A novel approach to error function minimization for feedforward neural networks

摘要

Feedforward neural networks with error backpropagation (FFBP) are widelyapplied to pattern recognition. One general problem encountered with this typeof neural networks is the uncertainty, whether the minimization procedure hasconverged to a global minimum of the cost function. To overcome this problem anovel approach to minimize the error function is presented. It allows tomonitor the approach to the global minimum and as an outcome severalambiguities related to the choice of free parameters of the minimizationprocedure are removed.