文档介绍：Olivier Capp´e,Eric Moulines and Tobias Ryd´en
Inference in Hidden
Markov Models
May 22, 2007
Springer
Berlin Heidelberg NewYork
Hong Kong London
Milan Paris Tokyo
Preface
Hidden Markov models—most often abbreviated to the acronym “HMMs”—
are one of the most essful statistical modelling ideas that have came up in
the last forty years: the use of hidden (or unobservable) states makes the model
generic enough to handle a variety plex real-world time series, while the
relatively simple prior dependence structure (the “Markov” bit) still allows
for the use of putational procedures. Our goal with this book is to
present a plete picture of statistical inference for HMMs, from
the simplest finite-valued models, which were already studied in the 1960’s,
to recent topics putational aspects of models with continuous state
space, asymptotics of maximum likelihood, putation and model
selection, and all this illustrated with relevant running examples. We want
to stress at this point that by using the term hidden Markov model we do
not limit ourselves to models with finite state space (for the hidden Markov
chain), but also include models with continuous state space; such models are
often referred to as state-space models in the literature.
We build on the considerable developments that have taken place dur-
ing the past ten years, both at the foundational level (asymptotics of maxi-
mum likelihood estimates, order estimation, etc.) and at putational
level (variable dimension simulation, simulation-based optimization, etc.), to
present an up-to-date picture of the field that is self-contained from a theoret-
ical point of view and self-sufficient from a methodological point of view. We
therefore expect that the book will appeal to academic researchers in the field
of HMMs, in particular PhD students working on related topics, by summing
up the results obtained so far and presenting some new ideas. We hope that it
will similarly interest practitioners a