An output coding approach for knowledge increasable artificial neural network

How to inherit the learned knowledge of existing neural networks without destroying their structure and functionality is a difficult problem. In this paper, we propose an output coding approach for building such a system, which fully utilizes the information gained from the component neural units. By coding the neural outputs, a neural network becomes a self-contained system. For a given pattern, such a neural network can correctly recognize or reject it or point out it is similar to patterns it has learned. Such information is useful for further decision. Experiments demonstrate it is a good approach for building a KIANN system. This is meaningful for utilizing the learned knowledge of existing neural networks and for large scale parallel processing.