The hybrid Hopfield-clique memory with perfect storage

A new model for associative memory, the hybrid Hopfield-clique network (HHcN) is proposed. It is based on Jagota´s Hopfield-clique network (HcN) and addresses its main problem, i.e., the existence of spurious memories. A new characterization of spurious memories based on graph theory is introduced. Given this characterization, it is possible to detect whether spurious memories are created. Once they are detected HHcN avoids them by creating new nodes in the network. The HHcN thus ensures perfect storage and recall for any arbitrary size collection of memories, and is space- and time-efficient on sparse ones (such as word-dictionaries). Experiments establish that the network sizes required to achieve perfect storage remain reasonably small