Minimum Message Length Hidden Markov Modelling

Tim Edgoose & Lloyd Allison

home1 home2
 Bib
 Algorithms
 Bioinfo
 FP
 Logic
 MML
 Prog.Lang
and the
 Book

MML
 Structured
  mixtures
  HMM
  DTree
  DGraph
  Supervised
  Unsupervised
IEEE Data Compression Conference, pp.169-178, 1998
 
Abstract: This paper describes a Minimum Message Length (MML) approach to finding the most appropriate Hidden Markov Model (HMM) to describe a given sequence of observations. A MML estimate for the expected length of a two-part message stating a specific HMM and the observations given this model is preseneted along with an effective strategy for finding the best number of states for the model. The information estimate enables two models with different numbers of states to be fairly compared which is necessary if the search of this complex model space is to avoid the worst locally optimal solutions. The general purpose MML classifier `Snob' has been extended and the new program `tSnob' is tested on `syntehetic' data and a large `real world' dataset. The MML measure is found to be an improvement on the Bayesian Information Criterion (BIC) and the un-supervised search strategy effective.
 
Paper:
[paper.pdf],
[pdf@DCC]['04]
 
Also see:
[seminar]
Coding Ockham's Razor, L. Allison, Springer

A Practical Introduction to Denotational Semantics, L. Allison, CUP

Linux
 Ubuntu
free op. sys.
OpenOffice
free office suite
The GIMP
~ free photoshop
Firefox
web browser

© L. Allison   http://www.allisons.org/ll/   (or as otherwise indicated),
Faculty of Information Technology (Clayton), Monash University, Australia 3800 (6/'05 was School of Computer Science and Software Engineering, Fac. Info. Tech., Monash University,
was Department of Computer Science, Fac. Comp. & Info. Tech., '89 was Department of Computer Science, Fac. Sci., '68-'71 was Department of Information Science, Fac. Sci.)
Created with "vi (Linux + Solaris)",  charset=iso-8859-1,  fetched Friday, 29-Mar-2024 12:37:34 AEDT.