Convergence time in infinite-range neural networks with parallel dynamics at zero temperature
Ido Kanter · Physical Review A · 1989
In simulations on the Little-Hopfield model it is found that the convergence time to a stable state close to one of the embedded patterns scales like c(${m}_{0}$)log$_{10}N$, where N is the size of the network and c(${m}_{0}$) depends on the initial macroscopic overlap ${m}_{0}$ with the pattern. In a related model known as the pseudoinverse model the convergence time to the pattern is much smaller than ${\mathrm{log}}_{10}$N. The results are compared with other possible pattern recognition methods.