Speaker: 肖逊，University of Otago，the Department of Mathematics and Statistics

Inviter: 胡庆培

Title: Stable and Efficient Estimation for Sparse Single-Index Models

Language: Chinese

Time & Venue: 2026.11.23 14:00-15:00 南楼N402

Abstract: Sparse single-index regression accommodates nonlinear relationships while retaining sparse and interpretable covariate effects. However, estimation methods relying on the exact Hessian can suffer from numerical instability and high computational costs. We propose a fixed-damping Gauss–Newton algorithm that replaces the exact Hessian with a positive-definite curvature approximation, avoiding the construction of dense Hessian matrices. We establish convergence guarantees under suitable conditions and extend the framework to sparse single-index regression with fixed left censoring. Simulation studies demonstrate improved numerical stability and computational efficiency relative to the exact-Hessian approach, together with improved variable selection and prediction performance. Applications to aero-engine vibration and HIV viral load data illustrate the practical utility of the proposed method.