PURPOSE: To perform effective conflict learning without using inner product operation for a distance reference by extracting pieces of difference information on an inputted vector and a weight vector group and comparing the both, determining a weight vector to be updated according to the result, and updating the vector.
CONSTITUTION: The information processor which performs the conflict learning consists of an element A which outputs the input vector, an element B which outputs the weight vector group, and an element C which corresponds to a detector that detects and outputs the quantity obtained by comparing the both. The element A outputs the n× n-dimensional input vector and the respective components have (x) gradations. The element B outputs M×M weight vectors. The respective weight vectors are n×n-dimensional corresponding to the input vectors and respective components have (x) gradations. The result of the comparison between the input vectors and respective weight vectors is outputted by the element C. Then M×M scalar quantities are outputted corresponding to the number of the respective weight vectors and the respective scalar quantities have (y) gradations.
TERAJIMA MIKIHIKO
HASHIMOTO TAKESHI
SHIRATANI FUMIYUKI
Next Patent: PASSWORD COLLATING METHOD OF COMPUTER