Learning Relations Using Genetic Algorithms
Learning Relations Using Genetic Algorithms
复制标题
使用遗传算法学习关系
DOI：
--
复制
发表时间：
1993
期刊：
影响因子：
--
通讯作者：
G. L. Bello
中科院分区：
文献类型：
--
作者：
A. Giordana;L. Saitta;M. Campidoglio;G. L. Bello
从例子中归纳概念描述需要探索大量的假设空间。遗传算法由于其多点搜索策略，为传统搜索算法提供了一个有吸引力的替代方案。本文描述了一个新的系统REGAL：它使用遗传算法来学习一阶逻辑概念描述。此外，它可以很容易地与演绎成分集成，以利用一个领域理论。提出了两种学习析取概念描述的方法：第一种方法是对经典的一次学习一个析取概念描述方法的改进，第二种方法是基于适应度共享的思想，根据生态位和物种的理论，尝试让亚种群自发形成。在人工领域上对这两种方法进行了比较。
Inducing concept descriptions from examples requires a large space of hypotheses to be explored. Genetic algorithms offer an appealing alternative to traditional search algorithms, because of their multi-point search strategy. In this paper, the new system REGAL is described: it uses genetic algorithms to learn first order logic concept descriptions. Moreover, it can be easily integrated with a deductive component, in order to exploit a domain theory. Two approaches to learning disjunctive concept descriptions are presented: the first one is a modification of the classical method of learning one disjunct at a time, whereas the second one is based on the idea of fitness sharing and tries to let subpopulations be spontaneously formed, according to the theory of the niches and species. The approaches have been compared on an artificial domain.