PhD Candidate (f/m/x) - Efficient Reasoning Architectures for Medical AI

HPI is recognized as one of the leading university institutes for computer science ( ). Our goal is to strengthen Europe\'s digital sovereignty with excellent research and practice-oriented teaching in computer science. At HPI, we are convinced that excellence is created when knowledge from different disciplines is collaboratively combined to create responsible and pioneering solutions for the future. In our research clusters "Systems," "Data and AI," "Foundations," "Digital Health," "Security," and "Business and Society" researchers from various disciplines work together on topics relevant to solving the challenges facing society. The Digital Engineering Faculty of the University of Potsdam and HPI offers a computer science engineering degree program that is unique in Germany. PhD Candidate (f/m/x) - Efficient Reasoning Architectures for Medical AI The position is temporary (3 years, with possibility of extension) and full-time, starting from July 2026. The Chair of Human-Centred Transformative AI at HPI develops AI that is small enough to be sovereign, capable enough to be useful, and trustworthy enough to be deployed in high-stakes health decisions. Our core research spans efficient reasoning architectures, interpretability and data attribution, and differential privacy — all aimed at enabling compact, private, and auditable medical AI. We work with leading clinical and research partners including Mount Sinai Health System in New York. Design and investigate parameter-efficient transformer architectures — including recurrent-depth, looped, and weight-sharing designs — that achieve high effective model depth without proportional parameter cost Develop adaptive compute allocation and early-exit mechanisms that allow models to dynamically adjust inference depth based on input complexity, enabling efficient deployment on resource-constrained devices Establish and systematically evaluate compact model baselines (sub-4B parameters) on medical reasoning benchmarks Investigate the learning-theoretic foundations connecting looping, chain-of-thought reasoning, and algorithmic reasoning in compact architectures Contribute to open-source evaluation toolkits and benchmarks for efficient medical AI Publish at top-tier venues (NeurIPS, ICML, ICLR, AAAI) and contribute to the group’s collaborative research programme Master’s degree (or equivalent) in computer science, machine learning, mathematics, or a closely related field Strong background in at least one of: transformer architectures, efficient deep learning, recurrent or state-space models, data attribution, or neural architecture design Solid mathematical foundations in optimisation, learning theory, or computational complexity Proficiency in Python and modern deep learning frameworks (PyTorch preferred) Strong publication record or demonstrated research potential (e.g. outstanding thesis) Interest in health and medical AI applications is a plus but not required Fluent English; German is welcome but not required What we offer A first-class, active and inspiring research environment at the intersection of efficient AI, privacy, and health The opportunity to produce research that truly has an impact — our group\'s mission is to build AI that empowers patients Collaboration with leading international partners including Mount Sinai Health System (New York) and HPI’s Digital Health Cluster Regular team events and workshops For problems of any kind (including those in your private life), you can use our external employee counseling service free of charge Attractive employee offers (corporate benefits) Possibility to work remotely (up to 2 days a week) 30 days of vacation plus two additional days off on Christmas Eve and New Year\'s Eve Numerous sports facilities (tennis and basketball courts, beach volleyball court, calisthenics park, and participation in running events) Workplace on the HPI campus in Potsdam-Babelsberg (Griebnitzsee S-Bahn station), with fast transportation connections to Berlin We look forward to receiving your application, including motivation letter, CV, relevant transcripts or degree certificates, a sample of your scientific work (e.g. thesis, excerpt or publication), reference letters (if available), as well as your salary requirements by May 31, 2026 . The expected start date is July/August, 2026, and interviews will be conducted in June 2026. Please apply exclusively via our career page. Contact Research Group View email address on click.appcast.io #J-18808-Ljbffr Plattnerfoundation