Learning scheme for recurrent neural network by genetic algorithm
Toshio Fukuda, Takashi Kohno, Takanori Shibata · 2002
Recurrent neural networks have dynamic characteristics and can express functions which depend on time. To apply these neural networks to the memory of robotic motions, i.e., trajectories of manipulators, it is necessary to determine appropriate network interconnection weights. A new learning scheme for recurrent neural networks using a genetic algorithm (GA) is presented and used to determine the interconnection weights. The GA approach is compared with backpropagation through time. Simulations illustrate the performance of the new approach.