Keynote Talk: Christian Ledig

From Images to Embeddings: Is End-to-End Learning Coming to an End?

Christian Ledig

Christian Ledig

Abstract

As advancements in artificial intelligence transform medical image classification, the established paradigm of end-to-end learning is facing scrutiny. This talk investigates a promising shift from raw images to embeddings derived from foundation models, which enhance robustness and sample efficiency. By utilizing these low dimensional, yet highly descriptive feature embeddings we can improve data privacy through more flexible data exposure, increase transparency with more interpretable classifiers, and maximize label efficiency, enabling models to learn from fewer annotated examples. We will discuss the significant potential of a paradigm shift to static feature extractors in our goal of building responsible AI systems that are more robust, flexible, and efficient while maintaining high accuracy.

Speaker’s Bio

Prof. Dr. Christian Ledig is a professor at the Otto-Friedrich-University Bamberg. Since April 2022, he has been leading the Chair of Explainable Machine Learning at the Faculty of Information Systems and Applied Computer Science. Prof. Ledig is a member of the management board of the Bamberg Center for Artificial Intelligence (BaCAI). He leads the xAILab Bamberg, which conducts research focused on explainable and robust machine learning approaches and their translation into practice, particularly in healthcare.

Prof. Ledig studied Applied Mathematics (Dipl. Technomathematik) at the Friedrich-Alexander University Erlangen-Nuremberg and earned his doctorate at Imperial College London, UK, in the field of Medical Image Analysis within the BioMedIA group. As part of various research projects, Prof. Ledig has worked at the Boston Children’s Hospital at Harvard Medical School in the USA and at the University of Cambridge in the UK.

Before his professorship, Prof. Ledig conducted AI research in the industry, including positions at Twitter in London and, most recently, in the US startup sector in New York and Boston. There, he led research teams and successfully developed FDA-regulated, AI-driven healthcare systems that are now approved and improving patient treatment.

He has more than 15 years of research experience in machine learning, computer vision, and image analysis in both industry and academia. His focus is on the development of AI-driven medical applications that enable better healthcare by reducing diagnostic errors and improving access to high-quality medical diagnoses. Throughout his career, Prof. Ledig has published two book chapters and more than 75 research articles, which have been cited over 25,000 times. This places him among the top 2% of the most-cited scientists worldwide. Additionally, he is active as a reviewer for various leading conferences and journals, such as the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Medical Image Computing and Computer-Assisted Intervention (MICCAI), and IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI).