7. Stability and safety

One of the problems of artificial neural network technology is to verify stability and, to further refine the practitioner's formulation, to verify reliability and consistency of behavior. This difficulty arises not only in the case of neural networks, but also in that of many adaptive control circuits (e.g. learning control loops).

Stability and safety are essential in practice.

A typical application of artificial neural networks in control is to improve a known existing conventional process. It is rare to completely replace conventional processes, or to break new ground.

The purpose of installing an artificial neural network is to take advantage of the learning possibilities and ease of model building and control, which conventional processes can only achieve inadequately or at the cost of costly effort. This is...