MODEL ESTIMATION DEVICE, MODEL ESTIMATION METHOD, AND MODEL ESTIMATION PROGRAM

摘要

A model estimation device (100) equipped with: a hidden variable variation probability calculation processing unit (104) that obtains parameters in a hidden variable model, and uses these parameters to calculate a constrained hidden variable variation probability, which is the posterior probability of hidden variables near a preassigned distribution; a model parameter optimization processing unit (105) that uses the constrained hidden variable variation probability to optimize the parameters in the hidden variable model; and an optimality determination processing unit (106) that determines whether a marginalized logarithm likelihood function which uses the optimized parameters is converging. When it is determined that the marginalized logarithm likelihood function is converging, the constrained hidden variable variation probability and the parameters used in that marginalized logarithm likelihood function are output.