PI: Andres Marquez

Artificial intelligence (AI) enables new data analysis capabilities that accelerate society’s transition in all its forms from a first principles, rules-based foundation to a data driven foundation. The transition to machine learning (ML) brings great opportunities but also great risks. An inherent risk of data driven outcomes is the difficulty in explaining or deriving ML model and input data that is precise and trustworthy. While active research is underway to shed some light into black box AI modeling, the work proposed here seeks to attack the “lack-of-trust” problem using a systems view approach.

We intend to achieve this by investigating multi-granular, low-overhead, and provenance tracing infrastructures capable of fingerprinting data driven outcomes to not only perform online signature generation but also watermarking, integrity, and security checking.