Forward smoothing and online expectation-maximisation in Gaussian linear state-space models
Sinan Yıldırım, Ali Taylan Cemgil · 2011
In this work, we studied forward-only smoothing recursion in Gaussian linear state-space (GLSS) models. We exploited a stochastic approximation of this recursion to develop an online version of the expectation-maximisation (EM) algorithm for GLSS models. We compared the performance of online EM with the conventional EM and demonstrated the advantages of its use in case of long data sequences.