3/2024 - Project explanation and research updates of the SKIAS project
As part of the SKIAS project, eight DLR institutes are working together on various building blocks for a major goal: safe AI for autonomous systems. Under the leadership of the DLR Institute of Optical Sensor Systems, selected fundamental methods and technologies are being researched in order to better understand artificial intelligence systems, make them safer and prepare their transfer to industry and society. The main focus here is on the reliability of AI predictions and possible uncertainties, sensor-related method implementations and suitable metrics for quantifying various quality characteristics.
We, as the Institute of Data Science, are represented in three different areas:
Together with the Institute of Robotics and Mechatronics, we are working on methods to quantify uncertainties in the predictions of neural networks. In particular, an impure data basis and changes in the distribution of the measured data play a major role for practical applications.
The use of AI as a software component also entails particular security risks. These are, for example, adversarial examples or membership inference attacks. We investigate how a secure design of an AI-based application can reduce such risks. We create additional security with formal verification of behavior and intelligent runtime monitoring.
Efficient and sensor-based processing of collected information is of central importance for many systems. For this purpose, we develop and evaluate various strategies to make machine learning methods more efficient and compact without having to accept significant reductions in the performance of the models.
Current status of SKIAS:
- We are in the final spurt for project completion at the end of 2024.
- Results in all three areas have already been achieved and a transfer to (further) applications and projects is already underway.
(Previous) Publications as part of SKIAS:
Project partners: