful in multi task reinforcement learning, to use teammate agents’ experience by doing simple interactions between each other. To improve performance of multi-task learning in a nondeterministic and dynamic environment, especially for dynamic maze problem, we use the past experiences of agents. Interactions are simulated by operators of evolutionary algorithms. Also it is switched to a chaotic exploration instead of a random exploration. Applying Dynamic Chaotic Evolutionary Q-Learning to an exemplary maze, we reach significantly promising results.
收录刊物: 2017年2卷3期
稿件基金:
浏览次数: 98
下载次数: 43
点击下载