A simulation based algorithm for optimal quantization of hidden Markov models
Vladislav B. Tadić, Arnaud Doucet · 2003
A general problem of multilevel scalar quantization of hidden Markov models is considered. An algorithm for the optimal selection of the quantization levels is proposed and its asymptotic behavior is analyzed theoretically and through simulations. The proposed algorithm is based on stochastic approximation and Monte Carlo gradient estimation.