Abstract
A novel algebraic neural network training technique is developed and demonstrated on two well-known architectures. This approach suggests an innovative, unified framework for analyzing neural approximation properties and for training neural networks in a much simplified way. Various implementations show that this approach presents numerous practical advantages; it provides a trouble-free non-iterative systematic procedure to integrate neural networks in control architectures, and it affords deep insight into neural nonlinear control system design.
Original language | English (US) |
---|---|
Pages (from-to) | 1605-1610 |
Number of pages | 6 |
Journal | Proceedings of the American Control Conference |
Volume | 2 |
DOIs | |
State | Published - Jun 2001 |
All Science Journal Classification (ASJC) codes
- Electrical and Electronic Engineering