Hidden Semi-Markov Models

Chapter 4

Implementation of HSMM Algorithms

Abstract

In the implementation of the general HSMM, one may face various practical problems. The first and the biggest problem is that the joint probabilities associated with observation sequence often decay exponentially as the sequence length increases, which results in that the implementation of the forward–backward algorithms by programming in a real computer would suffer a severe floating-point underflow problem. To overcome this problem, one usually uses a heuristic scaling method to scale the forward and backward variables, or transforms the forward–backward algorithm into the logarithmic form. This chapter introduces these two methods, derives a posterior formulation of the forward and backward ...