RIA: Learning Case Adaptation for Case-Base Reasoning

9409348 Leake基于案例的推理（CBR）系统从经验中推理：它们通过检索相关的先前案例来解决新问题，并执行案例适应知识，这些知识通常由系统开发人员提供，为单个任务手工编码。有效适应规则编码的困难被广泛认为是阻碍基于案例推理系统发展的一个严重问题。本研究通过开发一种自动学习模型来提高案例适应性来解决这个问题。该模型学习记忆搜索过程来实现一般自适应规则的操作。该模型从学习记忆搜索案例开始，跟踪所使用的记忆搜索过程及其结果。这些记忆搜索案例反映了特定的记忆搜索策略对特定适应性的适用性，检索学习到的记忆搜索案例为记忆搜索提供了特定的指导。通过一系列实验，包括“消融”研究和人类受试者适应质量的直接评估，评估了学习过程对适应绩效的影响。

9409348 Leake Case-based reasoning (CBR) systems reason from experience: They solve new problems by retrieving relevant prior cases and performing case adaptation knowledge is normally provided by the system developer, hand-coded for a single task. The difficulty of encoding effective adaptation rules is widely recognized as a serious impediment to the development of case-based reasoning systems. This research addresses that problem by developing a model of automatic learning to improve case adaptation. The model learns memory search procedures to operationalize general adaptation rules. The model starts results in learning memory search cases tracing the memory search processes used and their results. Those memory search cases reflect the applicability of particular memory search strategies to particular adaption, learned memory search cases are retrieved to provide specific guidance for memory search. The effects of the learning process on adaptation performance are evaluated by a series of experiments including "ablation" studies and direct assessment of the quality of the adaptations by human subjects.