GENERALIZATION AND CHAOS IN A LAYERED NEURAL-NETWORK

The generalization performance of a multi-state and a graded response layered attractor neural network trained with examples of low activity is established exactly for monotonic and non-monotonic input/output f unctions. Complex behaviour is found which goes from fixed-point attra ctors to chaos through a cascade of bifurcations, depending on an appr opriate threshold or cut-off parameter. The effect of the irregular be haviour on the generalization curves is explicitly demonstrated and ph ase diagrams for the recognition ratio of concepts cr in terms of the threshold/cut-off exhibit ordered (generalization), disordered (parama gnetic or self-sustained activity) and chaotic phases.