M3 Workshop Advances Federated AI for Biomedical Research
The Center for Translational Data Science (CTDS) and the Data Science Institute jointly hosted the Meshes of Midscale Models (M3) Workshop on January 28, bringing together researchers working at the frontier of AI in biomedical research.
A fundamental obstacle to using AI in biomedicine is the scarcity of high-quality data to train large language models, generative AI, and large quantitative models. The Meshes for Midscale Models (M3) Initiative takes a multidisciplinary approach to the challenge, integrating expertise in biology, medicine, computer science, machine learning, and economics. Rather than relying on massive proprietary models, it focuses on developing small and midscale AI models that can operate effectively in resource-limited environments. The approach is already showing results: earlier M3 work demonstrated that pretrained machine learning models can help diagnose skin cancers in settings where specialist dermatologists are scarce.
The morning opened with welcome remarks from Conrad Gilliam, Dean for Basic Science in the Biological Sciences Division, followed by sessions updating attendees on the Initiative’s current areas of research. Robert Grossman (Frederick H. Rawson Distinguished Service Professor of Medicine and Computer Science and Jim and Karen Frank Director of the Center for Translational Data Science) opened with an overview of AI Commons and AI Meshes. Additional presentations ranged from foundational research—Yuxin Chen on how data quality affects model pretraining, Tian Li on distributed and federated models—to applied tools like Steven Song‘s work on cancer modeling and Raul Castro Fernandez‘s exploration of incentives for data sharing.
There were also talks on AI tools like Query Augmented Generation Cohort Copilot (Aarti Venkat), and Gen3.2 AI Commons (Craig Barnes and Michael Lukowski). Afternoon discussions focused on AI Commons applications and infrastructure, including the Head and Neck Cancer Bank (Alex Pearson, MD, PhD and Sara Kochanny) and AI Commons to support applications for Hospital Medicine (David Meltzer, MD, PhD) and Emergency Medicine (David Beiser, MD). Breakout sessions gave attendees an opportunity to dive deeper into building AI Commons and associated models for hospital and emergency medicine applications.
“The impact of biomedical AI will grow substantially if we solve the problem of getting AI models the biomedical data they need at the scale they need,” said Robert Grossman. “M3 is working to make it easier to build, share, and connect AI models across institutions. We’re proving that with the right tools and data commons infrastructure, researchers everywhere can build models tailored to their specific needs.”
As AI plays a growing role in biomedical research, the M3 Initiative is working to ensure that powerful tools remain accessible and privacy-preserving—meeting the needs of researchers at major medical centers and low-resource settings alike. Researchers interested in the Initiative work can learn more at https://m3initiative.uchicago.edu/.
This article was originally published on the Data Science Institute website.
