Neural net based pattern recognition on the Graph Search Machine.

Thus far, neural networks have been implemented chiefly as mainframe simulations or in analog array processors (e.g., resistive nets.) Mainframe simulations have the disadvantage of being relatively slow while analog array processors are difficult to change. The Graph Search Machine (GSM), a programmable pattern matching device, is presented as an alternative for implementing neural network algorithms. As an example, the Hopfield Net Algorithm for character recognition has been implemented on a single GSM. The resulting performance is a recognition rate of about 29 characters per second, where a character is 8 by 8 pixels.