PhD Position in Multimodal Foundation Model Alignment for Oncology (Technical University of Munich)

PhD Position (TV-L E13, 100%) Multimodal Foundation Model Alignment for Oncology 27.02.2026, Wissenschaftliches Personal Your Role You will join Subproject 3 (Model Alignment and Optimization), led by PD Dr. Keno Bressem (https://scholar.google.com/citations?user=wIEgwbkAAAAJ&hl=en). Your work sits at the interface between raw foundation model capabilities and clinical utility. Specifically, you will: - Align foundation models with clinical reasoning using Direct Preference Optimization (DPO) and reinforcement learning from human feedback, building preference datasets together with clinicians - Build and run a Red Team process with physicians, computer scientists, and patient representatives to systematically identify failure modes, hallucinations, and clinical blind spots - Optimize models for real-world deployment through quantization, distillation, and model merging, ensuring that multi-billion parameter models run on hardware hospitals can actually afford Your Profile - Completed university degree (Master or equivalent) in computer science, mathematics, physics, medical informatics, or a related field - Strong programming skills in Python and experience with deep learning frameworks (PyTorch preferred) - Experience or strong interest in large language models, multimodal learning, or reinforcement learning from human feedback - Ability to work independently and collaboratively in an interdisciplinary team of clinicians, computer scientists, and biologists - Excellent communication skills in English (written and spoken); German language skills are advantageous but not required - Prior experience in medical AI or clinical data analysis is a plus What We Offer - A structured PhD program embedded in a large-scale, nationally funded research consortium with access to unique multimodal clinical datasets - State-of-the-art GPU infrastructure for training and fine-tuning large foundation models - An interdisciplinary research environment at one of Europe’s leading university hospitals - Remuneration according to the Collective Agreement for the Public Service of the Länder (TV-L) in salary group E13 (100%) - The position is initially limited to 3 years with the possibility of extension How to Apply Please send your application as a single PDF including a cover letter, CV, transcripts, and (if available) a list of publications to: PD Dr. Keno Bressem Department of Radiology, TUM University Hospital: E-Mail: keno.bressem@tum.de Die Stelle ist für die Besetzung mit schwerbehinderten Menschen geeignet. Schwerbehinderte Bewerberinnen und Bewerber werden bei ansonsten im wesentlichen gleicher Eignung, Befähigung und fachlicher Leistung bevorzugt eingestellt. Hinweis zum Datenschutz: Im Rahmen Ihrer Bewerbung um eine Stelle an der Technischen Universität München (TUM) übermitteln Sie personenbezogene Daten. Beachten Sie bitte hierzu unsere Datenschutzhinweise gemäß Art. 13 Datenschutz-Grundverordnung (DSGVO) zur Erhebung und Verarbeitung von personenbezogenen Daten im Rahmen Ihrer Bewerbung. Durch die Übermittlung Ihrer Bewerbung bestätigen Sie, dass Sie die Datenschutzhinweise der TUM zur Kenntnis genommen haben. Kontakt: keno.bressem@tum.de Mehr Information https://radiologie.mri.tum.de/de/node/2850 If you apply for this position please say you saw it on Medicinoxy

PhD Position (TV-L E13, 100%) Multimodal Foundation Model Alignment for Oncology

27.02.2026, Wissenschaftliches Personal

Your Role
You will join Subproject 3 (Model Alignment and Optimization), led by PD Dr. Keno Bressem (https://scholar.google.com/citations?user=wIEgwbkAAAAJ&hl=en). Your work sits at the interface between raw foundation model capabilities and clinical utility. Specifically, you will:
- Align foundation models with clinical reasoning using Direct Preference Optimization (DPO) and reinforcement learning from human feedback, building preference datasets together with clinicians
- Build and run a Red Team process with physicians, computer scientists, and patient representatives to systematically identify failure modes, hallucinations, and clinical blind spots
- Optimize models for real-world deployment through quantization, distillation, and model merging, ensuring that multi-billion parameter models run on hardware hospitals can actually afford

Your Profile
- Completed university degree (Master or equivalent) in computer science, mathematics, physics, medical informatics, or a related field
- Strong programming skills in Python and experience with deep learning frameworks (PyTorch preferred)
- Experience or strong interest in large language models, multimodal learning, or reinforcement learning from human feedback
- Ability to work independently and collaboratively in an interdisciplinary team of clinicians, computer scientists, and biologists
- Excellent communication skills in English (written and spoken); German language skills are advantageous but not required
- Prior experience in medical AI or clinical data analysis is a plus

What We Offer
- A structured PhD program embedded in a large-scale, nationally funded research consortium with access to unique multimodal clinical datasets
- State-of-the-art GPU infrastructure for training and fine-tuning large foundation models
- An interdisciplinary research environment at one of Europe’s leading university hospitals
- Remuneration according to the Collective Agreement for the Public Service of the Länder (TV-L) in salary group E13 (100%)
- The position is initially limited to 3 years with the possibility of extension

How to Apply
Please send your application as a single PDF including a cover letter, CV, transcripts, and (if available) a list of publications to:

PD Dr. Keno Bressem Department of Radiology, TUM University Hospital: E-Mail: keno.bressem@tum.de

Die Stelle ist für die Besetzung mit schwerbehinderten Menschen geeignet. Schwerbehinderte Bewerberinnen und Bewerber werden bei ansonsten im wesentlichen gleicher Eignung, Befähigung und fachlicher Leistung bevorzugt eingestellt.

Hinweis zum Datenschutz:
Im Rahmen Ihrer Bewerbung um eine Stelle an der Technischen Universität München (TUM) übermitteln Sie personenbezogene Daten. Beachten Sie bitte hierzu unsere Datenschutzhinweise gemäß Art. 13 Datenschutz-Grundverordnung (DSGVO) zur Erhebung und Verarbeitung von personenbezogenen Daten im Rahmen Ihrer Bewerbung. Durch die Übermittlung Ihrer Bewerbung bestätigen Sie, dass Sie die Datenschutzhinweise der TUM zur Kenntnis genommen haben.

Kontakt: keno.bressem@tum.de