Asynchronous learning dynamics in massively parallel recurrent neural networks

Summary form only given. A mathematical basis of the concurrent asynchronous relaxation method for the parallel learning of recurrent neural networks has been proposed. The condition for the asynchronous relaxation learning method of recurrent networks to converge in multiprocessor systems was developed based on partially asynchronous gradient descent optimization theory. The parallel learning of recurrent neural networks was successfully implemented on a CRAY X-MP using Macrotasking and an iPSC/2 using asynchronous communication. The recurrent neural network is trained to learn the behavior of a class of aperiodic or chaotic nonlinear differential-delay equations by Mackey and Glass