A study on merging mechanisms of simple hopfield network models for building associative memory

We are in the era of artificial intelligence which is fulfilling the vision of ubiquitous intelligent machines and systems. Just like in us humans, the most critical elements for machine intelligence are those related to associative memory and recall. It is urgent to have a structure or model that can emulate humanlike associative memory and recall functionality. This research is aimed at building a model based on a hybrid of neural networks like Hopfield network, recursive neural network (Recursive NN), and recurrent neural network (Recurrent NN). In our study, a Hopfield network is used as a core element for learning associative relations among concepts or objects and the Hopfield matrix as a basic unit memory for association knowledge. Hopfield networks in a sequence are recursively merged by applying a Recursive NN and associative relations between concepts/objects nodes in two Hopfield networks are learned and their sequences of merging are kept in a Recurrent NN. When there is a stimulus, the model can retrieve its associative concepts or objects as its recall. This paper shows the feasibility of the proposed model with some case study and a proof of concept prototype.