专家系统组织导论

• 作 者: MarkStefik JanAikins RobertBalzer JohnBenoit LawrenceBirnbaum FrederickHayes-Roth EarlSacerdoti 中原
• 出 处: 计算机科学
• 发表日期: 1985年03期
• 关键词: 专家系统 专家问题求解 推理技术 子任务 组合规则 穷举搜索 搜索空间 不精确推理方法 先验概率 知识表示

本导论介绍专家问题求解程序的组织。作者从限定的一类问题出发,引入特别简单的组织。为了使专家系统组织可行,要求输入数据是静态的,可靠的,也要求解的空间足够小,能够穷举搜索。这样,在十个更复杂组织的研究中,每次逐渐放宽这些假设。第一种情形给出处理不可靠数据和时变(time-varying)数据的技术。其它情形给出抽象解空间的建立与推理技术,和使用多路推理的技术。最后,就这些技术所涉及的范围比较了这些组织方法,并以当前专家系统的若干实例予以说明。

This introduction introduces the organization of the expert problem solver. The author starts with a limited set of questions and introduces a particularly simple organization. In order to make the expert system organization feasible, the input data is required to be static and reliable, and the space required for solution should be small enough to search exhaustively. In this way, these ten assumptions are gradually relaxed in each of the ten more complex organizational studies. The first case gives techniques to deal with unreliable data and time-varying data. Other cases give the construction and reasoning of abstract solution spaces and the techniques of using multiple inferences. Finally, these organizational approaches are compared in terms of the extent to which these technologies are covered and illustrated by several examples of current expert systems.