Statistics, probabilistic machine learning, and the occasional detour.

I'm Jan — a statistician and technical product manager in agricultural reinsurance based in Zurich. I recently completed my M.Sc. in Statistics at ETH Zurich, working on probabilistic AI, high-dimensional statistics, and causality. This is where I write up my projects and whatever I find interesting outside of work.

Download CV More about me

Does Catastrophic Forgetting Happen in Tiny Subspaces?

Machine Learning

Catastrophic forgetting appears to live in the same low-curvature bulk subspace where ordinary learning happens, tested on Permuted MNIST and Split-CIFAR.

Detecting Pneumonia from Chest X-Rays: CNNs, Integrated Gradients, and Grad-CAM

A ResNet-50 classifies chest X-rays for pneumonia, while Integrated Gradients and Grad-CAM show it attends to clinically sensible lung regions.

Detecting Heart Attacks from ECGs: Classical ML, Deep Learning, and Representation Learning

From logistic regression to transformers on ECG signals: how far feature engineering closes the gap to deep learning, and whether contrastive embeddings transfer.

Predicting Heart Disease: When Simple Models Rival Deep Learning

A Lasso regression, an MLP, and a neural additive model predict heart disease about equally well, and independently agree on which clinical features matter.

A Theoretical Primer on Proper Scoring Rules

Machine Learning

We examine the application of proper scoring rules in evaluating probabilistic forecasts. These rules provide a rigorous framework for assessing the quality of probabilistic…

Data-Driven Tactics: Analyzing Spain’s Playing Style in Euro 2024 Qualifiers

Sports Analytics

Passing networks, offside patterns, and defensive lines: what event and tracking data reveal about Spain’s Tiki-Taka in the Euro 2024 qualifiers.

A Visual Introduction to Copulas

Quantitative Finance

What copulas are, why they matter for modelling dependence, and how they are used in finance, with the heavy math left out.

Comparing Univariate and Multivariate Models for Value at Risk Forecasting

Quantitative Finance

Do multivariate GARCH models beat univariate ones at forecasting Value at Risk? Backtests say yes; predictive accuracy says it is complicated.