Re: [UAI] learning HMM structure

From: Stephen M. Omohundro (om3@worldnet.att.net)
Date: Wed Jun 13 2001 - 15:18:46 PDT

  • Next message: hpguo@cis.ksu.edu: "[UAI] KDD-2001 Call for Participation"

    > Hi,
    >
    > Can anybody point me to some references for learning the structure of
    > Markov chains and Hidden Markov Models from data?
    >
    > Also of interest would be some standard references discussing metrics or
    > scoring functions for HMMs.
    >
    > Thanks,
    > Denver.
    > --
    > http://www.sis.pitt.edu/~ddash

    The following references describe a technique for learning the
    structure of HMM's that works much better than simply applying EM to a
    totally connected structure and looking for non-zero probability
    transitions. The idea is to start with the maximum likelihood HMM for
    a given set of data. This has a node for each symbol of each training
    string and essentially looks like the collection of training strings
    themselves running in parallel from the start state to the final
    state. The algorithm then successively merges states together, always
    greedily picking the pair to merge which maximizes the increase in the
    Bayesian posterior probability. The merging stops when no merge can
    increase the posterior further. In practice, we found that it robustly
    finds the structure of many HMM's from very small amounts of data. A
    similar approach can also be applied to learning the structure of more
    complex stochastic grammars (such as stochastic context-free grammars).

    Andreas Stolcke and Stephen M. Omohundro, "Hidden Markov Model
    Induction by Bayesian Model Merging", in Advances in Neural
    Information Processing Systems 5, ed. Steve J. Hanson and Jack D.
    Cowan, J. D. and C. Lee Giles, Morgan Kaufmann Publishers, Inc., San
    Mateo, California, 1993, pp. 11-18.

    Andreas Stolcke and Stephen M. Omohundro, "Inducing Probabilistic
    Grammars by Bayesian Model Merging", Proceedings of the International
    Colloquium on Grammatical Inference, Alicante, Spain, Lecture Notes in
    Artificial Intelligence 862, Springer-Verlag, Berlin, September 1994,
    p. 106-118.

    Andreas Stolcke and Stephen M. Omohundro, "Best-first Model Merging
    for Hidden Markov Model Induction", ICSI Technical Report TR-94-003,
    January 1994.

    -- Steve

    Stephen M. Omohundro om3@att.net
    252 Hawthorne Avenue http://home.att.net/~om3/
    Palo Alto CA, 94301-1110 650-328-7605



    This archive was generated by hypermail 2b29 : Wed Jun 13 2001 - 15:20:10 PDT