Events Calendar

PhD Dissertation Defense - Elan Markowitz

Title: Integrating Knowledge Graphs and Large Language Models to Improve Factuality and Reasoning
 
Date and Time: Thursday, May 8th, 2025 | 2:30p   Location: GCS 202C
 
Committee Members: Aram Galstyan, Greg Ver Steeg, Bistra Dilkina, Antonio Ortega
 
Abstract: Large Language Models (LLMs) have rapidly emerged as the dominant paradigm in AI due to their powerful understanding of unstructured text, strong reasoning abilities, and highly general task completion capabilities. However, they also have limitations in terms of how they use knowledge. They are black boxes with internal reasoning that is hard to analyze; they hallucinate incorrect facts as if they are true; and they suffer from knowledge cutoffs based on when their training ends. Knowledge graphs naturally complement these weaknesses through providing vast, structured, up-to-date, information over both general and specific domains.  At the same time, knowledge graphs have limitations, such as incompleteness and limited reasoning, that can be complemented by Large Language Models. Ultimately, through better integrating these approaches, we will deliver more reliable and trustworthy AI systems.
 
In this dissertation, I present a body of research on combining Large Language Models and Knowledge Graphs to address many of their individual weaknesses. This includes topics such as addressing knowledge graph incompleteness through combining language models and more structured graph neural networks; Integrating LLMs and external knowledge graphs with advanced reasoning capabilities; Measuring how presentation and other factors impact Large Language Models' understanding of in-context Knowledge Graphs; and using Knowledge Graphs to improve model editing approaches for updating an LLM’s internal knowledge. 

Location: Ginsburg Hall (GCS) - 202C

Audiences: Everyone Is Invited

Contact: Elan Markowitz

This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor.

Add to Google Calendar