Academic Report of School of Mathematical Sciences [2025] No. 120  

(Series Report for High-Level University Construction No. 1222)  

Title: Neyman–Pearson Classifier with Successive Convex Approximation for Imbalanced Data  

Speaker: Professor Peng Heng (Hong Kong Baptist University)  

Time: 15:00–16:00, November 21, 2025 (Friday)  

Location: Classroom 1, Huixing Building, Yuehai Campus, Shenzhen University  

Abstract:  

The Neyman–Pearson (NP) paradigm in binary classification is developed as a new statistical approach for addressing the unequal importance of type I and type II errors in asymmetric statistical learning. It aims to create classifiers that minimize type II error while keeping type I error within a user-defined limit. However, most current NP classifiers involve a two-step process, where errors from the model estimation can accumulate during the second inference step, possibly resulting in unstable outcomes. This article is the first to attempt constructing NP classifiers using the difference-of-convex approximation and the successive convex approximation algorithm from an empirical optimization standpoint. The proposed one-step classifiers adhere to NP oracle inequalities, which are the NP paradigm’s equivalent to oracle inequalities in traditional binary classification. In addition to their appealing theoretical attributes, we demonstrate their numerical benefits in controlling prioritized errors through both simulations and real data analyses.  

Speaker Biography:  

Peng Heng is a Professor in the Department of Mathematics at Hong Kong Baptist University. He received his Ph.D. in Statistics from The Chinese University of Hong Kong in 2003 and conducted postdoctoral research at Princeton University from 2003 to 2006. His research primarily focuses on nonparametric and semiparametric models, model selection, high-dimensional data modeling, and mixture models. He is a member of the Institute of Mathematical Statistics (IMS), served as an Associate Editor for Statistica Sinica from 2011 to 2014, and is currently an Associate Editor for Computational Statistics and Data Analysis. He has also served as a reviewer for journals such as Annals of Statistics, JASA, JRSSB, Biometrika, and Statistica Sinica. He has published dozens of papers in top-tier international statistics journals, including Annals of Statistics, JASA, Biometrika, Statistica Sinica, TEST, and Computational Statistics and Data Analysis.  

All faculty and students are welcome!  

Host: Hu Xianghong  

School of Mathematical Sciences  

November 17, 2025