PURPOSE: To learn plural patterns at a high speed by attaining a constitution where the parts of networks having each information processing element belonging to an output layer as each start point have no common information processing element in the layers except an input layer.
CONSTITUTION: A neural network which performs the recognition of patterns has such a constitution where a network having an optional information processing element belonging to an output layer of the neural network as a start point and a network having another information processing element belonging to the output layer as a start point have not common information processing element in the layers except an input layer. For instance, both subnetworks which recognize the patterns 1 and 2 respectively share no layer except an input layer. In other words, no coupling coefficient is shared between both subnetworks. Thus no problem of interference is produced to the learning and therefore the overall learning efficiency is improved in the neural network.
KOBAYASHI YASUHIRO
Next Patent: DRIVING CIRCUIT FOR LIQUID CRYSTAL DISPLAY DEVICE