Felix Karl Michael Wagner
PhD in Computer Vision & Medical AI
Advancing AI through federated learning, parameter-efficient fine-tuning, domain adaptation, and privacy-preserving medical imaging systems at the University of Oxford.
About Me
I'm a final-year PhD student at the Department of Engineering Science, University of Oxford, supervised by Professor Konstantinos Kamnitsas. My research sits at the intersection of Computer Vision, Medical Imaging, and Privacy-Preserving AI, with the goal of making AI models more robust, adaptable, and usable in real-world healthcare settings.
Before starting my PhD, I completed both my bachelorβs and masterβs degrees in Computer Science at TU Wien, where I explored the combination of symbolic and subsymbolic AI in knowledge graphs. Alongside academia, I gained practical experience as a freelance software engineer and various internships, giving me a solid foundation in building and deploying real systems.
Research Focus
I am passionate about building robust AI systems that can be applied effectively in real-world medical environments. My research has focused on the following areas:
Medical Imaging AI
Brain MRI segmentation and analysis
Federated Learning
Privacy-preserving collaborative ML
Domain Adaptation
Handling distribution shifts
Foundation Models
Vision-Language model fine-tuning
3D Segmentation
Volumetric medical image analysis
Parameter-Efficient Fine-Tuning
Lightweight model adaptation techniques
π― Current Impact
- 4 papers accepted in 2025 at top-tier venues (CVPR 2025, WACV 2025, AAAI 2025)
- Honourable Mention as Outstanding Reviewer at MICCAI 2024
- Novel federated learning frameworks enabling cross-institutional AI collaboration, released as open-source on GitHub to advance reproducible research
News
Selected Publications
DIsoN: Decentralized Isolation Networks for Out-of-Distribution Detection in Medical Imaging
Pre-print: Under Review
We introduce DIsoN, a novel decentralized framework for out-of-distribution detection. By comparing test samples with training data without data exchange, DIsoN improves reliability without compromising privacy.
FΒ³OCUS--Federated Finetuning of Vision-Language Foundation Models with Optimal Client Layer Updating Strategy via Multi-objective Meta-Heuristics
IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) 2025 Spotlight
A novel layer updating strategy for parameter-efficient finetuning of Vision-Language Foundation Models in resource constrained federated settings. FΒ³OCUS optimizes client-specific layer importance and inter-client layer diversity using multi-objective meta-heuristics.
Feasibility of Federated Learning from Client Databases with Different Brain Diseases and MRI Modalities
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2025 Oral
We introduce FedUniBrain, demonstrating for the first time that a single federated model can segment multiple brain lesion types from databases with different MRI modalities, all without data exchange between institutions.
Post-Deployment Adaptation with Access to Source Data via Federated Learning and Source-Target Remote Gradient Alignment
Machine Learning in Medical Imaging (MLMI 2023) workshop at MICCAI 2023 Oral
We enable deployed models to adapt to new target distributions by accessing training data via federated learning. Our novel gradient alignment algorithm preserves privacy while enabling effective domain adaptation.
Why This Matters
In an era where AI is transforming healthcare, my research focuses on making AI in healthcare more safe, private, and reliable across diverse medical institutions. The goal is to enable broader access to advanced AI while protecting patient data and ensuring trust in real-world use.