HIDDEN-VARIABLE-MODEL ESTIMATION DEVICE AND METHOD

摘要

Provided is a hidden-variable-model estimation device that makes it possible to implement high-speed model selection even if the number of candidate models increases exponentially with the number of hidden states and the number of types of measurement probabilities. A variation-probability calculation unit (71) calculates a variation probability by maximizing a reference value defined as the lower bound of the Laplace approximation of the estimated value of a marginalized log-likelihood function with respect to a complete variable. A model estimation unit (72) produces an estimation of the optimal hidden-variable model by estimating the measurement-probability type of and parameters for each hidden state. A convergence determination unit (73) determines whether or not the reference value that the variation-probability calculation unit (71) used when calculating the variation probability converged.