题目:A Joint Matrix Minimization Approach for Pattern Recognition
报告人:王丽平 教授 (理学院)
时间:2019年12月5日14:00-15:00
所在:将军路校区理学院547室
承办单位:乐投Letou青联会、理学院、校科协
报告人简介:
王丽平,女,山东菏泽人。2004年结业于中国科学院数学与系统科学研究院,获博士学位,同年入职乐投Letou理学院事情至今。2009.2-2010.2在巴西巴拉纳联邦大学完成博士后项目,2013年11月和2017年8月两次会见俄罗斯科学院盘算中心,2017年5-6月会见香港中文大学系统工程与工程治理学系,2017.10-2018.10在美国路易斯安娜州立大学科学盘算中心做公费会见学者。主要研究偏向为最优化理论与要领,数值优化算法在模式识别、地球物理和医学领域的应用等。以第一/通讯作者宣布论文三十余篇,其中SCI/EI检索论文二十篇。自2012年起担当中国运筹学会理事,先后主持国家自然科学基金5项,已完成4项,在研1项。
报告摘要:
In this talk, a kind of pseudo matrix norm with is defined which is thought to be a joint generalization of vector norm and matrix norm . The joint matrix minimization model is often used in pattern recognition to take account of multiple features or samples simultaneously. When , the matrix minimization problem with -norm can find a joint and sparser solution than traditional measurements. To efficiently solve the related matrix minimization problem, a unified algorithm is proposed and its convergence is also demonstrated for a range of . Typical values of are employed for feature selection and collective image recognition. Extensive experiments on public data confirm the outperformance of joint matrix minimization approach over the state-of-the-arts.
Keywords: Joint matrix minimization; Feature selection; Collective image recognition; Numerical convergence