Lihu Chen - Exploring Factuality and Interpretability in LLMs: Confidence Estimation and Key Neuron Analysis
- Date: 16 décembre 2024 à 13h
- Salle: 65-66 304
Large Language Models (LLMs) possess vast amounts of knowledge within their parameters, which makes them highly effective in solving a wide range of tasks. However, they are susceptible to generating hallucinated answers in a confident tone. To address this, it is crucial to estimate the confidence levels of LLM outputs and gain a deeper understanding of the knowledge mechanism within LLM parameters. In this talk, I will present two recent worksaimed at tackling these challenges. First, I will introduce an approach to reconfidence LLM uncertainty from the grouping loss perspective. Second, I will discuss a method for identifying query-relevant neurons in Llama. We hope that our work can contribute to the development of more trustworthy and explainable LLMs.
Bio: Lihu Chen is a research associate at Imperial College London. His research primarily focuses on natural language processing (NLP) and large language models (LLMs). He is dedicated to developing efficient, reliable, and open-source models and tools, with a particular emphasis on information extraction and biomedical applications.