The segmentation (or classification) problem for pairwise Markov models (PMMs) is considered. In a PMM, the observation process and the underlying state sequence form a two-dimensional Markov chain. We examine examples of rather different types of PMMs. These include, for example, a model for two related Markov chains and a model that allows an inhomogeneous Markov chain to be represented as a homogeneous one.

In the segmentation problem, only one of the marginal processes is observed, and the task is to estimate the unobserved state path given the observations. Standard state-path estimators are the Viterbi path (a state sequence with the maximum state-path probability given the observations) and the pointwise maximum a posteriori (PMAP) path (a state sequence that maximizes the conditional state probability pointwise given the observations). Since these estimators have certain limitations, hybrid path estimators that interpolate between the PMAP path and Viterbi path are studied.

Please note (and please join): After the seminar, we will walk to Studenterbaren for a refreshment.