An on-line algorithm for generation of stable mappings in a generalized modular network SOM

Recently, we proposed a modular network SOM (mnSOM) in which each nodal unit of the conventional SOM is replaced by a neural network module. In the mnSOM, the learning algorithm is based on the batch learning SOM for generation of stability mapping, since the on-line algorithm of the SOM depends on initial condition and the scheduling of the neighborhood function. However, when the mnSOM is applied to an autonomous robot or a control system in which the unlabeled episode data are observed from sensors one after another, then the mnSOM should be used with on-line mode. In this paper, we propose an on-line mnSOM algorithm based on Bayesian theory in order to generate a stable map. Moreover, the availability of our proposed algorithm is shown with a result of a simulation.