Returning to FIG. 1, the learning termination determining section 15 determines the termination of the learning.

For example, when the difference between the weight W′_N×V that is the parameter before the update of the weight W′_N×V and the weight W′_N×V that is the parameter after the update of the weight W′_N×V is smaller than a threshold, the learning termination determining section 15 terminates the learning. The weight W′_N×V before the update is a value upon the termination of the activation of the parameter adjusting section 13 or upon the start of the activation of the adjusted parameter distributed representation learning section 14. The weight W′_N×V after the update is a value upon the termination of the activation of the parameter distributed representation learning section 14. A requirement for the termination of the learning is expressed by the following Inequality (1). W′_N×Vnew is W′_N×V after the update of W′_N×V. W′_N×Vold is W′_N×V before the update of W′_N×V. ε is a threshold. It is sufficient if the difference between the weight W′_N×V before the update and the weight W′_N×V after the update is determined to be sufficiently small based on the threshold.
W′_N×Vnew?W′_N×Vold<ε??(1)

When the difference between the weight W′_N×V before the update and the weight W′_N×V after the update is equal to or larger than the threshold, the learning termination determining section 15 causes the parameter adjusting section 13 and the adjusted parameter distributed representation learning section 14 to repeatedly operate.