LATENT VARIABLE OPTIMIZATION DEVICE, FILTER COEFFICIENT OPTIMIZATION DEVICE, LATENT VARIABLE OPTIMIZATION METHOD, FILTER COEFFICIENT OPTIMIZATION METHOD, AND PROGRAM

摘要

Provided is a technology which optimizes a latent variable by using a cost function obtained by taking into simultaneous consideration of a plurality of probabilistic assumptions. In this latent variable optimization device including an optimization unit which optimizes a latent variable ~w*, the optimization unit takes vj(1≤j≤J) as an auxiliary variable of the latent variable ~w* expressed as vj=Dj~w*+bj by using a matrix Dj and a vector bj, and cost terms of the auxiliary variable vj are expressed by using a probability distribution of the auxiliary variable vj, which is a log-concave distribution. The optimization unit optimizes the latent variable ~w* by solving a minimization problem of the cost function including the sum of cost terms of the auxiliary variable vj.