|
In
many recognition problems we are required to discover the underlying
structure in a sequence of observed symbols. Few example problem
areas are: speech recognition, optical character recognition,
part-of-speech disambiguation,gesture recognition from video sequences
and finding structural
motifs in DNA seqences.
A
mathematical formulation that has been successfully applied to
attack such problems is hidden Markov model (HMM). My talk
will be a tutorial on HMMs and is based on Rabiner's paper: "A
Tutorial on Hidden Markov Models and Selected Applications in
Speech Recognition, Prov. of IEEE, vol. 77, no. 2, pp. 257-286,
1989.
Topics
that I will be cover in my talk are:
-- The hidden Markov model formulation
-- Forward algorithm for estimating the total probability of a
model, given a sequence of observed symbols;
-- Viterbi algorithm for estimating the most likely state sequence,
given a sequence of observed symbols and a model;
-- Baum-Welch algorithm for estimating the model parameters, given
a sequence of observed symbols.
I
will cover the first two topics in the first talk, and the last
two in the second talk. The date and time of the second talk will
be announced later.
I
have a software implementation of all the algorithms. If you are
interested in getting a copy, please send me email at kanungo@cfar.umd.edu
|
