Multi-topographic neural network communication and generalization for multi-viewpoint analysis

This paper presents a new generic multitopographic neural network model whose main area of application is clustering and knowledge extraction tasks on documentary data. The most powerful features of this model are its generalization mechanism and its mechanism of communication between topographies. This paper shows how these mechanisms can be exploited within the framework of the SOM and NG models. An evaluation of the generalization mechanism based on original quality and propagation coherency measures is also proposed. A secondary result of this evaluation is to proof that the generalization mechanism could significantly reduce the well-known border effect of the SOM map.