Rationale presentation device, evidence presentation method, and evidence presentation program

摘要

PROBLEM TO BE SOLVED: To create a saliency map showing a basis for judgment even in a neural network handling various input data at low cost. A rationale presentation device 100 combines an input data "x" and a saliency map 122 "m" to create a composite data "y", a composite unit 103, and an input data "x" and a composite data "x". The inference unit 104 that creates the inference data "z" as a result of inferring from each of "y" using the first NN112, the inference data "z" of the input data "x", and the inference data "z" of the composite data "y". The map generation unit 102 has an evaluation unit 105 that calculates the error with the loss function “E1 (・)” and uses the calculation result as the loss score, and the map generation unit 102 refers to the second NN 121 used in the previous inference. The parameter of the second NN 121 is updated by reflecting the loss score. [Selection diagram] Fig. 1