A methodology for analysis of neural network generalization in control systems

In this article, a methodology for analysis of neural network generalization in control systems is presented. Rigorous definitions to quantify the generalization ability of a neural network in the context of system control are given. Utilizing these definitions, it is proved that a successfully trained neural network always generalize “well” to some extent. It is then shown that (i) specific conditions under which a neural network is guaranteed to generalize “well”, and (ii) the performance of the control system operating under those conditions, can be analytically determined using techniques from system sensitivity theory. The results of this work provide new tools for performance analysis of neuro-control systems, and represents a first step towards a rigorous framework for performance-oriented analysis and synthesis of neural networks for control