Statistics Seminars: An efficient algorithm for performing inferences in hidden Markov models with imprecise probabilities
26 May 2009 14:15 in CM107
We discuss hidden Markov models where the state transition probabilities, as well as the probabilities governing the observational process, may be imprecise, or interval-valued. These are special cases of credal networks, where we replace the usual notion of strong independence with the weaker epistemic irrelevance. We have derived an exact message-passing algorithm that computes updated beliefs for a variable in the network. The algorithm, which is essentially linear in the number of nodes, is formulated entirely in terms of so-called coherent lower previsions. We supply examples of the algorithm's operation, and report an application to on-line character recognition that illustrates the advantages of the model for prediction.
Contact firstname.lastname@example.org for more information