基于高阶逻辑的归纳逻辑程序设计学习算法及其应用研究

项目名称： 基于高阶逻辑的归纳逻辑程序设计学习算法及其应用研究

项目编号： No.61300098

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 李艳娟

作者单位： 东北林业大学

项目金额： 25万元

中文摘要： 归纳逻辑程序设计（Inductive logic programming，简称ILP）采用一阶逻辑表示经验数据和学习到的规则，克服了传统机器学习方法描述能力弱和无法利用背景知识的限制，近年来，顶级国际刊物《机器学习》相继出版了多期ILP的专刊，ILP逐渐成为机器学习领域的前沿研究课题。本研究进一步提高ILP的表达能力，采用高阶逻辑表示经验数据和学习到的规则，研究基于高阶逻辑的ILP学习算法和应用。首先，根据搜索策略不同研究确定性高阶ILP算法和非确定性高阶ILP算法。然后，针对实际应用中对数据标记的代价很高，容易获得的是无标记数据，在上述两个算法的基础上，研究利用无标记数据提高高阶ILP算法的分类性能。最后，将高阶ILP算法应用于图像语义分类系统中，采用高阶逻辑表示图像内部的空间关系。该研究为机器学习方法提供了更加深入的理论和方法，为人工智能的应用领域提供了强有力的技术支持。

中文关键词： 机器学习；归纳逻辑程序设计；人工蜂群算法；；

英文摘要： By using first-order logic to represent empirical data and learned rules, Inductive logic programming (ILP for short) overcomes two limitations of classical machine learning: a limited knowledge representation formalism which is essentially propositional logic and inability to use substantial background knowledge in the learning process. In recent years, the first-class international journal "machine learning" has published several special issues on ILP. ILP has been a hot topic of machine learning. The research further improves the expressive ability of ILP, adopts higher-order logic to represent empirical data and learned rules, and studies higher-order logic based ILP learning algorithm and its application. Firstly, according to search strategy, determined ILP algorithm and stochastic ILP algorithm are studied. Secondly, in practical applications, unlabeled data are readily available but labeled data are fairly expensive to obtain because they require human effort. Based on the two algorithms proposed above, the project investigates how to exploit unlabeled data to enhance classification performance of higher-order ILP algorithm. Finally, higher-order ILP algorithm is applied to image semantic classification system adopting higher-order logic to represent space relation. This research provides machine learnin

英文关键词： machine learning；inducitve logic programming；artificial bee colony；；