DFG Excellence: "Berlin Initiative for Research in the Area of Foundation Models” (2024-2029)
The DFG research impulse “Berlin Initiative for Research in the Area of Foundation Models”, aims of a better understanding of basic science and the practical optimization of machine learning systems, so called foundation models, in the fields of robotics, quantitative biology and Predictive medicine with the goal to better meet the challenges caused by demographic change in the world of work as well as in the diagnosis of diseases and medical care for the population. DATEXIS works on foundation models for medicine and improved textual patient representations.
BMBF Efficient language models for SMEs (More-With-Less) 2023-2026
Large language models are insufficiently adapted to the needs of German SMEs. The goal of the project is to develop methods and a framework for the efficient adaptation of language models for SME-specific applications. In the BHT subproject, data- and model-centric approaches will be explored and prototypically tested for three use cases: (1) automated text authoring for the finance industry, (2) document matching for legal tech companies, and (3) improved reading order recognition in legal and finance documents.
Partners: Merantix Labs GmbH, 2txt NLG GmbH, Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. (IAIS)
BMBF Machine Reading in Supply Chain Management KMUs (AI4SCM) 2023-2026
The goal of the collaborative project is to explore methods for detecting even rare SCM events in text-based sources and in the context of supply chain due diligence. Our methods require little training data from the SME and can be operated via dashboards. An important goal is highly customizable learning methods with few training data for niche domains. These include methods of Few-Shot, Transfer Learning, Multi-Task Learning, Prototypical Learning, and Meta Learning. The latter, in particular, also enable explainability. The basis of the risk indicators for this project is compliant with the Supply Chain Due Diligence Act, which comes into effect on Jan. 1, 2023, and must also be implemented by SMEs. These indicators include the holistic view of the company, such as economic relationships between suppliers and partners.
Partners: Neofonie GmbH, Ubermetrics Technologies GmbH
EU COMFORT: COMputational Models FOR patienTstratification in urologic cancers – Creating robust and trustworthy multimodal AI for health care (COMFORT) 2023 -2027
In Europe, more than two million men are living with prostate cancer (PCa), and about 80,000 die from it each year. To date, there are no accurate, robust, transparent and ready-to-use multimodal AI tools for PCa and KC and no standard-of-care options for non-invasive individual patient stratification according to prognosis and therapy response at the time of initial diagnosis. Current standard diagnostic procedures for PCa have led to overdiagnosis and overtreatment, while at the same time resulting in under- or missed diagnosis. For low-grade PCa, e.g., aggressive therapy can result in long-term side effects, such as sexual dysfunction or incontinence.
COMFORT sets out to develop and prospectively validate data-driven decision support tools based on cutting edge AI models, to improve detection, diagnosis and prognosis of patients with PCa and KC.
Partners: Charité Berlin - Department of Radiology and Comprehensive Cancer Center, Berlin, DE, University of Naples "Federico II" - Department of Radiology, IT, Technical University Munich - Department of Radiology and Institute for Artificial Intelligence in Healthcare and Medicine, DE. General University Hospital of Patras - Department of Urology, GR, School of Medicine of the Aristotle University of Thessaloniki (A.U.Th) - 1st Department of Urology, Laboratory of Medical Physics and Digital Innovation, GR, Servicio Madrileno de Salud, ES Hospital 12 de Octubre, Departament of Medical Oncology, ES, Berlin University of Applied Sciences and Technology (BHT) - Data Science +X Research Center, DE, Radboud University Medical Center Nijmegen (Radboudumc), NL. Quantib B.V., NL. Phönix PACS GmbH - PACS provider, DE. European Research and Project Office GmbH, DE. Umeå University - Responsible Artificial Intelligence, SE. European Cancer Patient Coalition, BE. MeVis Medical Solutions AG, DE
