深圳技术大学大数据与互联网学院数学系学术报告（九）

报告标题：Interior Quasi-subgradient Method with non-Euclidean Distances for Constrained Quasi-convex Optimization Problems

报告人：胡耀华副教授（深圳大学）

报告时间：2022年11月28日（星期一） 16:00—17:30

报告地点：（腾讯会议）会议号：804-277-699 会议密码：123456

邀请人：王洪博士

摘要：Quasi-convex optimization is fundamental to the modeling of many practical problems in various fields such as economics, finance and industrial organization, and subgradient methods are practical iterative algorithms for solving large-scale convex or quasi-convex optimization problems. In this talk, an interior quasi-subgradient method is proposed based on the proximal distance to solve constrained nondifferentiable quasi-convex optimization problems. The properties of the specific quasi-subdifferentials of quasi-convex functions, including nonemptiness, quasi-monotonicity, and outer semicontinuity, will be presented. The convergence properties, including the global convergence and iteration complexity, of the interior quasi-subgradient method are investigated under the assumption of the Holder condition of order p, when using the constant/diminishing/dynamic stepsize rules. Convergence rate results are obtained by assuming a Holder-type weak sharp minimum condition relative to an induced proximal distance.

报告人简介：胡耀华，现任深圳大学数学与统计学院副教授，博士生导师，香港理工大学兼职博导，国家优秀青年科学基金获得者，深圳市海外高层次人才。主要从事连续优化理论、算法与应用研究，包括非凸稀疏优化、广义凸优化、复合优化等。先后主持国家自然科学基金4项，省市级科研项目9项。在SIAM Journal on Optimization, Journal of Machine Learning Research, European Journal of Operational Research, Briefings in Bioinformatics, Transportation Research Part B/ Part E等国际期刊发表论文40余篇，授权3项国家发明专利，开发多个生物信息学工具包与网页服务器。