メタ知識で制御した学習システム

概要

論文の詳細を見る
There are many Inductive Logic Programming(ILP)systems proposed in recent years. Some of these systems try to employ the additional knowledge except the given background knowledge itself. Most of the existing systems only employ the contents of background knowledge. FOCL deeply depends on the typing constraints and the mode of background knowledge to diminish the search space ; GRENDEL translates the background knowledge by a bias representation language ADG to an explicit description, which is a language representation bias. It is necessary that a system employs the meta-knowledge as well as background knowledge itself. The meta-knowledge, a kind of knowledge in knowledge, is some control knowledge and the prerequisites of background knowledge. Our system ACTOR efficiently learns the first order logic clauses by employing the suitable constraints. The constraints are the ante-constraint and post-constraint. The ante-constraint comes from the application of the meta-knowledge and the post-constraint is from the utilization of MDL principle.
社団法人電子情報通信学会の論文
1996-03-11