Mohammad Mahmoody

Assistant Professor

University of Virginia

Title:

Watermarking LLM-generated texts

Time: Thursday, May 2nd 2024, 09:15 – 10:00

 

Abstract:

With the popularity and extensive uses of large language models (LLMs) watermarking the text generated by LLMs has become an important problem. In this talk, I will talk about the theoretical foundations and a formalism of this problem, and then I will discuss information-theoretic, private-key and public-key approaches to this task.

Biography:

Mohammad Mahmoody obtained his undergraduate degree from Sharif’s computer engineering department in 2004. He then got his PhD from Princeton’s computer science department in 2010. He joined Cornell as a postdoc associate and then University of Virginia as an assistant professor in 2013. His research interests are in foundations of theory of cryptography and trustworthy machine learning.