Nonstationary Newral Nets: a Bridge between Classical Newral Nets and Coupled Map Lattices?
We describe an attempt to construct a Newral Network which has the form of Globally Coupled Map and the stored patterns correspond to complex periodic or chaotic behaviour instead of stationary state (fixed point). The output of the network is formed by means of time averaging. It is shown the possibility to create both feed forward and attractor networks with nonstationary neurons.
Mathematical modelling in actual problems of science and technics