構造化領域での生成的概念形成と記述の扱い

概要

論文の詳細を見る
A new learning model called SCNC, working in structured domains consisting of objects with many components, is presented to induce concepts under the situation where no attributes are prespecified. Model SCNC receives observed data about the components, and outputs a concept hierarchy and attributes for describing concepts. The needed attributes are generated in the course of concept formation. One important problem in treating structured objects is what kind of description is to be employed. If any description is expressed in a combination of various literals for each component as seen in the previous models, the results should be understood via relations among the components. We here propose to employ descriptions with a single attribute for some specified components, and discuss the benefits of employing them in getting a compact and easy understandable concept hierarchy. The applicability of SCNC to the real world data is also examined from viewpoints of the costs and the resultant qualities.
社団法人人工知能学会の論文
1994-11-01