The project CausalAnomalies is about the development of new causally explainable anomaly detection algorithms and the application of these to DLR-specific problems. Finding anomalous states in time series is a ubiquitous goal - wherever sequential data are acquired and analyzed. These data are often high-dimensional, error-laden, incomplete, or unlabeled. And it is often not known what anomalous behavior of the data is relevant. Therefore, a first goal is to explore AI methods for anomaly detection. The next goal is to make these methods explainable to overcome the "black box" nature of many methods. For this purpose, causal inference methods will be used and adapted. The combination of common and new methods for anomaly detection with methods of causal inference will be done generically and application-independent. Subsequently, the developed methods will be transferred to different application cases at DLR, tested, and optimized if necessary.
Project runtime: 12/2021 - 12/2025
Spokeperson: Julia Fligge-Niebling, Sharmita Dey, Andreas Gerhardus