I work on deep learning with an emphasis on both safety and interpretability. Central to my research is assessing the information value of actions, detecting novelty, and learning compressed representations.
I earned my B.Sc. from Istanbul Technical University, receiving a DAAD/TEV scholarship for graduate studies in Germany. At Karlsruhe Institute of Technology (KIT), I completed my M.Sc. with "very good" and Ph.D. with "summa cum laude" distinctions. My dissertation won the UniDAS Science Award and was an IEEE ITS Society Best Dissertation Award finalist. After my Ph.D., I held a visiting researcher position at the University of Toronto. I'm currently serving as a research group leader at the FZI and KIT. For more details, please see this page.
I engineer and align representations in deep learning models. Key aspects of my work include leveraging representation quality metrics and invariance properties of features for learning. I apply these techniques to create more interpretable and controllable models.
My research covers model-based reinforcement learning, focusing on partially observable Markov decision processes and multi-armed bandits. My work includes novelty detection, information gathering, and curriculum learning to enhance learning efficacy.
I have worked extensively on the research and engineering of autonomous driving systems. I applied my machine learning research to autonomous driving and have also developed rule-based algorithms that run in real-time on autonomous car prototypes.