Thursday, June 22, 2023 | 09:00 am (CET) | Room: B04.1.06 | Lakeside Science & Technology Park
Ass.-Prof. Dr. Jan Steinbrener | Department of Smart System Technologies (Control of Networked Systems group) at Alpen-Adria-Universität Klagenfurt
Abstract: Deep neural networks (DNNs) have become an important tool in many fields of applications from image recognition to natural language processing and beyond, often outperforming human experts in their domains. Compared to heuristic, expert algorithms or shallow machine learning models, DNNs benefit from better prediction accuracy and better generalizability to unseen data. This comes at the cost of resource and data-intensive training of these models and a black-box-like behavior that does not provide information about underlying reasoning or uncertainty of the predictions. In robotics, DNNs have been successfully applied to diverse tasks such as state estimation, path planning, and control for various different platforms. This talk will explore the application of deep neural networks for sensor data processing with a particular focus on state estimation for robotic applications. End-to-end trainable deep neural networks that directly predict the desired state based on raw sensory inputs as well as hybrid models where the predictions of the DNNs are fused with other sensor data in a statistical state estimator will be discussed. Finally, strategies how to quantify model and task-based uncertainties of DNN predictions with the goal to improve the consistency of DNN-based state estimators will be presented.
Bio: Jan Steinbrener is an assistant professor on a tenure track position in the Control of Networked Systems group (CNS) at the University of Klagenfurt. He obtained his PhD in Physics in 2010 from Stony Brook University in Stony Brook, NY USA. After his PhD, he worked as a postdoctoral researcher at the Max-Planck Institute for medical research in Heidelberg, Germany, and then spent 5 years working in industry developing medical x-ray machines at Siemens Healthcare in Erlangen, Germany. Before joining CNS in 2019, he worked as a senior researcher at the Carinthian Tech Research Centre (now Silicon Austria Labs) in Villach Austria.
His current research focuses on combining machine learning approaches with classical methods for state estimation and navigation of autonomous systems. He has authored or co-authored more than 40 peer reviewed publications on novel imaging systems, image processing and reconstruction techniques, applied machine learning, machine learning algorithm development, and combination of machine learning with classical filters for state estimation. He currently holds 2 patents on image processing techniques.