报告题目： Some recent progresses on gradient methods

报 告 人：Assistant Prof. YaKui Huang Hebei University of Technology

照    片：

邀 请 人：刘三阳

报告时间：2020年12月1日（周二） 10：00-11：00

报告地点：南校区信远楼二区206

报告人简介：黄亚魁，河北工业大学副教授、元光学者、硕士生导师。从西安电子科技大学获得学士、硕士、博士学位之后，到中科院作博士后。长期从事梯度法理论及应用的研究，在国际权威的优化计算类和数据挖掘类期刊发表SCI论文10余篇。现主持国家自然科学基金青年项目1项，曾主持中国博士后基金1项，参与多项国家自然科学基金。兼任中国运筹学会数学规划分会青年理事、河北省运筹学会理事。

报告摘要： Since its proposition in Cauchy (1847), one milestone work along the gradient method is the Barzilai-Borwein (nonmonotone) method (1988), while another significant work is the Yuan stepsize in (2006), which leads to the appearance of the efficient Dai-Yuan (monotone) gradient method (2005). In this talk, some recent progresses on gradient methods will be introduced. We will present three new stepsizes which have two-dimensional quadratic termination property. Such a property can significantly improve the performance of the gradient method. By making use of the new stepsizes and the Barzilai-Borwein stepsizes, we develop new efficient gradient methods. Numerical experiments demonstrate that our proposed methods outperform the most successful gradient methods in the literature.