A high order neural network called the memory model has been established as a content addressable memory device in [J]. In contrast to Hopfield systems, the memory model can store any combination of any of the binary n-strings without spurious attractors and with one feedback loop per memory--solving the content addressable memory problem.
More generally, high order neural network models are capable of quickly reaching a great variety of finite choice decisions arising in optimization problems. We wish to build a dynamical system which has as its attractor trajectories feasible solutions. Also, we wish to design the regions of attraction so that optimal or near optimal solutions are inherently favored. Thus the attractors (regarded as answers or memories) are implicitly built into the system functions of the model and are discovered by repeated simulations.