入出力データの逐次生成による能動学習統合型最適化法

抄録

We propose the optimization integrating an active learning of input-output relations with successively produced learning data. Here, a relaxed problem with several constraints selected from among an infinite number of inequality constraints to assure uniform accuracy to input-output relations is supposed firstly, and then inequality constraints corresponding to newly produced data are iteratively added to the relaxed problem. In this process, a global optimal solution with higher accuracy can be obtained under the assurance of approximation accuracy to input-output relations.