確率的推論のモデル化と推論結果の信頼性を保証する推論方式について

概要

論文の詳細を見る
In the field of Artificial Intelligence (AI), the framework of belief networks (BN) was proposed by Pearl in the 1980s. BN has become popular within the probability and uncertainty community of AI. The knowledge representation of BN facilitates intuitive understanding of knowledge. On the other hand, we proposed a new reasoning method, i.e. deduction algorithm, which is more general than the previous methods in our last paper. In this paper, we shall divide the reasoning into two parts, i.e. deduction and induction. The induction is a calculation of the maximum likelihood estimator of each cell in a contingency table. Meanwhile, the deduction is a calculation of conditional probabilities using the maximum likelihood estimator when several marginal sums of the conditional probabilities are given. We shall show that our deduction algorithm is useful for the framework in this paper. Moreover, we shall propose a new reasoning method that guarantees a given reliability based on statistics by applying the interval estimation.
湘南工科大学の論文
2003-03-18