High-order Hopfield and Tank optimization networks

We employ high-order weights to extend the class of optimization problems that can be solved with neural networks. Hopfield and Tank networks are used; the associated energy function is a polynomial with order equal to the highest order weights in the network. As an example, we consider the problem of partitioning a graph into triangles. Simulation results indicate that multiple runs on a problem can be considered independent trials; high performance can thereby be achiebed feasibly.