I'm a final-year PhD candidate at Helmholtz Munich & TU Munich, specializing in efficient machine learning and computer vision for biomedical imaging.

My research focuses on building compute-efficient models (like zero-shot and one-shot learning) that perform in data-scarce environments. Additionally, I am currently developing foundation models for applications in the biomedical field. I have a proven record in the full research lifecycle, from initial concept to first-author publication at A* conferences (ECCV, ICCV, AAAI, MICCAI) and to deployed, production-ready solutions.

Portfolio / LinkedIn / Google Scholar

1. CoLIE (ECCV'24): One-shot model that outperforms fully-supervised methods in low-light image enhancement and achieves near real-time inference. There's 🤗 demo and Google Colab demo.
2. Noise2Detail (MICCAI'25): Ultra-lightweight 22k-parameter data-free denoising framework, 100-1000x smaller than typical denoising U-Nets, solving a key bottleneck for biomedical imaging where training data and compute are scarce. Google Colab demo
3. Privacy risks in Medical AI (AISec'23): One of the first exhaustive studies of membership inference attacks on semantic segmentation models. The code quantitatively benchmarks the vulnerability of popular architectures and analyzes the trade-offs of various defenses.