procedures for the processing of uncertainty of input data in neural networks.

摘要

In many applications, the input signals in neuronal networks are beset by considerable uncertainties. For this reason, training data are often not representative enough for the test phase data. A method of processing uncertainties in neuronal networks is proposed to solve this problem. Modified input signals are calculated from the input signals of neurons by linearly combining the input signals with neutral values. The coefficients of this linear combination are measures of the certainty of the input signals.