Zachary Izzo

Artificial intelligence (AI) has become ubiquitous in recent years. Enabled by increasingly advanced machine learning (ML) models and algorithms, AI has found applications from autonomous vehicles to clinical settings.  In high-stakes scenarios such as these, it is critical to understand the risks and uncertainty associated with using a particular ML algorithm. For instance, what are some tests that a practitioner can use to be reasonably certain that his/her algorithm will have a high accuracy once deployed? To what degree can the uncertainty about the future accuracy be quantified? Subtler questions also arise, such as how to preserve the privacy rights of users whose data is used to train a model. My research aims to formulate and answer these questions in a way which is useful to both theoreticians and practitioners by synthesizing mathematics/computer science/statistics theory with knowledge of clinical and regulatory needs.