Professor Syed Ejaz Ahmed, Brock University
"Deep Statistical Learning: Post-Shrinkage Strategies for High-Dimensional Regression"
Friday, 27 November 2026. 13:00 (note the change from the usual time)
Room 383 ASBS
Abstract
In this talk, we investigate estimation and prediction in regression models where the coefficient vector is potentially constrained to a candidate subspace, with applications spanning both low- and high-dimensional settings. We develop a post-shrinkage framework that combines an initial shrinkage or penalization step with a subsequent estimation stage that exploits the information contained in the candidate subspace. The proposed procedures encompass penalized, post-pretest, and post-shrinkage estimators, and their large-sample behavior is characterized through asymptotic distributional analyses of quadratic bias and risk.
We complement the theoretical analysis with extensive Monte Carlo simulations designed to assess the finite-sample performance of the proposed post-shrinkage procedures. Real-data applications further demonstrate their practical utility through comparisons with standard penalized estimators and modern machine learning methods.
Across substantial regions of the parameter space associated with commonly imposed sparsity structures, the proposed post-shrinkage strategies achieve consistent performance improvements relative to competing methods. Notably, their advantages remain evident even when the assumed sparsity structure is substantially misspecified or violated. These findings demonstrate the robustness of the post-shrinkage approach and highlight its potential as a flexible framework for estimation and prediction in both low- and high-dimensional regression. Some open research problems and promising directions for future investigation will also be presented to stimulate discussion and encourage research and collaboration.
Bio
Professor Ejaz Ahmed is a Distinguished Professor and internationally recognized leader in Statistics, Data Science, and Statistical Machine Learning. He is a Professor in the Department of Mathematics and Statistics at Brock University, Canada, where he previously served as Dean of the Faculty of Mathematics and Science.His work has been continuously supported by prestigious funding agencies, including the Natural Sciences and Engineering Research Council of Canada (NSERC), the Canadian Institutes of Health Research (CIHR), and the Ontario Centre of Innovation (formerly the Ontario Centres of Excellence, OCE).
Professor Ahmed's scholarly achievements have earned him numerous national and international distinctions, including the prestigious Bualuang ASEAN Chair Professorship. He has received numerous awards and honors, served as Editor and Associate Editor for leading scientific journals, held several adjunct and visiting professorships, and delivered invited keynote and plenary lectures at conferences and universities around the world. In 2011, he founded the internationally renowned workshop series on High-Dimensional Data Analysis, which has become a leading forum for researchers in the field.
Professor Ahmed is a Fellow of the American Statistical Association (ASA), an elected member of the International Statistical Institute (ISI), and a Fellow of the Royal Statistical Society (RSS). He has served as Review Editor of Technometrics for more than 15 years. Dedicated to mentoring the next generation of statisticians and data scientists, he has supervised more than 25 PhD and numerous Master's students, as well as Postdoctoral Fellows, many of whom now hold academic and research positions worldwide.
He is currently the President of the Business and Industrial Statistics Section (BISS) of the Statistical Society of Canada (SSC) and serves on the Board of Governors of the Canadian Statistical Sciences Institute (CANSSI). Recently, he been inducted as a Foreign Member of the Luxembourg Academy of Sciences and Arts (LASA).
For further information, please contact business-seminar-series@glasgow.ac.uk
We foster a positive and productive environment for seminars through our Code of Conduct
First published: 20 November 2026