Quand: le 1 juin 2017 à 14h
Où: en H218, dans le bâtiment Étoile à Télécom SudParis

Title: On segmentation with hidden, pairwise and triplet Markov models

Prof. Juri Lember
Tartu University
Estonia

Abstract: In a latent variable stochastic model, the segmentation problem in general consists of estimating the unobserved sequence of hidden variables.The classical examples include image segmentation and speech recognition in signale processing; also GpC islands, DNA copy numbers and the secondary structure of proteins in molecular biology.

We introduce a general statistical learning-based framework for studyng the segmentation. It turns out that most of the segmentation methods in practice fall into that framework. Then we study the possibility of applying the methods for wide class of stochastic models — pairwise and triplet Markov chains.