A method for generating digital signals in response to novel inputs

Abstract

A neural network architecture which has the capability of generating unique digital output signals in response to novel input signals is introduced. In this capacity, the network functions as an associative memory where the mapping between input and output signals is automatic and shaped by the previous training of the network. The generated output is shown to be a combination of digital signals already stored in the network. While many neural networks possess similar capabilities, the mapping between input and output cannot usually be controlled without destroying information already stored in the network. However, in this network, the mapping can be directly controlled without altering information previously stored. Use of this network for generating unique output signals in response to new inputs is demonstrated