The learning system includes a learning device and a teacher data creation device for the learning device. The learning device learns the neural network by error back propagation. The teacher data creation device, with respect to the input data, for each label, excludes from the positive evaluation indicating that the content of the input data matches the label, the sub-assessment indicating that the content of the input data does not match the label, and excluding the label to be learned Any one of the neglected evaluations shown is acquired to create teacher data. When learning a neural network for learning, the learning apparatus adjusts the weighting factor of the middle layer so that the recognition score of the positive or negative evaluation label and the positive or negative evaluation score are close together, and furthermore, Make sure that the recognition score does not affect the adjustment of the weighting factor in the middle layer.
展开▼