型付き論理プログラムの正例からの学習

概要

論文の詳細を見る
This paper describes a novel method to learn typed logic programs from positive examples. The attaching of type information to hypotheses is effective in avoiding over-generalization as well as enhancing readability and comprehensibility. IN many areas, such as NLP, type information plays a crucial role in achieving high performance. It is also typical that negative examples are not always available. RHB^+, our implementation of an inductive logic programming system that learns logic programs based on t-terms, makes use of type information to efficiently compute informativity form only positive examples and to judge a sopping condition. The new technique of the dynamic type restriction by positive examples lets covered positive examples decide the types appropriate for the current clause. RHB^+, written in the typed logic programming language LIFE, directly manipulates types as structured background knowledge when operations related to types are required. These features make RHB^+ efficient and effective in deciding types to be attached to variables in hypotheses. This leads to advantages over several previous learners, such as FOIL and PROGOL. Experimental results demonstrate RHB^+'s good performances for both artificial and real data.
社団法人人工知能学会の論文
2000-01-01