NSF POSE: Phase II: MLCommons Research for Science: Enabling Open-Source Ecosystems for Scientific Foundation Models by Community Standards and Benchmarks

This Pathways to Enable Open-Source Ecosystems (POSE) Phase II project aims to revolutionize machine learning (ML) for scientific discovery by establishing an open-source ecosystem (OSE) that focuses on identifying, developing, and applying foundation models for science. MLCommons, a non-profit consortium comprising over 60 corporations, universities, and government entities with more than 2000 members, will leverage its existing community and resources to create an OSE that facilitates access to, experimentation with, and modification of benchmarks, datasets, and models for machine learning research. The OSE will also offer courseware, training, and documentation, serving as a national beacon for linking industry, government, and academia to democratize artificial intelligence (AI) technologies. The project will enhance education and workforce opportunities by engaging a wide community of students, teachers, and underrepresented groups in AI creation. The OSE's foundation models will enable diverse users to construct custom AI systems without requiring high-level AI knowledge, meeting societal and national needs while driving scientific breakthroughs, improving public health outcomes, and contributing to the nation's economic growth and competitiveness. This Pathways to Enable Open-Source Ecosystems (POSE) Phase II project, MLCommons Research Open-source ecosystem (OSE), will advance the identification and development of foundation models for science, promoting scientific progress by facilitating the transfer of artificial intelligence (AI) models across scientific domains. The project will expand the community of scientific contributors through proactive outreach, enhancing intellectual access to applied AI technologies currently limited to large-scale private enterprises. To address the shortage of AI talent and the challenges of scaling AI across academic and industry verticals, the OSE will develop a new generation of open-source ML technologies, adhering to FAIR (Findability, Accessibility, Interoperability and Reusability) principles and identifying standard APIs (application programming interfaces) and ontologies needed to integrate multi-modal datasets into foundation models. The OSE materials will be searchable, and the expansion of MLCommons Research under this project result in more collaborators with broader skills across multiple industry verticals and academia, leading to a greater variety of data and model structures that will better challenge future AI systems and serve as a valuable resource for those developing AI models for new and emerging application domains. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.