Michael Johnson
On October 28, 2026, Meta announced $2 million in support for research at five U.S. universities. The form of assistance depends on the institution and project: it includes access to Meta models and technologies, collaboration with the company’s researchers, and funding for selected projects. The work spans medicine, physics, chemistry, biology, computer science, and mathematics.
The program includes the California Institute of Technology, The Ohio State University College of Medicine, Louisiana State University, Indiana University, and UTHealth Houston. Teams will use Muse models, including Muse Spark Open, Muse Wonder, Muse Spark, and Muse Glimmer, as well as the computer vision tools SAM, V-JEPA, and DINO. Meta says medical projects will run models within the institutions’ own infrastructure, keeping clinical and patient data under their control.
Clinical projects are being tested on real-world data
At The Ohio State University College of Medicine, a team is investigating how routine abdominal CT scans, longitudinal medical records, and available ECGs can be combined to assess future risk of heart failure with preserved ejection fraction. Other university projects focus on drafting radiology reports using multimodal data and comparisons with previous scans; evaluating quality metrics for those reports; and a separate study of how radiologists accept or reject AI recommendations in a simulated clinical setting. Another team is developing an agentic system that analyzes a patient’s cognitive history and helps clinicians choose the next informative steps when assessing Alzheimer’s disease.
At LSU, support covers five projects across four campuses: an inter-university registry of neurosurgical outcomes; Eyes-On, for monitoring infants with tracheostomy tubes; virtual reality training for clinicians; testing RF-based posture assessment systems that could be used for privacy-preserving fall detection; and causal AI to analyze outcomes of head and neck cancer treatment. Dr. Michael Dunham and Supratik Mukhopadhyay are involved in developing Eyes-On. According to LSU, the system is designed to track the tube’s position and warn of displacement; video is processed locally near the crib. Meta separately says a proof of concept for Eyes-On has been published and development is ongoing.
Indiana University is pursuing three areas. NeoSmartML+ explores multimodal tools for maternal and newborn care designed to run offline on low-cost Android devices. Another area studies how children learn words and visual concepts through everyday experience. The third focuses on adapting conversational agents to the communication patterns of people with dementia. At UTHealth Houston, researchers are studying local model deployment on the institution’s own GPU infrastructure, the performance of agent-based workflows in medical interfaces, and analysis of heterogeneous records for patients with glioblastoma. Meta describes UTHealth’s initial test on a sample of 65 cases as a workflow evaluation.
Scientific challenges, from mathematics to chemistry
At Caltech, researchers will test whether language models can help tackle open problems in mathematics and the physical sciences. One project is building a database of research-level mathematical problems and evaluating whether models can formalize conjectures, find lemmas, and produce machine-checkable proofs. Other areas include physical models, predicting chemical reactions from unpublished laboratory records, developing mathematical theories for biological systems, and creating formally verifiable scientific artifacts.
These projects offer different ways to evaluate results: clinical teams can compare assessments with medical data and specialists’ work, while scientific groups can compare them with experimental observations and formal proofs. In its program description, Meta outlines research plans and projects already underway; their practical value can be assessed through the methods and results teams publish as the work progresses.
Universities retain control of their data
According to Meta, universities conduct medical research independently in their own environments and retain control of clinical and patient data: the data are not transferred to Meta, processed by the company, or made available to it. This allows participants to investigate local applications of models to sensitive records, including medical images and text.
The program combines funding for selected projects with access to models and technical collaboration. Its results will depend on whether reproducible methods and published data demonstrate the value of models for specific research tasks—from clinical assessment and clinician training to mathematical proofs and scientific experiments.
