About me
Hello and welcome to my website! My name is Felix Wagner, and I’m a PhD student at the Engineering Department of the University of Oxford supervised by Professor Konstantinos Kamnitsas. My research focuses on Computer Vision for Medical Imaging, with a specific emphasis on Federated Learning and Domain Adaptation. I am particularly interested in addressing distribution shifts in heterogeneous decentralized datasets during collaborative model training. In real-world medical applications, it is crucial to develop methods that enable AI models to adapt and perform well in new, unseen environments. In general, my interest lies in enhancing the reliability of AI models in the medical sector
Before starting my PhD journey, I earned both my bachelor’s and master’s degrees at TU Wien in Computer Science, where I focused on combining symbolic and subsymbolic AI.
News
[08/23] My first paper ⭐ Post-Deployment Adaptation with Access to Source Data via Federated Learning and Source-Target Remote Gradient Alignment ⭐ got accepted in the Machine Learning in Medical Imaging (MLMI 2023) workshop at MICCAI 2023.
[08/23] Modality Cycles with Masekd Conditional Diffusion for Unsupervised Anomaly Segmentation in MRI from my colleague Ziyun Liang, in which I am coauthor, got accepted in Multiscale Multimodal Medical Imaging workshop in MICCAI 2023.
[09/22] I joined Professor Kamnitsas’ lab and started research on Federated Learning for Medical Imaging
[12/21] Completed my master’s degree at TU Wien with a master’s thesis on injecting symbolic knowledge into Knowledge Graph Embeddings
[09/21] Received the Angela-Krosik scholarship from the Anglo-Austrian Society
[09/21] Started first year of the Health Data Science CDT PhD programme in Oxford with one year of courswork in Statistics, Machine Learning, Ethics, Health and Biomedical research.
Publications
Post-Deployment Adaptation with Access to Source Data via Federated Learning and Source-Target Remote Gradient Alignment
Felix Wagner, Zeju Li, Pramit Saha, Konstantinos Kamnitsas
Machine Learning in Medical Imaging (MLMI 2023) workshop at MICCAI 2023
Usually deployed models cannot adapt reliably because we often do not have access to the training data anymore because of privacy conerns. In this paper we enable deployed models to access training data via federated learning without data exchange. We propose an algorithm to align gradients of source and target data to adapt deployed models to a target distribution.