2026.06.21 Colloquia Seminars
| Speaker |
Prof. Keming Yu, Brunel University of London |
| Title |
A Linear Extremile Regression and its Semi-Supervised Learning |
| Time |
2026.07.06 11:00-12:00 |
| Venue |
N226 |
| Abstract |
Modelling and predicting extreme events remain one of the most important challenges across a wide range of disciplines, including finance, climate science, environmental studies, and risk management, as the most significant impacts often arise from rare, high-consequence events rather than average conditions. Recent research has increasingly focused not only on estimating tail probabilities but also on understanding how covariates influence extreme outcomes. Extremile regression, a recently proposed least-squares analogue of quantile regression, provides a promising framework for analysing extreme responses. However, two important challenges remain. First, existing extremile regression methods are largely nonparametric and may suffer from the curse of dimensionality in high-dimensional settings, leading to data sparsity issues and substantial difficulties in achieving √n-consistent estimation. Second, labelled data are often scarce, costly, and time-consuming to obtain in applications involving extreme events, whereas unlabeled data are typically abundant. To address the first challenge, this paper derives a linear extremile regression to avoid directly estimating the unknown nonparametric component, thereby enabling the construction of a √n-consistent estimator for the finite-dimensional parameters of interest. To address the second challenge, we develop a semi-supervised learning framework that effectively incorporates unlabeled data to improve estimation efficiency and reduce overfitting, even when the assumed linear extremile regression model is misspecified. |
| Biography |
Professor Keming Yu is Professor and Chair of Statistics and Data Science at Brunel University of London. He was recognised among the world's top 2% of scientists in the Stanford University/Elsevier Single-Year Citation Impact Rankings in 2021, 2022, 2023, and 2025. In 2024, he was named a Highly Ranked Scholar by ScholarGPS. In 2026, he was ranked 196th in the UK and 3,054th globally in Research.com's ranking of the world's leading mathematicians. Professor Yu is internationally recognised as a pioneer of Bayesian quantile regression, with influential contributions to statistical methodology and data science. | |
