Mathematical Foundations of Artificial Intelligence

The National Science Foundation’s Mathematical Foundations of Artificial Intelligence (MFAI) program supports interdisciplinary research aimed at deepening the theoretical and mathematical understanding of artificial intelligence. Jointly sponsored by multiple NSF directorates—the Mathematical and Physical Sciences, Computer and Information Science and Engineering, Engineering, and Social, Behavioral and Economic Sciences—the program seeks to strengthen the mathematical principles underlying AI and machine learning systems. It builds on earlier NSF investments such as MoDL, SCALE MoDL, TRIPODS, and the Foundations of Machine Learning programs, which highlighted the need for rigorous frameworks to ensure reliable, interpretable, and trustworthy AI. The MFAI program invites proposals that focus on developing novel mathematical and theoretical approaches to understanding the capabilities, limitations, and emergent properties of modern AI methods. Areas of interest include, but are not limited to, deep learning, foundation models, statistical learning, federated learning, generative models, topology, complexity theory, algebraic geometry, dynamical systems, partial differential equations, and optimization theory. The intent is to uncover fundamental principles that could lead to explainable, transferable, and generalizable AI systems that advance knowledge and benefit society. Projects should propose creative, mathematically grounded research while ensuring practical relevance to the design and analysis of future AI systems. Applicants must submit full proposals through either Research.gov or Grants.gov, following NSF’s Proposal and Award Policies and Procedures Guide (PAPPG). No letters of intent or preliminary proposals are required. Each proposal should include a Collaboration Plan describing how team members from mathematics, computer science, engineering, and social sciences will coordinate research and share results. Principal Investigators (PIs) and co-PIs are limited to one submission per deadline. Successful applicants are expected to participate in annual PI meetings to exchange findings and build community connections across disciplines. Up to 15 awards are anticipated per annual competition, with total funding of approximately $8.5 million per year. Individual awards typically range between $500,000 and $1.5 million for projects lasting 36 months. Cost sharing is prohibited. Awards will be issued as standard or continuing grants depending on project scope and review outcomes. Standard NSF award conditions and reporting requirements apply, including annual project reports and a final public project outcomes report. Eligibility is limited to accredited U.S. institutions of higher education and qualified non-profit, non-academic research organizations directly associated with education or research activities. PIs must hold a tenured, tenure-track, or full-time paid research or teaching appointment at an eligible U.S. institution. For-profit organizations and individuals are not eligible to serve as PIs. Proposals will be evaluated using NSF’s two standard merit review criteria—Intellectual Merit and Broader Impacts—with additional attention to the quality of the mathematical innovation and collaboration plan. Applications are due annually by October 10 for 2024 and 2025 cycles, and October 9 for 2026. Awards are expected to begin the following calendar year. General inquiries can be directed to the program team at mfai@nsf.gov or (703) 292-5111. The full solicitation, NSF 24-569, is available on the NSF website.