Assessment of a Neural Network for the Correction of Measurement Errors

Many modern technology trends such as autonomous driving, image processing or speech recognition are prime examples for the application of machine learning methods. However, the application of these methods is often omitted in safety critical areas since verification and validation of the learned features are problematic. The following work uses a multilayer perceptron in order to correct measurement errors for an angular sensing system and suggests a methodology to extract the learned features. This is possible since the considered system is clearly defined with well known parameters, which is typically for most sensing applications.