学术报告
A Fast Smoothing Newton Method for Bilevel Hyperparameter Optimization in Logistic Classification
李庆娜（北京理工大学）
报告时间：2023年9月9日星期六 14：30-15：15
报告地点：沙河主楼E405
报告摘要: Logistic classification is a classical and well-performed learning method in machine learning. A regularization parameter, which significantly affects the classification performance, has to be chosen and this is usually done by the cross-validation procedure. In this paper, we reformulate the hyperparameter selection problem for logistic classification as a bilevel optimization problem in which the upper-level problem minimizes logistic loss of misclassified data points over all the cross-validation folds. The resulting bilevel optimization model is then converted to a KKT system with an extra lower bound constraint. To solve this system, we propose a smoothing Newton method, which is proved to converge to a strict local minimizer of the nonlinear system. Extensive numerical results verify the efficiency of the proposed approach.
报告人简介：李庆娜，北京理工大学数学与统计学院教授,博士生导师。湖南大学本科、博士，中科院数学与系统科学研究院博士后. .曾访问英国南安普顿大学，新加坡国立大学、香港中文大学等。主持国家自然科学基金青年、面上项目等. 任中国运筹学会数学优化分会理事和北京运筹学会理事。主要研究最优化理论与算法及应用。著有专著《多维标度方法》，教材《最优化方法》、《凸分析讲义》等三部。获2020、2021北京市高校优秀毕业设计指导教师荣誉称号。2021年获北京运筹学会优秀青年论文奖.
邀请人：崔春风
