Reliable Medical AI

Openings

Fully Funded PhD Position on Differential Privacy for Medical Deep Learning (m/f/d)

At the Reliable Medical AI group of the Lab for AI in Medicine (RWTH Aachen University and University Hospital RWTH Aachen), we are looking for a PhD student, funded by the German Research Foundation (DFG).

Medical foundation models are pretrained on large image collections and then finetuned on the patient data of a hospital. Differentially private (DP) finetuning formally guarantees that the model reveals little about any single patient. However, DP finetuning can lower the diagnostic performance more for some patient groups than for others. The mechanisms behind these differences are not well understood. The aim of this project is to identify these mechanisms and to develop DP finetuning methods that keep the diagnostic performance high for all patient groups.

Within this project, the PhD candidate will conduct independent research toward a doctoral thesis. Using theory and large-scale experiments, the candidate will analyze how per-sample gradient clipping, noise addition, and subsampling in differentially private stochastic gradient descent (DP-SGD) interact with the geometry of pretrained representations. The candidate will then use this analysis to develop new DP finetuning methods to improve the performance of the worst-performing patient group at a fixed privacy budget. Possible directions include reliability-aware training objectives and the controlled adaptation of pretrained layers.

The candidate will have access to a GPU cluster with 30 NVIDIA H100 and 16 RTX 6000 Blackwell GPUs at the Lab for AI in Medicine of Prof. Daniel Truhn, as well as to large-scale image datasets, so that they can focus on the core part of the project from the beginning instead of on data collection. The results are expected to be published in top-tier venues such as NeurIPS, ICLR, ICML, CVPR, Nature Portfolio, and MICCAI.

Requirements

  • A master's degree in computer science, electrical engineering, or a related field
  • A strong technical and theoretical background in machine learning
  • Publications at high-quality peer-reviewed venues are strongly preferred
  • Proficient in Python and a deep learning framework such as PyTorch
  • Fluent written and spoken English
  • Desirable: experience with differential privacy
  • Hands-on experience with agentic AI tools such as Claude Code is not mandatory but is a plus

The position

This is a 3-year full-time project with an on-site position in Aachen, Germany. The position is initially limited to one year, with a possibility of extension for 2 years afterwards. The earliest possible start date is February 1, 2027. The remuneration is based on TV-L, the standard collective wage agreement for public sector employees in German states (approx. €4,759.39 gross per month in case of TV-L E13/1 2026).

Application

Please send applications by email to Dr.-Ing. Dr. rer. medic. Soroosh Tayebi Arasteh (stayebiarast@ukaachen.de). The application should include the following documents as a single PDF file:

  • CV (max. 3 pages)
  • Cover letter (max. 1 page)
  • Master's transcript
  • 3 letters of recommendation

The applications will be reviewed on a rolling basis until the position is filled. We welcome applications from all qualified candidates, regardless of gender, nationality, ethnic or social origin, religion, disability, age, or sexual orientation.

Aachen, 21.10.2026