About

AI Researcher · Generative Models & Foundation Model Adaptation

"Specialized in advancing un-/self-supervised and generative machine learning techniques for medical image anomaly detection. I'm passionate about leveraging technology for better healthcare. I look forward to tackling the next pivotal AI challenge in the health domain."

Professional Summary¶

Senior AI researcher at the German Cancer Research Center (DKFZ) working on generative models, unsupervised anomaly detection, and foundation model development — bridging theoretical innovation and industrial-grade engineering. I currently lead research on building Vision-Language Models (VLMs) and establishing rigorous benchmarks for model robustness, with a focus on medical imaging.

I'm based in Heidelberg, Germany and work at the German Cancer Research Center (DKFZ).

Core Competencies¶

Medical Image Analysis: Segmentation, anomaly detection & localization, out-of-distribution detection, longitudinal modeling, VLM adaptation for radiology (MRI, CT, endoscopy, histopathology).
Generative AI & LLMs: VAEs, GANs, Flows, Auto-regressive methods, Vision-Language Model (VLM) & CLIP training and benchmarking.
Model Robustness & Evaluation: Unsupervised Anomaly Detection, Out-of-Distribution (OOD) Benchmarking, Advanced Performance Validation.
Engineering & Infrastructure: Distributed training pipelines, large-scale data curation, PyTorch ecosystem (trixi, batchgenerators), Docker-based evaluation systems, HPC cluster integration (SLURM/LSF).
Leadership: Sub-group coordinator (10+ students), co-supervision of 6+ PhD students, founder of international benchmarks (MOOD), open-source maintainer.

Skills & Languages¶

Python

Java

JavaScript

German — native

English — fluent (CEFR Level C1)

Experience¶

2022 — Present

PostDoc Researcher / Scientist

DKFZ — German Cancer Research Center, Heidelberg

Pioneering the development of generative models and setting benchmark standards that influence global research trajectories. Published influential papers with 1900+ citations (400+ first author).

Foundation Model Development: Spearheading the development of Large Vision-Language Models to specialized, dense medical domains. Developing novel approaches to handle embeddings from pretrained vision encoders for high-fidelity report generation.
Data Curation at Scale: Designed and implemented pipelines for large-scale data cleaning and OOD detection to curate training data for foundation models. Built the central infrastructure to handle online data curation of raw, unreviewed imaging data.
Benchmarking & Evaluation: Founded and established the MOOD Challenge, the first international benchmark for anomaly localization (80+ teams). Created a proprietary evaluation framework utilizing a live Docker-based submission system on on-premise clusters.
Leadership: Coordinating research strategy for a team of 6+ PhD students; headed a sub-group with 10+ students.

2017 — 2022

Graduate Researcher — Deep Learning & Anomaly Detection

DKFZ — German Cancer Research Center, Heidelberg

Built and improved generative models for anomaly localization. Set new state-of-the-art performance for 3D anomaly localization on medical images. Established the first benchmark and challenge for medical anomaly localization with over 80 international participating teams.

Methodological Innovation: Developed Context-encoding/Masking VAEs (ceVAE) and Gradient/Score-based Anomaly Localization, introducing inspectability to unsupervised models by visualizing why a model flags an input as anomalous.
Engineering Excellence: Early adopter of PyTorch (Jan 2017); initiated the department-wide migration from Theano. Refactored major codebases and maintained key open-source libraries (trixi for experiment tracking, batchgenerators for high-throughput 3D data augmentation).
Impact: Published influential papers (500+ first author citations) and set methodological foundations for anomaly localization in a field that previously had <3 active publications.

2017 (6 months)

Machine Learning Researcher

understand.ai, Karlsruhe

Played a crucial role in extending and enhancing multiple object detection and segmentation methods for autonomous driving.
Implemented 5+ backbone networks and temporal context strategies to optimize model performance for real-world inference constraints.

2016 — 2017

Student Researcher

FZI Research Center for Information Technology, Karlsruhe

Early Generative Models: Developed one of the first biologically inspired trainable spiking neural networks. Based on Restricted Boltzmann machines (RBMs) and Deep Belief Networks, enabling end-to-end training.

2015 — 2016

Research Assistant

FZI Research Center for Information Technology, Karlsruhe

Robust Vision: Developed a dynamic CNN-based object detection framework tailored for urban scenarios. Integrated domain adaptation and semi-supervised techniques, optimizing the solution's adaptability and performance in diverse real-world settings.

2014 (6 months)

Student Researcher

Fraunhofer IOSB

Developed an evolutionary algorithm and SVM-based approach for "human-interpretable" feature learning and object detection.

2013 (6 months)

Teaching Assistant

Karlsruhe Institute of Technology (KIT)

Assisted in teaching Computer Architecture, covering topics including C programming, caching, processor design, pipelining, virtual memory, and floating-point arithmetic.

2012 (2 months)

Intern

Software AG

Collaborated within the University Relations department on various projects.

Education¶

2017 — 2022

Ph.D. in Computer Science (Dr. rer. nat.)

Heidelberg University

Main focus: Generative methods (VAEs, GANs, Auto-regressive methods), Out-of-distribution and Anomaly-detection, Medical image analysis.
Thesis: *"Unsupervised Learning for Anomaly Detection in Medical Images"*
Fellowship: Helmholtz International Graduate School for Cancer Research (2017)

2014 — 2017

M.Sc. in Computer Science

Karlsruhe Institute of Technology (KIT)

Achieved a top-grade course performance: 1.0 (equivalent to GPA 4.0/4.0). Focus on Cognitive Systems. Master Thesis resulted in the **Best Paper Award, ICANN 2017**.

2011 — 2014

B.Sc. in Computer Science

Karlsruhe Institute of Technology (KIT)

Thesis: *"Supervised learning of human interpretable image features for bulk good sorting"* (Fraunhofer IOSB)

Awards & Recognitions¶

🏆

Winner, HackZurich 2020

Carbon Foodprint — IBM & SwissRe Workshop

🌙

Winner, HackZurich 2019

Orange — Moonshot Award

🥇

Winner, Medical Segmentation Decathlon 2018

Acknowledging main contribution by F. Isensee

📄

Best Paper Award, ICANN 2017

Spiking convolutional deep belief networks

🎓

Helmholtz Fellowship 2017

International Graduate School for Cancer Research

💰

Grand Final Winner 2017

PayHero — POST/bank Hackathon — 20,000€

🥈

Runner-up, Audi Autonomous Driving Cup 2016

Urban scenario perception

🎤

Runner-up, Best Presentation IPCAI 2018

Crediting main contribution by T. Ross

📂 See the Projects page for detailed descriptions and tech stacks of all hackathon and side projects.

Outreach & Organizational Roles¶

2021–2024: Lead Organizer for MOOD (Medical Out-of-Distribution Analysis Challenge) in 2021, 2022, 2023, and 2024.
2021–2022: Co-Organizer for FeTS (Federated Tumor Segmentation Challenge).
2021–2023: Program Committee for DART Workshop at MICCAI.
2023–2024: Program Committee for VAND Workshop at CVPR.
2020: Founder & Organizer of the inaugural Medical Out-of-distribution Analysis Challenge (MOOD) at MICCAI.

Peer Review Contributions¶

Reviewer for esteemed conferences and journals:

MICCAI — Medical Image Computing and Computer Assisted Intervention
MIDL — Medical Imaging with Deep Learning
IEEE TMI — Transactions on Medical Imaging
Medical Image Analysis (MIA)
SPIE JMI — Journal of Medical Imaging
SPIE JARS — Journal of Applied Remote Sensing
NeurIPS Medical Imaging Workshop
ML4H @ NeurIPS — Machine Learning for Health
DART Workshop @ MICCAI

Beyond Research¶

Job offer as SWE at Microsoft (Office 365) in 2017
Organizer of the Heidelberg Triathlon 2024–2026
Passionate swimmer & triathlete
I also like coding in my free time