A Priliminary Study of LoRA Experts for Personalizaed Clinical Summarization

Presentation Time: 05:30 PM - 07:00 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Artificial Intelligence, User-centered Design Methods
Primary Track: Applications

Clinicians document and summarize critical patient information in Electronic Health Records (EHRs), but existing summarization models fail to address specialty-specific needs. Single-task models lack scalability, while multi-task models suffer from task interference. To improve personalized clinical summarization, we propose a novel approach leveraging the Llama3.1 model with Mixture of Low-Rank Adaptation (LoRA) Experts. This method enhances accuracy and flexibility, enabling tailored summaries for different specialties without compromising performance or manageability.

Speaker(s):
Mengxian Lyu, Master
University of Florida

Author(s):
Mengxian Lyu, Master - University of Florida; Yonghui Wu, PhD - University of Florida;