Method and network arrangement for obtaining the output signals gradient of a given network for processing time discrete signals in relation to their network parameters

摘要

A method for obtaining the gradient of the output signals (y) of a network (NW) for processing time-discrete signals with respect to the network parameters ( alpha ), in which the network consists of a first subnetwork (N1) which only has delay-free elements, to which one or more input signals (x) are applied and which outputs the output signal or signals (y), and a second subnetwork (N2) which only exhibits delaying elements, which is supplied with one or more signals from the first subnetwork and which feeds these signals delayed back to the first subnetwork, in which method further output signals corresponding in each case to the gradient of the output signals with respect to the network parameters are formed which are equal to the sum over all signals which result from the logical combination of all input signals (x), all signals (c) delayed by the second subnetwork and all network parameters via a first logic function and over all signals which result from the logical combination of all input signals, all signals delayed by the second subnetwork and all network parameters via a second logic function multiplied by internal signals which are delayed in accordance with the second subnetwork, in which method the first transfer function is equal to the gradient of the transfer function of the first subnetwork with respect to the network parameters, in which method the second logic function is equal to the gradient of the transfer function of the first subnetwork with respect to the signals delayed by the second subnetwork and in which method the internal signals are equal to the gradient of the signals (d) output by the first subnetwork to the second subnetwork with respect to the network parameters. IMAGE