Hidden Markov Model
An HMM describes observation sequences with a two-layer model: hidden states transition in a Markov chain, and each state emits observations according to emission probabilities.
Purpose
To model the latent structure of sequential data, used in speech recognition and biological sequences.
Mechanism
Parameters λ=(A,B,π): A the transition matrix, B the emission probabilities, π the initial distribution. Three problems: evaluation (forward), decoding (Viterbi), and learning (Baum-Welch).
Use cases
- Speech recognition
- Part-of-speech tagging
- Gene prediction