主题:【鲁棒与随机优化系列讲座(七)】Some Applications of Gaussian Mixture Models in Optimization under Uncertainty
主讲人:同济大学胡照林教授
主持人:英国威廉希尔公司徐亮教授
时间:2021年5月20日(星期四)14:00-15:00
直播平台及会议ID:腾讯会议 会议ID:553 562 939
主办单位:英国威廉希尔公司 科研处
主讲人简介:
胡照林,同济大学经济与管理学院教授、同济大学青年百人计划入选者。他分别于浙江大学数学系和香港科技大学工业工程及物流管理系获得学士和博士学位。他的研究兴趣包括随机优化,仿真理论和实践,机器学习,金融风险管理,不确定环境决策等。在MS, OR, JOC, IIE, NRL, WSC等主流期刊和会议发表论文。研究获得国家自然科学基金优青、面上、青年基金的资助。曾获得2012 Institute of Industrial Engineers Pritsker Doctoral Dissertation Award, 3rd place。
内容提要:
In this talk, we discuss using Gaussian mixture model (GMM) based statistical learning to support optimization under uncertainty. First, we use GMM to conduct input modeling for some stochastic optimization problems involving risk measures. We consider the linear portfolio optimization problems with some risk measures, e.g., conditional value-at-risk and entropic risk measure, and show that under GMM the problems admit analytical structures. We also consider several classes of chance constrained programs (CCP), and discuss how to solve the problems by integrating the structures of CCPs and GMM. Second, we use GMM to learn the system response surface in simulation optimization. We take GMM as sampling distributions and design integrated random search algorithms to solve simulation optimization problems.
本讲座介绍基于高斯混合模型的统计学习方法,并用于支持不确定环境下的优化。首先,利用高斯混合模型建模不确定参数,并用于解决带风险测度的随机优化,如带风险测度的投资组合优化。然后,用高斯混合模型来学习仿真优化的系统相应曲面。