大数据环境下的增强学习综述  

An Overview of Reinforcement Learning in Big Data

作　　者：仵博[1] 冯延蓬[1] 孟宪军[1] 江建举[1] 何国坤[1] 

机构地区：[1]深圳职业技术学院教育技术与信息中心,广东深圳518055

出　　处：《深圳职业技术学院学报》2014年第3期71-75,共5页Journal of Shenzhen Polytechnic

基　　金：广东省自然科学基金项目(S2011040004769);深圳市科技研发资金项目(JCYJ20120617134831736)

摘　　要：在大数据应用领域,如何快速地对海量数据进行挖掘是当前大数据应用基础研究的热点和难点,也是制约大数据真正应用的关键.而机器学习是解决该问题的有效途径,本文综述抽象增强学习、可分解增强学习、分层增强学习、关系增强学习和贝叶斯增强学习等五类增强学习方法的研究进展,分析了它们的优势和缺点,指出将监督学习或半监督学习与增强学习相结合是大数据机器学习的有效方法.In the field of big data application, processing the huge amount of data is an issue of great concern and a hard nut to crack in big data application basic research. It is also the main factor that affects the application of big data. Nevertheless, machine learning offers an effective approach to solving this problem. This paper reviews the research on abstract reinforcement learning, factored reinforcement learning, hierarchical reinforcement learning, relational reinforcement learning, and Bayesian reinforcement learning, analyzes their advantages and disadvantages respectively, and points out that combining supervised learning or semi-supervised learning with reinforcement learning is an effective method for machine learning in big data.

关 键 词：大数据 增强学习 维数灾 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]