The Use of State Space Control Theory for Analysing Feedforward Neural Networks
Multi-layer neural networks have the ability to approximate any nonlinear function [l], and are therefore used for a variety of purposes. However, in many cases, the neural network is treated as a black box, since the internal mathematics of a neural network can be hard to analyse. As the size of a neural network increases, its mathematics becomes more complex and hence harder to analyse. This chapter examines the use of concepts from state space control theory, for the analysis of feedforward neural networks. The concepts used in this chapter are observability, controllability and stability. Some can be applied completely to feedforward neural networks and others have little or no meaning in the context of neural computing. Each concept will be examined and its used for analysing feedforward neural networks discussed.
KeywordsState Space Hide Layer Training Algorithm Feedforward Neural Network State Space Representation
Unable to display preview. Download preview PDF.
- K. Ogata, 1967, “State-space analysis of control systems.”, Prentice-Hall.Google Scholar
- Y.Z. Tsypkin, 1971, “Adaptation and Learning in Autonomous Systems.”. Academic Press Inc.Google Scholar
- M. Vidyasagar, 1993, “Nonlinear Systems Analysis.”, Prentice-Hall International Editions.Google Scholar
- J.C. Mason, 1984, “Basic Matrix Methods.”, Butterworth and Co.Google Scholar