A Simulation Based Algorithm for Optimal Quantization of Hidden
Vladislav B. Tadi, Arnaud Doucet · 2003
A general problem of multi-level scalar quantization of hidden Markov models is considered. An algorithm for the optimal selection of the quan- tization levels is proposed and its asymptotic behav- ior is analyzed theoretically and through simulations. The proposed algorithm is based on stochastic ap- proximation and Monte Carlo gradient estimation.