EMアルゴリズムの基礎 Fundamentals of the EM Algorithm
The EM (Expectation-Maximization) algorithm is a general-purpose stable procedure for maximum likelihood estimation in a wide variety of situations described as incomplete-data problem. Incomplete- data problems where the EM algorithm has been succesfully applied include not only evidently incomplete- data situations, for example, there are missing data, grouped observations, but also a whole variety of situations where the incompleteness of the data is not natural or evident.<BR>In this article, at first, I summarize maximum likelihood estimation and formulation of the EM algorithm. Subsequently, I briefly mention the properties of the EM algorithm, and two applications where the typical probablistic models are assumed. Lastly, I introduce some problems, which arise from applying the EM algorithm to the complex situations, and the examples of the solutions against them.
放射線医学物理 19(3), 148-161, 1999-09-30
Japan Society of Medical Physics