Berlin Initiative for Applied Foundation Model Research

• Subject Area: Computer Science
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 528483508

Demographic changes pose major challenges for our society, three of which are at the heart of our work. (1) Our workforce needs to be productive with less human labor available. In the future, robots may complement human labor. (2) An increasingly elderly population will require more medical care. Ideally, machines will provide doctors with a second opinion to prevent overlooking rare diseases. (3) Additionally, basic research in computational cell biology can help to support the diagnosis of diseases caused by bacterial infections in the long term. In response to such challenges caused by major demographic changes, we choose three research scenarios: Robotics and Behavioral Learning, Quantitative Biology, and Predictive Medicine. AI systems working across these scenarios require learning data efficiently from multiple modalities, scarce data, with sparse distributions, and in very high dimensionality. Foundation models (FMs) can address these requirements. This new, successful paradigm enables the training of one model on broad data at scale and adapting it to many applications. Some of these powerful models already generate photo-realistic images from text input or perform real-time communication on a large scale. However, adapting these models to further applications is not trivial, which blocks their use for a majority of industrial, medical and scientific use-cases, which is exactly the overarching issue we want to address in our RI.Therefore, our scientific goals focus on understanding (1) how these FMs can be adapted with small, sparse training data and (2) novel brain-inspired architectures. We would like to understand (3) how these models can learn continuously, (4) how we can learn complex motion by demonstration, and finally (5) how we can holistically evaluate their robustness, fairness, and explainability in an automated fashion.Our multidisciplinary team at Berlin's major HAWs (BHT and HTW) is well equipped to solve the proposed difficult research questions. With over 40 PhD students and over 25 professors already working across three research centers, our measurable impact is proven by more than 50 publications at CORE A* and A venues, nine PhD graduations in the last three years, and more than four million euros annual funding. We established infrastructure facilities for deep learning (DL), robotics, and quantitative biology, and created several startups offering applications of AI. Finally, we established a network of world-class partners in machine learning (ML) and our research areas. Our structural goal is to build the Institute for Interdisciplinary Research on Foundation Models in Berlin around this already strong nucleus, with a focus on HAW-oriented, interdisciplinary basic research. This structure will attract top talent to the capital and provide PhD and postdoc training in different fields. We will translate basic research in FMs into applied research and make novel models publicly available.

• DFG Programme: Research Impulses
• Applicant Institution: Berliner Hochschule für Technik (BHT)
• Participating Institution: Hochschule für Technik und Wirtschaft Berlin
• Spokesperson: Professor Dr. Felix Gers
• Participating Researchers: Professor Dr. Felix Biessmann; Professor Dr.-Ing. Ivo Boblan; Professorin Dr. Elisabeth Grohmann; Professor Dr.-Ing. Kristian Hildebrand; Professor Dr.-Ing. Hannes Höppner; Professor Dr.-Ing. Alexander Löser; Professorin Dr. Simone Reber; Professor Dr. Erik Rodner; Professorin Dr.-Ing. Amy Siu