逐次推定過程の情報理論的基礎づけ

概要

論文の詳細を見る
This paper attempts to establish a theoretic basis for sequential estimation problems from the viewpoint of information theory. The problem which is dealt with consists in that one has to make a decision which point of a set of categories an observed sequence of random variables belongs to. It is supposed that the distributions of the random variables depend on a point of the finite set of categories. Much attentions are paid to constructing the quantities of equivocation with respect to decisions,which satisfy some axiomatically required conditions of goodness measure. A necessary and sufficient condition for an equivocation to be minimized by the Bayes decision rule is obtained when the number of categories is larger than two. In the simplest case when only two categories are possible the Bayes decision rule minimizes an infinite class of equivocation quantities.
一般社団法人情報処理学会の論文
1969-03-15