Speakers

Prof. Chengqing Zong

IEEE Fellow, ACL Fellow and CAAI Fellow

Research fellow of Institute of Automation, Chinese Academy of Sciences (CAS)

Chengqing Zong is a Research Fellow (Professor) of the Institute of Automation, Chinese Academy of Sciences. He received his Ph.D. degree from the Institute of Computing Technology, CAS, in March 1998. From May 1998 to April 2000, he was a postdoctoral research fellow at the Institute of Automation, CAS, and he has been working at the institute since completing his postdoctoral program. In 1999 and 2001, he worked at the Advanced Telecommunications Research Institute (ATR) in Japan as a guest researcher. In 2004, he visited CLIPS-IMAG in France as a visiting scholar.
He is a Fellow of the IEEE, a Fellow of the Association for Computational Linguistics (ACL), a Fellow of the Chinese Association for Artificial Intelligence (CAAI), and a Fellow of the China Computer Federation (CCF). His research interests include NLP, MT, and cognitive language computing etc. He has published over 200 papers in top-tier conferences and journals. He has authored and co-authored six books and translated two books in cooperation with his colleagues.
This year he is serving as the President of ACL. He served as the General Chair of ACL 2021, a top-tier conference. He has received numerous awards, including the Second Prize of the National Scientific and Technological Progress Award of China in 2015.
 

Abstract: In recent years, Large Language Models (LLMs) such as ChatGPT and Deepseek have developed rapidly and become international technology hotspots. They have had a great impact on artificial intelligence research and development, and even on society as a whole. However, how do these LLMs perform in practical tasks, such as machine translation (MT)? How will natural language processing technologies, including MT, develop in the future? This talk will present a performance analysis of ChatGPT and Deepseek in MT, then introduce some research work on MT conducted by speaker’s research group, and finally provide the speaker’s understanding and outlook on the future technological development of LLMs and MT.

Prof. Haoran XIE

World's Top 2% Scientists in (1) Artificial Intelligence and (2) Education, Stanford University

Lingnan University, Hong Kong, China

Prof. XIE Haoran received a Ph.D. degree in Computer Science from City University of Hong Kong and an Ed.D degree in Digital Learning from the University of Bristol. He is currently a Professor and the Person-in-Charge of Division of Artificial Intelligence, Acting Associate Dean of the School of Data Science, and Director of LEO Dr David P. Chan Institute of Data Science, Lingnan University, Hong Kong. His research interests include natural language processing, computational linguistics, artificial intelligence in education, and educational technology. He has published 438 research publications, including 262 journal articles. His Google Scholar citation count is 22654, with an h-index of 61 and an i10-index of 206. He is the Editor-in-Chief of Natural Language Processing Journal, Computers & Education: Artificial Intelligence, and Computers & Education: X Reality, and the Co-Editor-in-Chief of Knowledge Management and E-Learning. He has been selected as the World's Top 2% Scientists by Stanford University.

Abstract: Aspect-Based Sentiment Analysis (ABSA) poses challenges in structured sentiment prediction, particularly in handling implicit expressions and multi-element relationships. This talk presents three key approaches leveraging Large Language Models (LLMs) to enhance sentiment analysis: (1) RVISA, a reasoning-verification framework combining Decoder-Only and Encoder-Decoder LLMs for ABSA; (2) MT-ISA, a Multi-Task Learning framework with automatic weight learning for adaptive sentiment inference; and (3) STAR, a stepwise task augmentation framework for Aspect Sentiment Quad Prediction. By leveraging LLMs for reasoning, verification, and adaptive learning, we can enhance the explainability, robustness, and accuracy of sentiment analysis tasks across diverse datasets. The talk will discuss model architectures, experimental findings, and future research directions in explainable AI for sentiment analysis.

Prof. Xiaojun Quan

School of Computer Science and Engineering

Sun Yat-sen University

Xiaojun Quan is a professor at School of Computer Science and Engineering, Sun Yat-sen University (SYSU) since July 2017. Before joining SYSU, he was a research scientist in the data analytics department (DAD) at Institute for Infocomm Research (I2R), A*STAR, since January 2014. He was also a postdoctoral fellow in the Department of Chinese, Translation and Linguistics, City University of Hong Kong, from Sep 2012 to Dec 2013. He received his Ph.D. in Computer Science in 2012 from City University of Hong Kong, where he was among the several recipients of the School of Graduate Studies Scholarship, which is regarded as the highest honor for research students. He was a visiting scholar at Purdue University in 2011 collaborated with Prof. Luo Si and at Rutgers, The State University of New Jersey in 2010 collaborated with Prof. Hui Xiong. He received his master degree in computer science from University of Science and Technology of China in 2008.

Abstract: Alignment techniques are essential for enabling large language models (LLMs) to generate outputs that align with human preferences. However, their effectiveness often diminishes when applied to weak language models (WLMs), likely due to the limited capacity of these models. Instead of directly applying existing alignment techniques to WLMs, we propose leveraging well-aligned strong LLMs to guide the alignment process, facilitating the transfer of human preference knowledge to weaker models. In this talk, we will introduce our work on heterogeneous model fusion, a technique designed to transfer knowledge and capabilities from structurally different but well-aligned teacher LLMs to weaker student models. We will also present our series of projects, including FuseLLM and FuseChat 1.0/2.0/3.0, which aim to enhance alignment in WLMs through this approach.



Prof. Yan Song

University of Science and Technology of China

Dr. Song Yan is a professor at the School of Information Science and Technology at the University of Science and Technology of China and an affiliate professor at the University of Washington. His research focuses on artificial intelligence, encompassing natural language processing, information retrieval and extraction, and text representation learning, among others. Dr. Song has published over 100 papers in top-tier international journals and conferences on artificial intelligence. His work has been frequently featured in authoritative AI publications and conferences, such as the international journal Computational Linguistics and its annual meeting (ACL), the annual conference of the Association for the Advancement of Artificial Intelligence (AAAI), the Conference on Empirical Methods in Natural Language Processing (EMNLP), and the International Joint Conference on Artificial Intelligence (IJCAI), to name a few. Dr. Song has also served as a program committee member and senior area chair for leading AI conferences over the years. Previously, Dr. Song held positions as a researcher at Microsoft, a chief researcher at Tencent, the executive dean of the Innovation Works Greater Bay Area Research Institute, and an associate professor at The Chinese University of Hong Kong (Shenzhen). He was one of the founding team members of the "Microsoft Xiaoice" project.

Abstract:  As a medical multimodal information processing task, Radiology Report Generation (RRG) aims to automatically produce accurate, coherent diagnostic reports from radiographs, which significantly reduce radiologists' workloads, enhance diagnostic accuracy, and improve clinical workflows. Research for RRG witnesses the rapid development in the area of natural language processing from different aspects, so that enjoys the upgrade of methodologies from model, knowledge, and evaluation perspectives, while facing different challenges as well. In this talk, we present RRG research advances in recent years, focusing on progress in modeling techniques, knowledge integration, and evaluation frameworks, respectively, by emphasizing paradigm change from small models of encoder-decoder architecture to large language models via decoder-only process, from feature engineering on shallow patterns to knowledge integration of pathology information, from language generation evaluation on linguistic fluency to comprehensive assessment covering clinical correctness and explainability. In tackling the challenges that exist in the aforementioned perspectives in RRG development, this talk also introduces our representative work to advance model architectures, knowledge integration, evaluation metrics, respectively.