Posts

We hosted Dr. Marius Herr from the University of Tübingen, who presented FLAME — Federated Learning and Analyses in Medicine, a platform developed within the privateAIM initiative. He outlined how patient data can be analyzed in a privacy-preserving manner and demonstrated how our methods could be applied to clinical data through FLAME.

Azza joins our ML sub-team as a doctoral researcher working on uncertainty quantification and estimation for trustworthy AI. She brings experience in applied machine learning and language model engineering and will support our ongoing research on reliable ML methods in oncology.

This work presents Spectral Uncertainty, a new way to decompose uncertainty in large language models. Using the Von Neumann entropy, the method distinguishes aleatoric from epistemic uncertainty and incorporates detailed semantic structure in model outputs.

A proud moment: Florian has been listed as a Highly Cited Researcher 2025, placing him among the top 1% of scientists worldwide. This recognition reflects the shared work, ideas, and energy that move our lab forward every day.

Hendrik joins our ML sub-team, working on machine learning methods that combine large-scale EHR data with other data types such as genomics. He focuses on causality and uncertainty estimation and will support our ongoing methodological work.

Yussuf joins our bioinformatics sub-team as a doctoral researcher working on interpretable probabilistic machine learning models for multimodal data integration. He applies these methods to projects on novel mRNA technologies for colorectal and pancreatic cancer and will support our ongoing research efforts.

We present a unified framework that combines expressive neural embeddings with interpretable multi-output Gaussian processes for single-cell genomics. Joint representations of cells and genes reveal meaningful links between cell clusters and their marker genes via an interpretable gene-relevance map.

We introduce ModelAuditor, a self-reflective agent that simulates clinically relevant distribution shifts and produces interpretable reports on likely failure modes. Across multiple medical imaging domains, it recovers up to 25% of performance lost under shift while providing actionable deployment insights.

We present the first systematic framework for application-driven validation of posterior-based methods in inverse problems. Adapting concepts from object detection enables mode-centric validation with interpretable, application-focused metrics, demonstrated on multiple medical imaging use cases.

We introduce LVLM-Aided Visual Alignment (LVLM-VA), which aligns small vision models with human domain knowledge using large vision-language models. A bidirectional interface translates model behavior into natural language and expert instructions into image-level critiques, improving performance while reducing fine-grained feedback needs.