An assembly neural network model with a new recognition algorithm is described. The network is artificially partitioned into subnetworks according to the number of classes that the network has to recognize. The features extracted from input data are represented in neural column structures of the subnetworks. Hebb's assemblies are formed in the column structures of the subnetworks by means of modification of connections' weights. A generalization process takes place within each subnetwork of the assembly network separately which results in formation of an adequate description of every recognized class inside its own subnetwork. A computer simulation of the network is performed. The generalization phenomenon is explored in special experiments on the character recognition task.
展开▼
