An asynchronous weightless neural network for discrete relaxation problems

This paper proposes a weightless neural network for solving discrete relaxation problems. The network implements a parallel version of a new efficient sequential algorithm for the consistent labeling problem. The proposed neural network consists of a feedback interconnection of three layers of very simple neurons that operate in an asynchronous fashion. Because of the simple structure of the neurons, the neural network can be implemented using highly regular AND/OR VLSI arrays. The asynchronous operation eliminates the need for using clocks or placing registers between the different layers and along the feedback paths of the network, thus resulting in very fast convergence time.