题  目：Optimize a function based on inaccurate function values

报告人： Zaikun Zhang （The Hong Kong Polytechnic University）

时  间：2018年6月20日 11:00-13:00

地  点：西大楼308报告厅

摘 要:In real-world optimization problems, it happens very often that the derivatives of the objective functions are not available. These problems can be solved by algorithms that require only function values, also known as derivative-free optimization algorithms. But what if the function values are inaccurate?

This talk focuses on the direct search method, one of the simplest derivative-free optimization algorithms, and study the behavior of direct search working with inaccurate function values. Several questions will be addressed. 1) What kind of solution can we obtain by direct search if the function values are inaccurate? 2) What is the worst case complexity to attain such a solution? 3) Given the inaccuracy in the function values, when to terminate the algorithm in order to guarantee the quality of the solution and also avoid "over-optimization"? 4) If we have several oracles that provide function values with different levels of accuracy, how to use them effectively and efficiently? This talk is based on a joint work with F. Delbos (IFPEN, Paris), S. Gratton (CERFACS and IRIT-ENSEEIHT, Toulouse), and B. Pauwels (IFPEN, Paris).

报告人简介：张在坤，香港理工大学应用数学系助理教授。张在坤 2007 年本科毕业于吉林大学数学学院，2012 年博士毕业于中科院数学与系统科学研究院，导师是袁亚湘院士。他毕业后访问科英布拉大学、欧洲科学计算研究与高等培训中心和图卢兹大学，从事博士后研究，于2016 年加入香港理工大学应用数学系。张在坤的研究领域为最优化算法与理论，主要研究兴趣是无导数方法、随机化方法、子空间方法和带有噪声的优化问题，目前主持主持香港研究资助局 ECS 项目一项以及 PROCORE---香港/法国合作研究项目一项 (与图卢兹大学及空客公司 IRT Saint Exupéry 研究所合作)。

邀请人：陶敏  老师