The present invention is that about bottommost represents multiple weight to tensor tensor sum low level (inferior grade) in a process of upper decomposition and used by decomposing tensor and obtain the data by the corresponding acquisition of the specific relation intensity of cynapse-of learning performance of neural network method, it is used to obtain an approximate weight and tensor, and it includes the weight and structure that process is the artificial neural network ANN re-executed that cynapse-specialized training of weight, which corresponds to approximate tensor adjustment described above,.
展开▼
