论文部分内容阅读
文章分析了FOIL（first－orderinductivelearner）递归谓词学习算法理论上的不足以及由此导致的应用范围的局限，并通过两个例子给予详细说明．为了克服这一缺陷，文章引入了反映递归规则集R与实例空间E本质关系的实例图H（R．E）和实例序的概念，奠定了算法的理论基础．在此基础上，给出了基于实例图的FOILPlus算法．算法通过对悬例、悬弧的操作把握住实例序，自然而然地防止了病态递归规则的产生，从而保证了FOILPlus可以不受常量序限制地完成学习任务；同时，算法的时空复杂度较之FOIL算法没有增加．FOILPlus算法已经编程实现，并用它尝试了两个FOIL学习失败的递归任务，都获得了成功．
The paper analyzes the theoretical deficiencies of the recursive predicate learning algorithm of FOIL (first-orderinductivelearner) and the limitation of the application range of the algorithm. The two examples give a detailed explanation. In order to overcome this shortcoming, the article introduces the concept of instance graph H (R.E) and instance sequence that reflect the essential relationship between recursive rule set R and instance space E, which lays the theoretical foundation of the algorithm. On this basis, the FOILPlus algorithm based on the instance graph is given. The algorithm grasps the sequence of instances by suspending and suspending arc operation, and naturally avoids the generation of morbid recursion rules, thereby ensuring that FOILPlus can accomplish learning tasks without being limited by the order of constants. Meanwhile, the space-time complexity of the algorithm is better than that of FOIL The algorithm did not increase. The FOILPlus algorithm has been programmed and used it to try two recursive tasks that FOIL failed to learn. Both were successful.