PROBLEM TO BE SOLVED: To provide a learning method of a weight parameter in a neural network capable of performing learning by using various and large amounts of data while protecting privacy.SOLUTION: In a neural network system comprising a plurality of pieces of terminal equipment 1 and a management device 2, each of the plurality of pieces of terminal equipment comprises: a storage part 12 for storing image data; an image identification apparatus 13 for learning a weight parameter for performing image identification; and a first control part 14 for transmitting the learnt weight parameter to a management device, and for transmitting data for learning generated on the basis of the image data to the other terminal equipment or the management device, and for receiving the data for learning generated by the other terminal equipment. The image identification apparatus learns the weight parameter by using not only the image data of its own storage part but also the data for learning generated by the other terminal equipment. The management device comprises a second control part 22 for receiving the learnt weight parameter from the plurality of pieces of terminal equipment, and for selecting the optimal weight parameter.SELECTED DRAWING: Figure 1
展开▼