Topic: From
Word Embeddings to Large Language Models: Evolution and Prospects
Speaker: Ying-Jia
Lin Ph.D. in Computer Science and Information Engineering form National Cheng
Kung University
Time: 2024/10/15 (Tue) 14:10-16:00
Venue: The
Management Building, 11F, AI Lecture Hall
Join Online: https://gqr.sh/NU8B
About the Speaker:
Dr. Ying-Jia Lin is a postdoctoral
researcher at National Tsing Hua University. He received his PhD from the
Department of Computer Science and Information Engineering at National Cheng
Kung University in 2024. Prior to that, he obtained his MS from the Institute
of Biomedical Informatics at National Yang-Ming University in 2019 and his BS
in Biomedical Sciences from Chang Gung University in 2017. His current research
focuses on text summarization, model compression, and BioNLP. Ying-Jia Lin has
published in top AI/NLP conferences, such as AAAI, EMNLP, and AACL. He is an
honorary member of the Phi Tau Phi Society, and he won two Best Paper Awards at
TAAI in 2022 and 2019.
Abstract:
This presentation explores the evolution of
Natural Language Processing (NLP) from the foundational concept of word
embeddings to the emergence of large-scale language models like GPT. In the
first part, we will journey through the history of NLP, highlighting key
developments that have led to the current state of the field. The second part
critically examines whether GPT has really solved the challenges of Natural
Language Generation, using text summarization as a case study. We will discuss
architectural issues inherent in GPT models, such as those related to the
Key-Value (KV) cache, and examine knowledge limitations, particularly in the
application of GPT to medical text reports. The role of Retrieval-Augmented
Generation (RAG) in addressing these challenges will also be explored. This
talk aims to provide insights into the advancements and remaining hurdles in
NLP, offering perspectives on future directions and prospects.
Organizers: College
of Intelligent Computing & Artificial Intelligence Research Center
※ No registration needed.