CPS196 - Fall 1999

Hidden Markov Models

Reading: Manning and Schütze, Part of Chapter 10

Background: Like MDPs, Hidden Markov Models (HMMs) are a finite-state model with probabilistic transitions. However, whereas MDPs have a notion of actions and rewards, HMMs have a notion of "observations".

We'll talk about how HMMs provide a well reasoned way of dealing with state uncertainty, and how they are applied in some aspects of language processing. We'll work through a "leet speak" repair example.

Questions:

Notes


Modified: Wed Oct 27 21:01:46 EDT 1999 by Michael Littman, mlittman@cs.duke.edu