Memory capacity of artificial neural networks with high order node connections

The authors have undertaken to understand and quantify the performance of artificial neural networks in the simpler and more restricted domain of associative memory. They have constructed models of artificial neural networks based on several different node-to-node interconnection schemes. The authors have simulated these models and quantified their performance as associative memories. The storage capacity, as determined by the retrieval accuracy, has shown large increases with the higher-order node-to-node connection models. The storage capacity of a four-node connection model (the output of each quartet of nodes is input to every other node) is 46 times larger than a conventional two-node (i.e. Hopfield) model with the same number of nodes.<>