The invention relates to a method (100) for training an artificial neural network, ANN (1), which translates one or more input variables (11) into one or more output variables (13), by means of learning data sets (2) that comprise learning input variable values (11a) having measurement data, and associated learning output variable values (13a), the method comprising the steps of: • mapping (110) learning input variable values (11a) from at least one learning data set (2) onto output variable values (13) by means of the ANN (1); • processing (120) deviations of the output variable values (13) from the respective learning output variable values (13a) in accordance with a cost function (14) to form a measure of the error (14a) of the ANN (1) when processing the learning input variable values (11a); • determining (130), from the error (14a), by backpropagation, changes in parameters (12), the execution of which, when learning input variable values (11a) are further processed by the ANN (1), is likely to improve the evaluation of the thus obtained output variable values (13) by the cost function (14), and applying (140) said changes to the ANN (1); • wherein a subset (13*) of the output variable values (13) is excluded (131) at least from consideration in the backpropagation (130).
展开▼
