MODEL GENERATION DEVICE, REGRESSION DEVICE, MODEL GENERATION METHOD, AND MODEL GENERATION PROGRAM

摘要

A model generation device according to one aspect of the present invention implements machine learning of a neural network module including computation for extracting an element satisfying a predetermined condition from a target set. During machine learning, the model generation device performs extraction computation as it is in a feed forward phase of the neural network module, and in a feed backward phase, substitutes the extraction computation with differentiable alternate computation and calculates an approximate gradient corresponding to differential calculation of the extraction computation by differential calculation of the substituted alternate computation.