Abstraction and Hierarchical Problem Solving in Case-Based Reasoning

基于案例推理的抽象和分层问题解决

• 批准号: 9502152
• 负责人: Luther Branting
• 金额: $ 13.41万
• 依托单位: University of Wyoming
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-09-01 至 1998-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9502152&HistoricalAwards=false
• 关键词: Abstraction; Hierarchical; Problem; Solving; Case

The research component involves case-based reasoning (CBR) a technique that stores and later applies cases of known problems and their solutions to solve new problems. Since few stored cases will exactly match a new problem, most cases will be related at best by analogy to the new one. In that case, it is necessary first to retrieve an appropriate analogous case, check to see that it matches the new problem in some essential ways, then adapt it to the new problem in such a way as to infer a solution. In storing and using cases, it is necessary to do some abstraction, either at the time that they are stored away for later use or at the time retrieved, to avoid having to store large numbers of very similar but slightly different cases. This project examines the advantages of developing a hierarchy of abstraction for each stored case, a technique called "stratified CBR". The stratified CBR technique will be applied to a typical search task (route planning), two configuration tasks (artifact design and goal-directed activity planning), and an analytical task (legal reasoning), in order to test hypotheses about the superiority of the stratified CBR technique and to gain experience in using the technique. During the period of the grant, an educational plan will be developed that exposes students to the insights, perspectives, and tools of AI by (1) educating undergraduates about AI's basic methods and concepts and about AI's philosophical, social, and cognitive consequences, and (2) educating graduate students in other fields about the AI techniques and tools applicable to their fields.

研究部分涉及基于案例的推理（CBR），这是一种存储已知问题的案例并在以后应用它们的解决方案来解决新问题的技术。 由于很少有存储的案例与新问题完全匹配，因此大多数案例最多是通过类比与新问题相关。 在这种情况下，首先需要检索一个适当的类似 在一个案例中，检查它是否在某些基本方面与新问题相匹配，然后将它调整为适合新问题，从而推断出解决方案。 在存储和使用案例时，有必要进行一些抽象，或者在存储案例以备将来使用的时候，或者在检索案例的时候，以避免存储大量非常相似但略有不同的案例。 这个项目研究了为每个存储的案例开发抽象层次结构的优点，这是一种技术， 称为“分层CBR”。 将分层CBR技术应用于典型的搜索任务（路径规划），两种配置 任务（工件设计和目标导向的活动 规划）， 和 分析 任务 （法律的推理）、 在 检验假设的顺序 关于 总结了分层CBR技术的优越性和应用经验。在拨款期间，将制定一项教育计划，通过以下方式让学生接触到人工智能的见解、观点和工具：（1）教育本科生了解人工智能的基本方法和概念，以及人工智能的哲学、社会和认知后果，以及（2）对其他领域的研究生进行适用于其领域的人工智能技术和工具的教育。