Keynote Speaker
Professor Xiaodan Fan, PhD
Embracing AI for Bioinformatics Research: Statistician’s Opportunities and Mission
Abstract
Statisticians have long been key players in bioinformatics, benefiting from the fact that bioinformatics is a big data area. But Artificial intelligence (AI) has recently excelled with its staggering predictive power in most big data areas including bioinformatics. However, there is still a critical gap between predictive performance and scientific understanding - a gap that the statisticians are uniquely positioned to fill.
This talk explores statistician’s opportunities and mission in bioinformatics research. On one hand, we can leverage highly flexible AI architectures to infer complex relationships from heterogeneous datasets that traditional parametric models cannot capture. I will use both a genomic sequence example in virus mutation study and a protein structure example in computational immunology to illustrate our efforts in this direction. On the other hand, it is our mission to integrate core statistical principles with modern AI to produce more accurate and reproducible results. In bioinformatics, where decisions may directly impact human health, uncertainty quantification is especially important. We will discuss how to perform Bayesian uncertainty quantification efficiently for deep learning frameworks.
Biography
Xiaodan Fan is a professor in Statistics at the Chinese University of Hong Kong. He earned a BE in Automation and an MS in Pattern Recognition & Intelligent Systems from Tsinghua University, China, followed by a PhD in Statistics from Harvard University. His research interests encompass probabilistic modeling, machine learning, statistical computing, and their applications in biomedicine. He has served as an associate editor for the Journal of Computational and Graphical Statistics since July 2012 and as an editorial board member for Frontiers in Genetics since April 2015. Additionally, he was a guest editor for the IEEE/ACM Transactions on Computational Biology and Bioinformatics from November 2014 to April 2018. Fan has reviewed for several prestigious journals, including the Journal of the American Statistical Association and Bioinformatics, and has actively participated in organizing numerous academic conferences in his field.
|
Keynote Speaker
Associate Professor Han-Ming Wu, PhD
S-IGTD: Supervised Image Generator for Tabular Data for CNN-based Deep Learning Multiclass Classification
Abstract
The application of Convolutional Neural Networks (CNNs) to tabular data is fundamentally constrained by the permutation invariance of features, which lack intrinsic spatial locality. While recent methods like the Image Generator for Tabular Data (IGTD) map features onto 2D grids, their unsupervised nature prioritizes global similarity, often preserving non-discriminative noise at the expense of class-specific signals. This paper introduces the Supervised Image Generator for Tabular Data (S-IGTD), a framework that restructures feature rearrangement through between-group correlation matrices to embed discriminative signals into the grid topology. We provide a rigorous theoretical development of the method, establishing the asymptotic consistency of the S-IGTD estimator and deriving specific topological divergence conditions under which the supervised topology strictly enhances the signal-to-noise ratio of local image patches. Empirical validation on simulated and real biological datasets confirms that S-IGTD yields statistically significant improvements over unsupervised topology generators and established baselines. These results elucidate that treating the structural representation of data as a supervised optimization problem enables CNNs to effectively leverage spatial inductive biases, offering a statistically grounded methodology for representation learning of tabular data.
Biography
Han-Ming Wu is an Associate Professor in the Department of Statistics at National Chengchi University, Taipei, Taiwan. He received his Ph.D. in Statistics from National Chiao Tung University and later worked as a postdoctoral researcher at the Institute of Statistical Science, Academia Sinica. Before joining NCCU, he served as a professor in the Department of Statistics at National Taipei University. Prof. Wu has been actively involved in the international statistical community and served as the Chair of the Young Statisticians Committee of the International Statistical Institute (ISI) from 2017 to 2019, during which he organized the 2nd ISI Young Statisticians Workshop (YS-ISI2019) in Kuala Lumpur. He has long been promoting statistical data analysis using the R programming language. His research interests include high-dimensional data analysis, information visualization, microarray data analysis, statistical computing, R/Java software development, machine learning, and symbolic data analysis.
|
Invited Speaker
Professor Jung Jin Lee, PhD
Web-based Education for Statistics and Data Science using eStat, R, and Python
Abstract
Well-known statistical packages, such as SAS, SPSS, R, and Python, are widely used for Statistics education. They are good for data processing and research but do not have enough modules for Statistics education. A web-based free software at www.estat.me, eStat, specially designed for teaching Statistics. The eStat has numerous simulation modules and web data processing capabilities, including all statistical distributions, the central limit theorem, confidence intervals, hypothesis testing, analysis of variance, nonparametric tests, regression analysis, and forecasting. The eStat is also linked to a web-book system with embedded modules that include many examples, presentation files, and lecture videos. The eStat system has recently been extended to support Data Science education as well as Statistics, encompassing data visualization, supervised learning, and unsupervised learning. The Data Science modules also include R and Python commands as well as eStat for working with data, such as multidimensional tables, principal component analysis, decision trees, Bayes classification, neural networks, and k-means clustering. The demonstration of teaching Statistics and Data Science will be followed by a brief introduction to eStat.
The eStat system has been adopted for teaching at several universities worldwide, including Korea, Japan, Azerbaijan, and Uzbekistan, and it has proven very useful for both students and lecturers. We invite scholars to exchange interesting ideas, experiences, and modules to eStat for a desirable Statistics and Data Science education.
Biography
Professor Jung Jin Lee is an accomplished statistician, computer scientist, and educator with more than four decades of contributions to research, teaching, and academic leadership. His expertise spans multivariate statistical analysis, classification methods, data mining, information retrieval, multiple-criteria decision making, and statistical software development. He earned his Ph.D. in Operations Research from Case Western Reserve University, USA. Professor Lee previously served for many years at Soongsil University, Korea, where he held several leadership positions including Chair of the Statistics Department, Dean of Academic Affairs, Vice President for Foreign Affairs, and Dean of the Graduate School. He is currently a Professor in the Department of Mathematics and Advisor to the Rector at New Uzbekistan University. Professor Lee is also recognized as the founder and developer of eStat, an innovative web-based statistical education and analysis system designed to support interactive learning of statistics.
|