The EM Algorithm and Extensions

Rating:
Summary: Excellent text
Review: Excellent text on the EM algorithm. Covers theory as well as a number of applications. Clearly written. Historical accounts and examples make reading delightful. I would have found it sweeter if it covered applications in time series. It was only inevitable that everyone's favorite application couldn't be included because of their sheer multitude.

I guess this is also the only text available on the subject, as of now!

Rating:
Summary: great introduction to the EM algorithm
Review: Geoff McLachlan is well known for his books on discrimination and mixture models. He is an excellent writer who is very thorough in his description of the literature. The EM algorithm is a great invention which dates back to the seminal paper in JRSS by Dempster, Laird and Rubin. It was originally devised to handle likelihood estimation in the face of missing data. Standard applications also include truncated and censored data. Clever application of the missingness in the data structure have allowed it to be applicable to mixture model estimation and other problems where the missing information is not so obvious. It is used for Bayesian estimation, image processing, in random effects models, latent variable structures and log linear models. The technique and its applications are covered in parts of various books on specialized statistical topics or on statistical computing. However, this is the first and only text that is dedicated to the algorithm itself and its wide variety of applications. It is a perfect reference source for the EM algorithm and its various modifications. It provides a thorough and unified treatment for the subject.

Rating:
Summary: Learn about EM? Read the relevant papers but not this book
Review: I tried to read the whole introductive chapter a couple of times but I couldn't understand what is EM about, the used terminology and the basic definitions. The authors say that the book is for theoriticians and practicioners, but I do think it is not appropriate for both categories, unless the reader has been involved in writing papers on this topic. I have enough background knowledge in probability theory and in mathematics but it seems that I have to read all the relevant literature before going a step ahead. In my opinion this book is wide useless for people who do not know EM algorithm.