Summary
This NSF CAREER project aims to develop the mathematical tools required to build robust Artificial Intelligence (AI) systems that remain reliable in unpredictable, real-world conditions. While current AI models are highly effective in controlled settings, they can be fragile when faced with unexpected data shifts or adversarial attacks where data is geometrically manipulated to cause errors. The project will advance the state-of-the-arts by shifting from defensive methods that only patch specific vulnerabilities to a more universal approach that provides transferable certificates of robustness
