Recurrent neural networks for temporal learning of time series

The learning and performance behaviors of recurrent 3-layer perceptrons for time-dependent input and output data are studied. In the first task, the net learns the association of various input functions with corresponding target functions. In the recall phase, at the output, the net provides approximations to target trajectories for corresponding noise-corrupted input functions. In the second task, the net is trained to continue a trajectory that has been presented with some noise for a fixed interval. The theoretical framework of the authors´ investigations is the unified treatment of neural algorithms for time-dependent patterns. To cope with the increased learning time, the Ring Array Processor is used