Concepts Learning based on Non-monotonic Logic

本研究的目的是设计一个使用非单调逻辑的概念学习系统。我们特别使用了TMS的显性定向回溯和麦卡锡的限制。我们在以下1中描述OUT系统的出线。不一致的识别：当我们发现知识库中的不一致时，我们必须识别不一致的根源，以便使知识库保持一致。依赖关系导向回溯用于此目的。由于我们将不一致定义为指定谓词的不相容约束，所以上述标识的搜索空间变得不太大。在确定了不一致的来源后，为了保持一致性，我们附加了一个新的谓词。通过限定新引入的谓词，我们得到了该谓词的显式定义。换言之，我们获得了更精确的知识库。如果我们将这些方法应用到ISA层次结构中，新引入的谓词对应于一个新的概念。对谓词的归类给我们一个新概念的定义。

The purpose of this research is to design a concepts learning system using nonmonotonic logic. We use TMS's depebdency directed backtracking and McCarthy's circumscription especially. We describe the out-line of out system in the following.1. Identification of inconsistency: When we find inconsistency in a knowledge base, we must identify the souce of inconsistency in order to be the knowledge base consistent. The dependency directed backtracking is used for this purpose. Since we difine inconsistency as imcompatibility constraints for specified predicate, search space of the above described identification becomes not too large.2. After identification of the source of inconsistency, in order for consistency, we attach a new predicate. By circumscribing the newly introduced predicate, we obtain an explicit definition of the predicate. In other words, we obtain more precise knowledge base.3. If we apply these method to isa hierarchy, newly introduced predicate corresponds to a new concept. Corcumscription of the predicate gives us a definition of the new concept.