Abstract: Visualization technology transforms data into graphics, allowing people to intuitively perceive patterns and characteristics of big data. Traditional visualization techniques enable users to explore data from different angles through human-computer interaction, but this process often requires significant human effort and lacks exploration direction. Large model-driven visual analytics offers a new paradigm of human-intelligence collaboration, exploring methods for intelligent data analysis. In this process, we use visualization as a bridge for human-intelligence collaboration, enabling people to perceive data and convey their ideas to the model through interaction. We will share methods for large model-driven visual storytelling and data analysis, exploring the excavation and transmission of intelligent data insights.

Speaker Bio: Young researcher and doctoral supervisor at the School of Data Science, Fudan University, and head of the Fudan University Visual Analytics and Intelligent Decision Laboratory (FDUVIS). He is a high-level talent introduced by Shanghai and a Fudan Zhongying Scholar. Previously, he served as a research scientist at the Fraunhofer Institute for Intelligent Analysis and Information Systems (Fraunhofer IAIS) in Germany and a postdoctoral researcher at the University of Bonn. He holds a bachelor's degree from Fudan University (2011) and a PhD from Peking University (2017). His research focuses on big data visualization and visual analytics, interactive artificial intelligence, with key areas including AI+VIS, large model-driven visual analytics, social media analysis, autonomous driving, fintech, and digital twins. He has published over 100 papers, including more than 40 articles in top international human-computer interaction conferences and journals such as IEEE VIS, IEEE TVCG, ACM CHI, CSCW, and UIST. He has been nominated for the AI2000 Ten-Year Most Influential Visualization Research Award (top 100 globally). He has led and participated in over ten national and provincial-level projects. He serves as an international program committee member for IEEE VIS, associate editor for the IEEE CG&A journal, youth editorial board member for the Visual Informatics journal, chair of the IEEE PacificVis paper (VizNotes), chair of the ChinaVis Data Analysis Challenge, and co-chair of the VGI Geovisual Analytics Workshop on geographic and spatiotemporal visual analytics. For more information, please visit http://fduvis.net.

Abstract: Against the backdrop of rapid development in large language models (LLMs), understanding, optimizing, and utilizing these models has become a crucial research topic in artificial intelligence and visualization. This talk explores the role of visual analytics in trustworthy LLM research. Taking LLMs as the research object, we examine how visual analytics can support dataset construction, training process optimization, and model interpretability analysis, addressing their massive parameters, complex structures, and high-dimensional data generated during training and inference. Through systematic frameworks, technical innovations, and case studies, this talk aims to demonstrate the increasingly close integration between LLMs and visual analytics, inspiring new paradigms for future human-computer collaborative analysis.

Speaker Bio: Dazhen Deng is a "Hundred Talents Program" researcher at the School of Software, Zhejiang University, and a Qizhen Outstanding Young Scholar. His main research focuses on large language models and visual analytics, conducting series of studies on LLM-driven visual analytics, LLM security visual analytics, and applications of visual analytics in sports, urban studies, and other fields. He has published over 20 papers in CCF-A conferences/journals including IEEE VIS, IEEE TVCG, ACM KDD, ACM CHI, and UIST. He received Best Paper Nominations at IEEE VIS 2022 and 2024, and the First Prize of Zhejiang Province Science and Technology Progress Award in 2023. He serves as a reviewer for authoritative conferences such as IEEE VIS and ACM CHI, and as a program committee member for PacificVis TVCG Journal Track. He leads/participates in multiple national projects including National Natural Science Foundation and National Key R&D Program, as well as various provincial and municipal projects. His research has been successfully applied to China's table tennis big data analysis platform, supporting the national team in preparing for major international competitions with significant social benefits.

Abstract: As AI prediction models demonstrate strong knowledge discovery potential in scientific research, understanding their reasoning process through visual analytics and screening the most experimentally valuable predictions has become a core challenge in 'AI for Science' practices. This talk focuses on complex biological mechanism prediction tasks, proposing and iteratively building two logically connected visual analytics systems to enhance prediction interpretability and efficient result screening under human-machine collaboration. The first system focuses on analyzing and validating model reasoning paths, helping experts identify potentially overlooked biological mechanism predictions based on known mechanisms, leveraging visualization to enhance understanding and support mechanism reasoning. The second system introduces a 'hypothesis-driven' scientific discovery concept, integrating large language models' knowledge retrieval and reasoning capabilities with structured knowledge graphs to assist users in proposing, interpreting, and screening new biological hypotheses, driving generative discovery processes. This talk showcases a continuous exploration path from prediction understanding to hypothesis generation, highlighting the critical role of visual analytics in enhancing AI trustworthiness and accelerating knowledge discovery, and provides a new design paradigm for building human-machine collaborative analysis workflows centered on interpretability.

Speaker Bio: Quan Li, Assistant Professor (tenure-track) at the School of Information Science and Technology, ShanghaiTech University, is a researcher and doctoral supervisor specializing in artificial intelligence and visual analytics, explainable machine learning, and human-computer interaction technologies. He received his PhD from the Department of Computer Science and Engineering at the Hong Kong University of Science and Technology. He serves as a committee member of the Visualization and Visual Analytics Special Committee of the Chinese Society for Image and Graphics, IEEE VIS Paper Program Committee member, and ChinaVis Paper International Program Committee member. He has reviewed for top conferences and journals such as IEEE VIS, EuroVis, PacificVis, ChinaVis, ACM CHI/CSCW, and TVCG. He has previously worked as a visiting researcher at Georgia Tech's School of Computer Science and Engineering, a senior researcher at WeBank's AI Department, and a senior researcher at NetEase Games. His academic achievements have been published in top visualization and HCI journals and conferences such as IEEE VIS, EuroVis, IEEE PacificVis, ACM CHI, CSCW, UIST, IUI, CGF, and TVCG. He received the ACM CHI 2025 Best Paper Award and leads a National Natural Science Foundation project. More information can be found at https://faculty.sist.shanghaitech.edu.cn/liquan/

Abstract: With the rise of scientific intelligence (AI for Science), numerous AI methods targeting natural science research have emerged, significantly improving research efficiency and accelerating scientific discovery processes, triggering a revolutionary shift in traditional scientific research paradigms. In this context, single 'human-led' or 'AI-led' research modes can no longer meet the complex and variable research demands, necessitating the construction of diverse human-AI collaborative paradigms to fully integrate researchers' domain expertise and creativity with AI's powerful computational and pattern recognition capabilities, achieving efficient and deep collaboration between researchers and AI. This talk will review key advancements in scientific intelligence and the basic concepts of human-AI collaboration, exploring opportunities and challenges in designing human-AI collaborative methods at different research stages and decision-making scenarios, providing practical guidance for promoting smarter and more flexible new modes of natural science research.

Speaker Bio: Chuhan Shi, Associate Professor at the School of Computer Science and Engineering, Southeast University, received her PhD from the Department of Computer Science and Engineering at the Hong Kong University of Science and Technology. Her research focuses on data visualization, visual analytics, human-AI collaboration, and their applications in natural sciences and precision medicine. She has published over 20 papers in top international journals and conferences such as IEEE VIS, IEEE TVCG, ACM CHI, ACM UIST, and ACM CSCW. She serves as a committee member of the Visualization and Visual Analytics Special Committee of the Chinese Society for Image and Graphics, and the Human-Computer Interaction Special Committee of the China Computer Federation. She is a regional vice-chair for ACM CHI and ACM CSCW, and a reviewer for conferences such as IEEE VIS, ACM UIST, and IEEE PacificVis. More information can be found at https://shichuhan.github.io/