Karl-Philipp Kortmann

Karl-Philipp Kortmann, Moritz Fehsenfeld, Mark Wielitzka

Sensor and control data of modern mechatronic systems are often available as heterogeneous time series with different sampling rates and value ranges. Suitable classification and regression methods from the field of supervised machine learning already exist for predictive tasks, for example in the context of condition monitoring, but their performance scales strongly with the number of labeled training data. Their provision is often associated with high effort in the form of person-hours or additional sensors. In this paper, we present a method for unsupervised feature extraction using autoencoder networks that specifically addresses the heterogeneous nature of the database and reduces the amount of labeled training data required compared to existing methods. Three public datasets of mechatronic systems from different application domains are used to validate the results.

Max-Heinrich Laves, Sontje Ihler, Karl-Philipp Kortmann et al.

The model uncertainty obtained by variational Bayesian inference with Monte Carlo dropout is prone to miscalibration. In this paper, different logit scaling methods are extended to dropout variational inference to recalibrate model uncertainty. Expected uncertainty calibration error (UCE) is presented as a metric to measure miscalibration. The effectiveness of recalibration is evaluated on CIFAR-10/100 and SVHN for recent CNN architectures. Experimental results show that logit scaling considerably reduce miscalibration by means of UCE. Well-calibrated uncertainty enables reliable rejection of uncertain predictions and robust detection of out-of-distribution data.

Max-Heinrich Laves, Sontje Ihler, Karl-Philipp Kortmann et al.

Model uncertainty obtained by variational Bayesian inference with Monte Carlo dropout is prone to miscalibration. The uncertainty does not represent the model error well. In this paper, temperature scaling is extended to dropout variational inference to calibrate model uncertainty. Expected uncertainty calibration error (UCE) is presented as a metric to measure miscalibration of uncertainty. The effectiveness of this approach is evaluated on CIFAR-10/100 for recent CNN architectures. Experimental results show, that temperature scaling considerably reduces miscalibration by means of UCE and enables robust rejection of uncertain predictions. The proposed approach can easily be derived from frequentist temperature scaling and yields well-calibrated model uncertainty. It is simple to implement and does not affect the model accuracy.