Final Progress Report on Robust and/or Adaptive Filtering by Neural Networks
Final rept. 1 Jul 1999-30 Jun 2002
MARYLAND TECHNOLOGY CORP ELLICOTT CITY
Pagination or Media Count:
The purpose of this project is to develop a general and systematic approach to robust andor adaptive filtering in the presence of uncertain environmental parameters. Mathematical justification, intuitive understanding and numerical confirmation of risk-averting neural networks for general robust processing with various degrees of robustness have been achieved. Those of neural networks with long- and short-term memories for general adaptive processing have also been accomplished. A novel method of training neural networks that is effective in avoiding poor local minima has been discovered. This discovery is a major breakthrough in the development of neural computing. Robust neural filters have been mathematically justified and numerically tested. General adaptive filtering and general robust adaptive filtering turned out to be much more difficult than expected. Nevertheless, schemes for them by neural computing, which are mathematically natural and convincing, have finally been conceived.
- Numerical Mathematics