Meta Supports AI Research at Five Universities

Meta says it is supporting research at five U.S. universities through access to its AI models and technologies, collaboration with Meta researchers, and funding for selected projects. The institutions are California Institute of Technology, The Ohio State University College of Medicine, Louisiana State University, Indiana University and UTHealth Houston. The support varies by institution and project. Meta says university faculty chose their research questions, designed the studies and decided how to use the tools. For healthcare projects, universities will run Meta models in their own environments and retain control of patient and clinical data; Meta says that data will not be shared with or made available to the company.

The projects will use Meta's Muse Spark Open, Muse Wonder, Muse Spark and Muse Glimmer models, alongside computer vision systems Segment Anything Model (SAM), V-JEPA and DINO, SeamlessM4T, and some wearable technologies. Meta describes Muse Spark Open as its most capable open-weight model and Muse Glimmer as a 30B open-weight model that can run on a single consumer GPU. Meta says the work spans medicine, physics, chemistry, biology, computer science and mathematics, including medical training, patient outcomes and research in low-resource environments.

At Caltech, researchers will test whether language models can support original work in five scientific disciplines. Projects include MathDB Frontier, a community-curated database and benchmark built on MathDB's more than 86,900 research-level open problems; a constrained method for building physical representations; a chemistry benchmark based on unpublished electronic lab notebooks; automated construction and testing of mathematical models for biological active matter; and TorchLean, a Lean 4 framework for producing formally checkable scientific and machine-learning artifacts. Researchers will access Meta's Muse models through its first-party API.

Ohio State projects will investigate using abdominal CT scans, medical records and, when available, ECG data to identify risk of heart failure with preserved ejection fraction before diagnosis. Other work will draft chest X-ray reports using multiple sources of patient information, correct bias in measures used to grade model reports, and study when radiologists accept or reject AI drafts. A clinician-guided multi-agent system will analyze cognitive histories and clinical evidence for Alzheimer's diagnosis, identify uncertainty and suggest next clinical actions. Ohio State's Memory Disorders Clinic sees over 5,600 patient visits a year; an estimated 7.4 million Americans aged 65 and older have Alzheimer's disease and related dementias.

Louisiana State University projects include a multi-institution neurosurgical outcomes registry curated from clinical text, with SAM and DINO used to process medical images and Muse models running on institutional infrastructure. Another team is developing a computer vision system to detect displaced tracheostomy tubes and signs of distress in infants; a proof-of-concept has been published. A virtual reality medical training prototype uses Meta open-weight models and retrieval-augmented generation (retrieving relevant information to ground generated responses) for hands-free coaching, with plans to add SeamlessM4T for nearly 100 languages. A separate project will validate camera-free radio-frequency pose estimation using skeletal labels generated from synchronized RGB images, with the aim of avoiding motion-capture equipment. The supplied article text ends partway through its description of an LSU project on causal AI for head and neck cancer.

Meta says it is supporting research at five U.S. universities through access to its AI models and technologies, collaboration with Meta researchers, and funding for selected projects. The institutions are California Institute of Technology, The Ohio State University College of Medicine, Louisiana State University, Indiana University and UTHealth Houston. The support varies by institution and project. Meta says university faculty chose their research questions, designed the studies and decided how to use the tools. For healthcare projects, universities will run Meta models in their own environments and retain control of patient and clinical data; Meta says that data will not be shared with or made available to the company.

The projects will use Meta's Muse Spark Open, Muse Wonder, Muse Spark and Muse Glimmer models, alongside computer vision systems Segment Anything Model (SAM), V-JEPA and DINO, SeamlessM4T, and some wearable technologies. Meta describes Muse Spark Open as its most capable open-weight model and Muse Glimmer as a 30B open-weight model that can run on a single consumer GPU. Meta says the work spans medicine, physics, chemistry, biology, computer science and mathematics, including medical training, patient outcomes and research in low-resource environments.

At Caltech, researchers will test whether language models can support original work in five scientific disciplines. Projects include MathDB Frontier, a community-curated database and benchmark built on MathDB's more than 86,900 research-level open problems; a constrained method for building physical representations; a chemistry benchmark based on unpublished electronic lab notebooks; automated construction and testing of mathematical models for biological active matter; and TorchLean, a Lean 4 framework for producing formally checkable scientific and machine-learning artifacts. Researchers will access Meta's Muse models through its first-party API.

Ohio State projects will investigate using abdominal CT scans, medical records and, when available, ECG data to identify risk of heart failure with preserved ejection fraction before diagnosis. Other work will draft chest X-ray reports using multiple sources of patient information, correct bias in measures used to grade model reports, and study when radiologists accept or reject AI drafts. A clinician-guided multi-agent system will analyze cognitive histories and clinical evidence for Alzheimer's diagnosis, identify uncertainty and suggest next clinical actions. Ohio State's Memory Disorders Clinic sees over 5,600 patient visits a year; an estimated 7.4 million Americans aged 65 and older have Alzheimer's disease and related dementias.

Louisiana State University projects include a multi-institution neurosurgical outcomes registry curated from clinical text, with SAM and DINO used to process medical images and Muse models running on institutional infrastructure. Another team is developing a computer vision system to detect displaced tracheostomy tubes and signs of distress in infants; a proof-of-concept has been published. A virtual reality medical training prototype uses Meta open-weight models and retrieval-augmented generation (retrieving relevant information to ground generated responses) for hands-free coaching, with plans to add SeamlessM4T for nearly 100 languages. A separate project will validate camera-free radio-frequency pose estimation using skeletal labels generated from synchronized RGB images, with the aim of avoiding motion-capture equipment. The supplied article text ends partway through its description of an LSU project on causal AI for head and neck cancer.