Neural network, latent parameter learning device, latent parameter generation device, variable conversion device, method and program for these

摘要

PROBLEM TO BE SOLVED: To provide a random variable conversion technique capable of guaranteeing that a function describing a relationship between input/output data in a neural network is injective. In a neural network including one or more layers including one affine transformation unit and one activation function calculation unit, all the singular values of the weight matrix used in the affine transformation performed by the affine transformation unit are positive. In the activation function used in the activation function calculation unit, both the upper bound coefficient and the lower bound coefficient of the Lipschitz smoothness are positive. [Selection diagram] Fig. 6