CAREER: Scalable, General-Purpose, and Automated Secure Computing for AI Infrastructure

Modern cloud computing and artificial intelligence systems support scientific discovery, public services, economic activity, and security, but many valuable data sources are difficult to use because they contain private, regulated, proprietary, or security sensitive information. Current approaches often either expose data during computation through complex access control or protect data at a cost that limits efficiency, scalability, generality, or usability. This project aims to develop secure computing methods that allow artificial intelligence infrastructure to process protected data and models with reduced exposure during computation. The work is expected to serve the national interest by advancing the science of trustworthy computing, supporting safer use of sensitive data in data sensitive fields, and contributing to national health, prosperity, welfare, and security. The project will also support education and workforce development through course and laboratory modules, student mentoring, public tutorials, open-source software, engagement with practitioners, and outreach for students from kindergarten through twelfth grade. This project will study scalable, general purpose, and automated secure computing for artificial intelligence infrastructure. It addresses three challenges: improving the efficiency and scalability of protected and secure computation, supporting a broader range of protection needs and artificial intelligence functions, and making secure deployment easier for developers. The research will investigate scalable homomorphic computing and algebraic invariance methods for secure dense and sparse linear algebra, together with hardware adapted kernels for integer and modular arithmetic on commonly used computing accelerators. It will study methods to protect data, models, or both, and to support nonlinear functions through combinations of homomorphic computing and confidential computing when appropriate. It will also develop automated model level and data level transformations that map high level artificial intelligence workloads to secure implementations with less specialized cryptographic programming. The methods will be evaluated using comprehensive workloads and through cyberinfrastructure collaborations, with software, tutorials, and educational materials shared with research and practitioner communities. 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. NSF Award ID: 2544047 | Program: 01002930DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT,01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Qian Lou | Institution: The University of Central Florida Board of Trustees, ORLANDO, FL | Award Amount: $348,110 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2544047 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2544047.html