BIOINFORMATICS SEMINAR SERIES
https://bioinformatics.udel.edu/seminar
Toward Improving Fairness in Health LLMs
Rahmat Beheshti, PhD
Assistant Professor
Department of Computer and Information Sciences
University of Delaware
Abstract:
The surge in the adoption of large language models (LLMs) across various sectors of society has brought much excitement, fear, and uncertainty about their impact. In healthcare, various units have started adopting LLMs in the hope of finding long-sought solutions to challenges such as provider burnout and processing of claims or preauthorizations. One significant uncertainty regarding LLM applications in high-stakes applications like healthcare (Health LLMs) is their impact on the fair treatment of individuals and shifting healthcare disparities. While there is a growing body of work aiming to evaluate fairness concerns in Health LLMs, structured and scalable methods to identify the scope of such fairness risks or to offer ways to mitigate those in Health LLMs are not available. In this talk, I will further discuss the significance of developing such methods and several of our recent projects in this area. In particular, I will cover a method for scalable evaluation of health LLMs and another one for effective alignment of those.
Short Bio:
Rahmat Beheshti is an Assistant Professor in the Department of Computer and Information Sciences at the University of Delaware (UD). He has a joint appointment with the Epidemiology Program and is a resident faculty member of the Data Science Institute at UD. He leads the “healthy lAife” research lab.
