Learning temporal sequences by complex neurons with local feedback

To deal with temporal sequences is a very important and difficult problem for application of neural networks. We propose a multilayer network using complex neurons with local feedback (MNCF). A complex neuron can keep previous information by using the phase component. We derive a simple learning algorithm based on backpropagation for temporal sequences. It is shown in computer simulations that the proposed network has better ability than conventional real ones, including Elman's network.