报告地点：行健楼学术活动室526
邀请人：吴奕飞教授
Abstract: This report examines the future of statistical modelling in the context of increasingly complex data. It begins by highlighting high-profile forecasting failures (e.g., New Coke, the 2016 US Election, and Bank of England inflation forecasts) to illustrate the real-world consequences of models that fail to capture critical factors like emotional attachment or unexpected supply chain shocks.
As the data landscape grows more complex, the core methodological challenges include high-dimensional feature selection, strong correlations, limited sample sizes, and model (in)stability. To address these, the presentation advocates for a shift toward building trustworthy models through several key approaches: (1) Robust Methodologies: Employing techniques like regularization and advanced screening methods (e.g., Air-HOLP) to handle high-dimensional correlated data, improve prediction, and enhance interpretability. (2) Stability and Uncertainty Quantification: Emphasizing the assessment of feature selection stability through methods like stability selection and Bayesian inference. This allows for quantifying uncertainty (e.g., with Bayesian credible intervals) and incorporating domain expert knowledge to improve the reliability of selected features. (3) Transparency and Interpretability: Moving beyond simple prediction to create final models that are interpretable, with a strong emphasis on understanding the underlying data-generating process.
The modern statistical methodology is essential for enabling better scientific discovery and responsible, data-driven decisions in fields like health, environmental science, and policy. The future of the field lies in developing robust, stable, and interpretable models through close collaboration between methodological researchers and domain experts.
报告人简介： Professor Samuel Muller is an esteemed statistician and higher education leader, currently serving as the Executive Dean of the Faculty of Science and Engineering at Macquarie University since February 2025. With a Ph.D. in Mathematics from the University of Bern, he has held prominent academic positions, including Head of School, Head of Department, and Associate Dean. His leadership extends to professional organizations, having served as President of the International Biometric Society – Australasian Region. His research focuses on: variable selection and feature extraction; statistical model building; high-dimensional data analysis; robust estimation methods; multiplatform data classification.
Professor Muller is the Editor-in-Chief of the Australian and New Zealand Journal of Statistics, reflecting his commitment to advancing statistical research. He is recognized internationally as a Fellow of the American Statistical Association, an Elected Member of the International Statistical Institute, and a Member of the Honor Society of Research Scientists and Engineers (Sigma Xi). His work bridges statistical theory and practical applications, contributing to research and policy development across various scientific and industry sectors.
