We introduce a new learning strategy employing a modular structure. Such a system consists of the so-called GREN-network approximating the unknown objective function and the actually trained BP-network.
