HMM Network Instance

An HMM network instance is a lattice-based directed conditional probability network that abides by a hidden Markov network metamodel.

• AKA: Hidden Markov Model, Hidden Markov Graph, Viterbi Lattice.
• Context:
  • It can (typically) be composed of:
    • HMM States.
    • HMM State Transitions with HMM State Transition Probabilitys.
    • HMM Observations. (a Probabilistic Network Input Node?)
    • an HMM Start Probability.
    • an HMM Emission Probability.
  • It can be associated to Hidden Markov Metamodel.
  • It can be a Finite-State Sequence Tagging Model.
  • It can be represented with:
    • a Viterbi Lattice Structure.
    • a Weighted Directed Graph.
    • a Weighted Finite State Automata.
    • a Stochastic Finite State Machine.
  • It can be produced by a Hidden Markov Modeling System (that applies an HMM training algorithm to solve an HMM training task)
• Example(s):
  • 
  • 
• Counter-Example(s):
  • a Conditional Random Field Model Instance.
  • a Markov Logic Network.
  • a Recurrent Neural Network Model Instance.
• See: Undirected Probabilistic Network.

---

---

References

2005

• (Cohen & Hersh, 2005) ⇒ Aaron Michael Cohen, and William R. Hersh. (2005). “A Survey of Current Work in Biomedical Text Mining.” In: Briefings in Bioinformatics 2005 6(1). doi:10.1093/bib/6.1.57
  • Zhou et al. trained a hidden Markov model (HMM) on a set of features based on

2004

• http://www.cassandra.org/pomdp/pomdp-faq.shtml
  • Michael Littman's nifty explanatory grid:

Markov Models		Do we have control over the state transitons?
		NO	YES
Are the states completely observable?	YES	Markov Chain	MDP Markov Decision Process
	NO	HMM Hidden Markov Model	POMDP Partially Observable Markov Decision Process

1997

• (Shin, Han et al., 1995) ⇒ Joong-Ho Shin, Young-Soek Han, and Key-Sun Choi. (1995). “A HMM Part-of-Speech Tagger for Korean with wordphrasal Relations". In: Proceedings of Recent Advances in Natural Language Processing (RANLP 1995)
  • QUOTE: The trained hidden Markov network reflecting both morpheme and word-phrase relations contains 712 nodes and 28553 edges.

1993

• (Tanaka et al., 1993) ⇒ H. Tanaka & al. (1993). “Classification of Proteins via Successive State Splitting of Hidden Markov Network.” In: ProceedingsW26 in IJCAI93,

---