Entropy and Decision

The paper discusses the concept of entropy in hypothesis selection, and the conflict with maximum-likelihood methods, and in Markovian processes, wherein there is some conflict with implied steady-state behaviour.

Abstract

AbstractThis paper deals with the irrelevance of entropy to decision. The entropy measure fails to discriminate between states and their significance, as is indicated by a study of a reconnaissance example for which the entropy measure has been suggested. If the reduction of uncertainty is an explicit objective, then it is important to examine the properties of uncertainty measures, and it is queried as to whether some of those uniquely characterizing entropy are relevant. Other measures have some of the properties. The paper then discusses the concept of entropy in hypothesis selection, and the conflict with maximum-likelihood methods, and in Markovian processes, wherein there is some conflict with implied steady-state behaviour.