Abstract: Traditional machine learning model data quality governance follows the process of collecting first and then governing. However, with the rapid development of large models, expanding training data through data generation has become a new research direction. This report will combine existing research work and personal experience to explore the transformation of research thinking in the era of large models.

Speaker Bio: Changjian Chen is currently an Assistant Professor and doctoral supervisor at the School of Information Science and Engineering, Hunan University, and Deputy Director of the Hunan Provincial Key Laboratory of Blockchain Underlying Technology and Applications. He has conducted in-depth research on visual analysis and quality improvement of machine learning data for a long time. In recent years, he has published more than 20 academic papers in domestic and international journals and conferences such as IEEE TVCG, The Lancet Digital Health, and IEEE VIS, including more than 10 first-author/corresponding author CCF A-class papers. Due to his work on data quality improvement, he was invited to serve as the archival chair of IEEE VIS (CCF A) and as a (senior) program committee member of international conferences such as IEEE VIS, IEEE PacificVis (TVCG track), and ACM MM. He has led or participated in 7 projects including National Natural Science Foundation and major science and technology projects in Hunan Province. He received the Excellent Doctoral Dissertation Incentive Program of the Chinese Society for Graphics. His research results have been applied to Shanghai Data Exchange, Hunan Meteorological Observatory, CEC Wanwei, Beijing Winter Olympics, aerospace, and other units.

Abstract: This report will combine the speaker's academic growth experience from doctoral student to young teacher, sharing how visualization research comprehensively shapes researchers' comprehensive abilities. As a typical interdisciplinary research, visualization research not only requires solid technical foundation, but also needs to cultivate innovative thinking, teamwork, academic expression and other multi-dimensional abilities. The report will try to analyze five key links in the research process: the conception and screening of innovative ideas, the formation of interdisciplinary teams, the engineering implementation of research plans, the writing and publication of academic papers, and the presentation and reporting of research results, briefly discussing how these research experiences can be transformed into lasting career competitiveness.

Speaker Bio: Zikun Deng, Associate Professor and doctoral supervisor at the School of Software Engineering, South China University of Technology, fixed researcher at the Ministry of Education Key Laboratory of Big Data and Intelligent Robotics and the National Key Laboratory of Subtropical Architecture and Urban Science. He received his PhD from the National Key Laboratory of CAD&CG at Zhejiang University in 2023. His research interests focus on visualization and visual analytics, data mining, human-computer interaction, and related technology applications in smart cities, digital twins, and industrial software. He has led or participated in National Natural Science Foundation projects, National Key R&D Program projects, Guangdong Natural Science Foundation projects, and Guangdong Regional Joint Key projects. In the past 5 years, he has published more than 10 high-level papers in high-level journals and conferences such as IEEE TVCG, Computational Visual Media, IEEE VIS, and ACM CHI. He serves as a reviewer for journals and conferences such as IEEE TVCG, Journal of Computer-Aided Design & Computer Graphics, IEEE VIS, ACM CHI, and PacificVis, and received the Third Prize of the Lu Zengyong CAD&CG High-Tech Award.

Abstract: This report will introduce a series of systematic research I conducted during my doctoral studies on "how to combine data visualization and human-computer interaction technologies to improve data science process efficiency." The work focuses on the widely used analysis tool Computational Notebook (such as Jupyter Notebook), aiming to support data analysts with different needs and backgrounds to more efficiently complete result communication and knowledge transfer. Although these tools have become standard in data science, they still face many challenges in promoting content understanding and collaborative communication. Although such tools have become important infrastructure for data science work, there are still many challenges in content understanding, structural organization, and collaborative communication. To this end, we propose a series of design strategies, from information generation within cells, logical associations between cells, to the overall structural overview of the entire Notebook, gradually improving its user experience and communication efficiency. At the same time, I will also combine my own experience of gradually growing from a research novice to a doctoral researcher in the VisLab team, sharing some practical mechanisms of the team in student training.

Speaker Bio: Yanna Lin, Postdoctoral Researcher at Hong Kong University of Science and Technology, supervised by Professor Huamin Qu. She has received the Hong Kong Government's highest award "Hong Kong PhD Fellowship Scheme" (HKPFS), published more than ten papers in international top conferences and journals in visualization and human-computer interaction fields (such as IEEE VIS, ACM CHI, IEEE TVCG), and received Best Paper Nomination Awards from IEEE VIS and ChinaVis. She serves as a program committee member for conferences such as CHI, CIKM, and ChinaVis, and has long served as a reviewer for international top conferences and journals such as IEEE VIS, TVCG, and UIST. Her research focuses on data analysis, visual analytics, and human-computer interaction, committed to optimizing exploratory data analysis workflows. For more information, please visit her personal homepage: https://yannahhh.github.io.

Abstract: Natural language-based human-computer interaction patterns are becoming increasingly popular and widely used in complex task scenarios such as creative design and data analysis. However, the interpretability of natural language models has gradually become a key factor limiting the improvement of human-computer collaboration efficiency. The "black box" characteristics of models make it difficult for users to understand the model's behavioral characteristics and processing logic, resulting in low efficiency in processes such as constructing model inputs and evaluating model outputs. This research addresses model interpretability issues in multiple complex task scenarios, proposing new visual analytics methods to help users explore the model's input space, diagnose the model's output process, and evaluate the model's safety and reliability, thereby improving model interpretability and human-computer collaboration efficiency.

Speaker Bio: Yingchaojie Feng, PhD from Zhejiang University, supervised by Professor Wei Chen. Research interests include visualization, natural language processing, and human-computer interaction. The main goal is to improve human-computer collaboration efficiency through designing visualization interaction systems. Related achievements have been published in journals and conferences such as TVCG, ACL, and TKDE.