一种逻辑强化学习的tableau推理方法  

Tableau reasoning method based on logical reinforcement learning

作　　者：刘全[1] 崔志明[1] 高阳[2] 陈道蓄[2] 姚望舒[1] 

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006 [2]南京大学软件新技术国家重点实验室,江苏南京210093

出　　处：《智能系统学报》2008年第4期355-360,共6页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60673092,60775046);教育部重点资助项目(207040);中国博士后科研基金资助项目(20060390919);江苏省高校自然科学基金资助项目(06KJB520104);江苏省博士后科研基金资助项目(060211C);江苏省现代企业信息化应用支撑软件工程技术研究中心开发项目(SX200804)

摘　　要：tableau方法是一种具有较强的通用性和适用性的推理方法,但由于函数符号、等词等的限制,使得自动推理具有不确定性.针对tableau推理中封闭集合构造过程具有盲目性的问题,提出将强化学习用于tableau自动推理的方法.该方法将tableau推理过程中的逻辑公式与强化学习相结合,产生抽象的状态和活动.这样一方面可以通过学习方法控制自动推理的推理顺序,形成合理的封闭分枝,减少推理的盲目性;另一方面复杂的推理可以利用简单的推理结果,提高推理的效率.The tableau method is a reasoning method with high universality and applicability. However, given the restrictions of function symbols and equations, there remains a great deal of uncertainty in automated reasoning. In order to remove blind reasoning in the construction of a closed set for tableau reasoning, a method was developed to introduce reinforcement learning into tableau reasoning. Reinforcement learning was combined with the logical formulae in tableau reasoning to produce abstract states and actions. On the one hand, reasoning sequences in auto reasoning can be controlled by the learning method to form reasonable closed branches and reduce the blindness of reasoning. On the other hand, simple reasoning results can be reused in the complex reasoning system to improve reasoning efficiency.

关 键 词：逻辑强化学习 tableau推理 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]