On the existence of unstable bumps in neural networks (Q1949280)

The nonlinear integro-differential equation of Hammerstein is interpreted as modeling the behavior of a single layer of neurons in a neural network, the solution representing the local activity of a neuron and the integral kernel the coupling between neurons. Under some special assumptions on the integral kernel as well as on the nonlinear function of the neurons' firing rate, the existence of unstable bump solutions to the stated model is proved. Actually, common kernels as `Mexican hat' function or Gauss function do satisfy the conditions. Under additional assumptions, the referred bump solution is a Lyapunov-unstable equilibrium of the model, belonging to the space of continuous functions vanishing at infinity.