Classification of time series with hidden Markov models: Unsupervised learning and self-organization
V. Breuer, Günter Radons · Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics · 1996
Unsupervised learning and feature recognition with hidden Markov models (HMM's) is investigated. The well-known Baum-Welch algorithm is utilized to tune the free parameters of the HMM. The local state probability distribution of the model controls the recognition. For a simple problem we show analytically the self-organization of a feature detector. In a numerical simulation we present a detector for two-dimensional textures that perceives, recognizes, and restores disturbed textures.