[Auto-adaptative network separator of sources].
This article defines a new convergence criterion for the modelization of networks with modifiable synapses, as presented by Jutten, Herault, Ans ([1], [2]). The network adaptation law, empirically determined in [1], [2], is reconsidered: a theoretical law concerning the evolution of system coefficients and outputs is demonstrated and validated on simulations. The system always converges towards a unique attractor, as predicted by the theoretical convergence criterion. This system evolution towards a particular stable state enhances the interesting physical aspect of this modelization.
Humans↗