题   目：ADMM-based Bilevel Descent Aggregation Algorithm for Sparse Hyperparameter Selection

报告人：丁彦昀  

单   位：深圳职业技术大学

时   间：2026年2月7日 15:00

地   点：龙子湖校区九章学堂南楼C座302

摘   要：This paper proposes a novel bilevel optimization framework for a class of nonsmooth convex sparse optimization problems, addressing the key limitation of the commonly required Lower-Level Singleton (LLS) assumption. The framework integrates the Alternating Direction Method of Multipliers (ADMM) for efficiently solving the lower-level problem with a Bilevel Descent Aggregation (BDA) algorithm for exploring the hyperparameter space in the upper level. A primary contribution is a new convergence analysis, which proves that the proposed ADMM-BDA algorithm achieves global convergence under significantly relaxed conditions, moving beyond the restrictive LLS assumption. Numerical experiments on synthetic and real-world data demonstrate the superior effectiveness and robustness of ADMM-BDA compared to state-of-the-art methods, particularly when the lower-level problem involves elastic-net penalized statistical models.

报告人简介：丁彦昀，博士，讲师，2023年加入深圳职业技术大学。研究方向为最优化理论与方法，在iScience（Cell子刊）、Optim. Methods Softw.、J. Math. Imaging Vis.、J. Nonlinear Var. Anal.等期刊发表论文十余篇。