EVENT

Talk on "LLM Analysis, and Enhancement from a Probabilistic Standpoint" by Dr. WANG Xun

We are pleased to inform you about the upcoming seminar by Dr. WANG Xun titled : "LLM Analysis, and Enhancement from a Probabilistic Standpoint" Everyone interested is cordially invited to attend!

Title:

LLM Analysis, and Enhancement from a Probabilistic Standpoint

Abstract:

Large-scale language models have garnered attention due to their near-human level capabilities in understanding and generating language.
We conduct an analysis of these models to explore the potential limits of their language processing abilities. Considering the inherent properties of language models, we explore several improvement strategies, such as multi-agent systems and retrieval-augmented generation, to enhance their performance in real-world applications.
Nevertheless, challenges remain in processing long texts, avoiding hallucination effects, and reducing over-inference.
These challenges reveal the limitations of data-driven probabilistic models and delineate the current boundaries of large-scale language model capabilities. By comparing the models with human linguistic abilities, we explore potential directions for improving model structures and further understand the nature of language as a complex system.

Speaker Bio:

Dr. WANG, Xun received his B.A. in Applied Linguistics, B.S. in Statistics, M.S. in Computer Science from Peking University and PhD in Informatics from Kyoto University in 2010, 2010, 2013, and 2017, respectively. He was affiliated with NTT Communication Science Lab, Umass Amherst/Lowell and Veterans Affairs.
His research primarily focuses on discourse analysis, LLM analysis, and their practical applications.

Time/Date:

11:00 - 12:00 July 2 (Tuesday), 2024

Place:

Room 1208/1210 NII

Contact:

Please register from the link below.
https://forms.office.com/r/CFXZhwJFZC