Convergence time and finite size effects in neural networks
G. A. Kohring · Journal of Physics A Mathematical and General · 1990
The time needed for convergence to a stable state starting from an initial state having macroscopic overlap with a stored state is studied numerically for networks up to N=10 5 neurons. Although such systems are much larger than those traditionally used in studying neural networks, they are still too small to extract reliable information about the asymptotic behaviour of the convergence time.