E-Mail: firstname.lastname@tum.de
ORCID:
0009-0006-6397-3100Address:
Institut für Informatik (I7) Technische Universität München Boltzmannstr. 3 D-85748 Garching bei München / Germany
Sabine Rieder
Office: 03.11.39
E-Mail: firstname.lastname@tum.de
ORCID: 0009-0006-6397-3100
Address:
E-Mail: firstname.lastname@tum.de
ORCID:
0009-0006-6397-3100Address:
Since 2021, I have been a PhD student at the Technical University of Munich; in 2024, I also enrolled at the Masaryk University in Brno for a joint program between the two mentioned universities.
Previously, I was a researcher at Audi AG in Ingolstadt.
My research interests revolve around the safety of Neural Networks.
I am focused on two areas: (i) the abstraction of such systems to speed up verification and (ii) runtime monitoring since verification is not yet applicable to larger neural networks.
Research Areas
Runtime Monitoring of Neural Network
We develop runtime monitors for neural networks to improve their reliability.
Abstraction of Neural Network
We abstract neural networks to improve verification speed.
Machine Learning for LTL Synthesis
We apply machine learning to the LTL synthesis problem.
Tools
MONITIZER
We create a tool (Monitizer) that optimizes monitors on a NN for a specific task.
Publications
|
2024 |
Gaussian-Based and Outside-the-Box Runtime Monitoring Join Forces
RV 2024 |
|
Monitizer: Automating Design and Evaluation of Neural Network Monitors
International Conference on Computer Aided Verification |
|
|
2023 |
Guessing Winning Policies in LTL Synthesis by Semantic Learning
CAV 2023 |
|
Runtime Monitoring for Out-of-Distribution Detection in Object Detection Neural Networks
FM 2023 |
Student Projects
Open Projects
Evaluation of State-of-The-Art Runtime Monitoring Techniques
In this project, we want to compare different runtime monitoring techniques. We want to evaluate them on several types of unseen data and datasets to investigate the performance and capabilities of the techniques. Experiments will be carried out with the help of Monitizer.
|
Decision Trees for Monitoring Neural Networks
We want to explore the possibilities of using Decision Trees as a more understandable monitor for Neural Networks, potentially in combination with other monitors.
|
Ongoing Projects
Yifei Xu: A Framework for Evaluation of Runtime Monitors for Object Detection Neural Networks
While Object Detection Neural Networks are of high practical relevance, there is no standard way to evaluate them yet. In addition, not many monitors for this type of NN are known. In this thesis, we develop a framework to evaluate such monitors and present basic network monitoring methods.
|