Sharon Li

• Program

  AI Safety Science

• Institution

  University of Wisconsin-Madison

• Location

  USA

Dr. Sharon Li is an Assistant Professor at the Department of Computer Sciences at the University of Wisconsin–Madison, involved with Machine Learning@UW-Madison and a faculty affiliate at the Data Science Institute. Her previous experience includes a postdoctoral role at Stanford University’s Computer Science department, collaborating with Christopher Ré, and a PhD from Cornell University under John E. Hopcroft. Her research focuses on the algorithmic and theoretical foundations of safe and reliable AI, particularly in handling out-of-distribution data, quantifying uncertainties, and understanding foundational model mechanisms. Sharon and her team are currently developing responsible foundation models, striving to align these with human needs and values effectively.

Her funded project aims to establish a new framework for providing formal confidence guarantees for Large Language Model (LLM) outputs, utilizing conformal prediction theory. Traditionally effective in simpler contexts, applying conformal prediction to generative LLMs introduces challenges due to their sequential nature and expansive output possibilities.