01
Welcome to the Reliable Medical AI research group!
About Us
Researchers and students in the Reliable Medical AI group develop reliable and privacy-preserving AI methods for analyzing clinical data, including medical images, text, and speech. Reliable Medical AI is led by Soroosh Tayebi Arasteh and focuses on verifiable clinical decision support, learning from model-generated supervision, representation learning, and privacy-preserving deep learning.
Research Directions
We work across vision, language, and speech at hospital scale. The methods are retrieval-augmented agents, self-supervision under model-generated labels, the geometry of frozen encoders, and differentially private training.
Latest news
-
August 28, 2026
The ARISTOTLE proposal to the Federal Ministry of Research, Technology and Space is approved. Over the next 3 years the group will work on it with the Technical University of Munich and the University Hospital of TUM.
-
August 26, 2026
RSNA published an interview with Soroosh on how cognitive biases mislead radiology language models. Read it
-
June 9, 2026
The group presented 6 abstracts at Baiosphere Medical 2026 in Erlangen.