Absolute stability criterion for discrete time neural networks
Bruno Cessac · Journal of Physics A Mathematical and General · 1994
We give an absolute stability criterion for additive neural networks with discrete time dynamics, i.e. we show that there exists a value for the gain parameter of the sigmoidal transfer function below which the system admits only one fixed point, attracting all trajectories. As an example, we compute this value in the case of random synaptic weights and a fully connected net, in the thermodynamic limit.