M. Malyutov

Retrospective Training Slow HMM-SCOT Emissions Model

The Baum–Welsh recurrent ML estimation of HMM parameters has been successfully applied to speech recognition. Its application to Genome modeling is questionable since assigning independence and equal probabilities to emissions from the same part of genome is a rough approximation. We develop a hybrid slow HMM switching model with SCOT emissions which might be a more realistic model for Genome, analysis of combined authorship of literary works, seismological data or financial time series with piecewise volatility. Our combined online and offline segmentation stage estimates time regions with constant HMM states using homogeneity test for SCOT emissions strings. This is made recurrently in parallel on a cluster of computers.

KEYWORDS: SCOT models, Baum–Welsh algorithm, HMM, regime switching model