Hardware/Software Co-design Implementation of On-Chip Backpropagation

An on-line trained neural controller with a fuzzy learning rate of the Levenberg–Marquardt algorithm for speed control of an electrical drive with an elastic joint

“…Most of popular adaptation algorithms need information about dependence of cost function on weights changes (it is often represented by the Jacobian matrix with first-order partial derivatives) [2]. The exact level presenting the influence of several weights on output error is obtained mostly using propagation of the error through neural network [3]. The second group of training algorithms based on the second order derivatives of the cost function according to weights (represented by Hessian matrix).…”