Neural networks with prescribed large time behaviour
Sergei Avgustovich Vakulenko, Peter V. Gordon · Journal of Physics A Mathematical and General · 1998
The generalized Hebb rule (with a non-symmetrical synaptic matrix) allows us to create simple neural networks with complicated large time behaviour. These networks can simulate, in a sense, any dynamics and, in particular, can generate any hyperbolic attractors and invariant sets. The explicit mathematical algorithm allows us, by adjusting the network parameters (the neuron number, coupling matrix and thresholds) to obtain a network with given large time dynamics.