Events Calendar

PhD Dissertation Defense - Jun Yan

Title: Identifying and Mitigating Safety Risks in Language ModelsAbstract: Recent advancements in language models have revolutionized the field of Natural Language Processing, reshaping human-technology interactions. As these models become increasingly integrated in our daily lives, concerns about their safety risks have also escalated. In this thesis defense, I will present my work on identifying and mitigating safety risks in language models that could lead to system malfunctions and undermine user trust. My research addresses three key questions: (1) What threats can adversaries induce by poisoning the training data of language model classifiers? (2) Can practitioners reliably detect compromised language model classifiers before deployment? (3) What novel threats does data poisoning pose with the emergence of generative large language models? In conclusion, I will discuss future directions for the development of safer language models.
 
Committee Members: Prof. Xiang Ren (Chair), Prof. Robin Jia, and Prof. Morteza Dehghani  
 
Date: Monday, July 8th, 2024Time: 12pm – 2pm  
 
Location: Ronald Tutor Hall of Engineering (RTH) - 306  
 
Zoom Link: https://usc.zoom.us/j/6633659669  

Location: Ronald Tutor Hall of Engineering (RTH) - 306

Audiences: Everyone Is Invited

Contact: Ellecia Williams