Ramneet Kaur is a Ph.D. candidate at the PRECISE Center at the University of Pennsylvania. She is mentored by Insup Lee and Oleg Sokolsky. Her research interest lies in the intersection of Cyber-Physical Systems, Safety Guarantees, and Machine Learning. Specifically, her work provides statistical guarantees on the detection of novel scenarios where machine learning models are prone to make mistakes, with a focus on safety-critical cyber-physical systems applications. Ramneet's research has been published in top-tier cyber-physical systems and machine-learning conferences, including HSCC, ICCPS, ICAA, and AAAI. She won the best paper award in ICCPS, 2022. Her ICCPS papers have been rewarded with the Reusable and Reproducible Badge. She has served as the reviewer for multiple conferences and journals, including TCPS, ICCPS, Safecomp, RV, and, Neurips. She is also the program committee chair of DESTION, IEEE Workshop on Design Automation for CPS and IoT.
With the remarkable performance of machine learning models such as Deep neural networks (DNNs) across domains, there is significant interest in deploying these models in CPS with safety guarantees. The black-box nature of DNNs makes it difficult to interpret their uncertain predictions on perturbed inputs or even inputs from novel environments from training. This limits the trustworthy deployment of these models in high-assurance systems with dynamically changing environments such as autonomous driving. My research focuses on providing solutions for the reliable deployment of learning components in safety-critical CPS. Specifically, I have developed tools for monitoring distribution shifts and physical attacks on the inputs to learning components with bounded false alarm rates. I have also proposed solutions to quantify the performance of DNNs in novel environments as an assurance measure for the deployment of these models in the real world. Currently, I am working on runtime monitoring of the system-level safety specifications for CPS with learning components while considering distribution shifts at inference time.