Generalized HMMs (GHMMs)
Stephen N. Winters-Hilt · 2021
The generalized clique Hidden Markov Models (HMM) begins by enlarging the primitive hidden states associated with individual base labeling to substrings of primitive hidden states or footprint states. This chapter provides a description of the Baum–Welch algorithm in the adaptive Hidden Semi-Markov Models (HSMM) formalism followed by a description of the Viterbi algorithm in the adaptive HSMM formalism. The sliding-window clique overlap is much more significant than with the standard HMM, giving rise to many more table look-ups on eij -transition tables. Use of the meta-HMM formalism resolves complications due to heavy-tail duration distributions and weak contrast. The standard HHMM-with-Duration (HMMD) replaces the equation with a pi(d) that models the real duration distribution of state i . Table chunking methods for the dynamic programming algorithms have been developed that involve only a single-pass computation analogous to the Viterbi algorithm. The Viterbi algorithm efficiently calculates the most probable state path.