Unsupemised learning is the great promise of the future. In such training the network is provided with inputs but not with desired outputs. Unsupemised learning is used for the principal component networks. This paper describes a nerw method for training of the recirculation networks. Such method is ...