Several robust algorithms for parametric optimization of hidden Markov models are presented. These combine aspects of Fabian's 'sign' algorithm, two-time scale stochastic approximatton and certain techniques for estimating the gradient (or related quantit
展开▼