e-mail: ad2 at cec dot wustl dot edu
|
|
Alan Kaplan
e-mail: ad2 at cec dot wustl dot edu |
|
Inferring Latent Variables: Hidden Markov Models and Expectation Maximization Algorithms Often data exhibits observable properties that are dependent on one or more critical latent factors. The concept of state can aid in exploiting or discovering these factors. Such data can be modeled with finite mixture models, which assume certain probability density functions conditioned on unknown states; or more generally, hidden Markov models, which assume that state transitions form a Markov chain. The expectation maximization (EM) algorithm is used for estimating the parameters of these models. In this session, the preceding concepts will be introduced. |
|