Scientific Computing Researcher
Fengyun Yu
M.Sc. Scientific Computing, Universität Heidelberg (2023-2026)
I want to build trustworthy machine learning for health and biomedicine, systems whose answers you can inspect, not just score. More importantly, I want to build systems that can learn from data. My thesis at the German Cancer Research Center (DKFZ) asked what a spatial-transcriptomics graph model has actually learned, and whether higher accuracy really means better biology.
Based in Heidelberg. Away from the keyboard, I'm usually running or cooking.
Research Interests
Where machine learning meets health, and where I try to ask what a model actually knows.
Interpretable Geometric Deep Learning
Opening up graph neural networks for spatial transcriptomics: what their attention learns, and why accuracy metrics alone can decouple from biological meaning.
Causal Inference
Discrete choice experiments and clustering on health data, from COVID-19 vaccine preferences (n=12,000) to COPD patient phenotypes (n>10,000).
Health Economics
Modeling the global economic burden of chronic disease and benchmarking mortality risk prediction on large cohorts such as UK Biobank.
Latest from the Blog
Notes on research, engineering, and life.
2026-10-08
Attention mechanism in recruitment and promotion (academia): from references to peer reviews
From refeerences letter to peer review, we are accustomed to identify individual by the attention strength from the big names. In AI era, is there any better way to achieve that?
2026-08-25
Hello from the New Blog Editor
Testing the new Milkdown-based blog editor end to end.