A parallel hybrid learning approach to artificial neural nets

The requirements of well chosen applications are of great importance for developing new parallel computer architectures. The algorithms presented are implemented on the EDS (European Declarative System) parallel computer. By using a hybrid approach of genetic and gradient descend algorithms in an appropriate manner the advantages of both methods are combined. It is shown how artificial neural networks can be modelled to make the application of the hybrid learning paradigm possible. This hybrid learning approach was implemented in a simulation environment (NNSIM) and compared with standard learning algorithms on a phoneme recognition example.<<ETX>>