Federal Grant Award Summary

Cornell University received a $670,000 National Science Foundation (NSF) Engineering program (CFDA 47.041) CAREER grant awarded April 1, 2026, with completion targeted for March 31, 2031. This five-year project, performed in Ithaca, New York, aims to develop mathematical frameworks and computational tools for building robust artificial intelligence (AI) systems capable of maintaining reliability under real-world conditions including data distribution shifts and adversarial attacks. The research leverages optimal transport theory to construct ambiguity sets around data manifolds learned by generative models, framing the interaction between a learner and generative adversary as a game-theoretic problem to resolve trade-offs between computational tractability and certified robustness in distributionally robust optimization.

Primary deliverables include the development and open-source release of a software library called "SKDRO" that implements the project's mathematical innovations in robust learning methodology. The award also supports creation of educational materials for students and establishes new mathematical connections between data geometry and robust learning. By enabling AI models to transfer safety guarantees across different deployment environments rather than relying on defensive patches for specific vulnerabilities, this project contributes to more reliable AI technologies supporting national prosperity and security objectives aligned with NSF's emphasis on innovation and workforce development in engineering research.