Training of a class of recurrent neural networks
Khaled M. Shaaban · 2002
This paper presents design and analysis of a modified version of the Hopfield network which is called a Cascade Recurrent Network (CRN). This network has a single or multilayer feedforward (FF) structure with synchronous input-output feedback. System dynamics are determined by the characteristics of this FF structure. First, a formal definition of the mapping generalization is provided. Using this definition, and the classical definition of stability, the mapping-stability relation is developed in the form of a correspondence between CRN stability properties and FF mapping characteristics. On the basis of this stability-mapping relation, a new synthesis technique is developed.