The present invention achieves a neural network capable of detecting abnormality in the neural network, while reducing redundancy of scale. This neural network optimization system is characterized by including: a definition data analysis unit that analyzes trained neural network definition data; an inter-node dependency analysis unit that generates, on the basis of a result of analyzing the trained neural network definition data, dependency information indicating dependency between nodes in a trained neural network defined by the trained neural network definition data; and a sensitive node extraction unit that extracts, on the basis of the dependency information, a sensitive node in the trained neural network.
展开▼