报告题目:Optimization Models and Methodologies for Binary/Integer Matrix Factorization
报告人：彭积明 教授
报告时间：10:15-11:15
报告地点：广知楼B104
报告题目：Optimization Models and Methodologies for Binary/Integer Matrix Factorization
报告时间：2015年5月15日周五上午10:15-11:15
报告地点：广知楼B104
报告人：美国休斯敦大学工业工程系彭积明教授
报告摘要：
零一整数矩阵分解是大规模高维数据处理中的一个重要模型.本报告中, 我们第一次引入约束零一整数矩阵分解.在这里我们要求矩阵乘积是零一矩阵，并提出相应的交替更新算法。在每次迭代中，我们要解一个特定零一线性优化问题去更新涉及的矩阵。我们探求该特定零一线性优化子问题和聚类分析的内在联系, 用之于开发一种有效的近似算法.对低维相关矩阵, 我们的算法极大地改进了现有的零一矩阵分解算法。对高维相关矩阵,，我们用随机化技术开发一种线性时间近似算法, 并估计解的质量。数值实验来表明新的模型可以达到原零一整数矩阵分解的目的, 新的算法可以十分有效地找到高质量的逼近解。如果时间允许，我们将进一步讨论如何扩展到其他整数矩阵分解模型和算法。
报告人简介：
Jiming Peng is an associate professor in the department of industrial engineering, University of Houston. He received his PhD degree in operations research in 2001 from Delft University of technology, the Netherlands. Previously he worked in McMaster University in Canada, and University of Illinois at Urbana-Champaign.
His research interest covers several branches in optimization, with a recent focus on the development of effective algorithms for large-scale non-convex and mixed integer programming, with applications to big data, finance and computer vision. He has published a research monograph and about sixty peer-reviewed papers. He and his student have received numerous awards for their research contribution in optimization and financial engineering including Stieljes prize in Holland (2001), finalist of Tucker prize (2003), primer research excellence award from Ontario (2003), first runner-up for the annual Morgan Stanley Prize for Excellence in Financial Market (2012), best research paper award in financial service, Informs (2013).