Interpretability and Adaptation of Language Models

Prof. Nakashole will present this lecture remotely; the link to join is https://umich.zoom.us/j/96905725744. Local attendees may attend in 3725 Beyster Building.

Abstract: The first part of the talk will focus on interpretability of LLMs on structured data processing tasks.
Due to their ability to compute rich representations of natural language sequences, large language models (LLMs) are now commonly used to process structured data found in formats such as databases, and knowledge graphs. The structured data is first transformed into sequential token streams, ignoring its inherent non-linear properties. Our research aims to shed light on how linearization impacts the interpretation and handling of structured data in language models.

The second part of the talk will discuss our ongoing efforts to adapt language models to new linguistic contexts, focusing on a Bantu language.

Bio: Ndapa is an Associate Professor of Computer Science at the University of California, San Diego. Her research is on Natural Language Processing (NLP).

Before UCSD, she was a postdoctoral fellow in the Machine Learning department at Carnegie Mellon University, where she worked with Tom Mitchell. She obtained her PhD from the Max Planck Institute for Informatics, and Saarland University, where her advisor was Gerhard Weikum. Her dissertation was awarded the Otto Hahn Medal by the Max Planck Society. Her work has been awarded an NSF CAREER award.