Jan Schuchardt

Research interests. My research is focused on providing provable guarantees for the trustworthiness of machine learning methods at both training and inference time. In particular, I am interested in methods that account for most real-world data not being generic collections of numbers, but structured data: Grids (images), sequences (language / time series), or graphs (social networks / databases). My main focus is on pobabilistic methods, which introduce randomness into the training algorithm or model’s prediction to provide statistical guarantees for their robustness to input modifications. Depending on how we randomize and which modifications we consider, these statistical guarantees ensure privacy, safety, fairness, and various other nice properties for machine learning systems.

Current position. I work as a research scientist in the Morgan Stanley Machine Learning Research Department, where I continue to publish on trustworthy ML. I am currently based in London. In my applied work, I currently focus on generative models for (irregularly sampled) multivariate time series.

Education. In 2026, I completed my Ph.D. in computer science at TU Munich’s Data Analaytics and Machine Learning Group. If you’re interested in how probabilistic certificates of adversarial robustness (“randomized smoothing”) relate to probabilistic certificates of privacy (“differential privacy”), and how to specialize them to graph neural networks, you can check out my dissertation here.