当前位置: 首页 > 学术活动 > 正文
L0-Norm Regularized Methods for Multivariate Time Series Segmentation
时间:2026年11月16日 19:25 点击数:

报告人:黄玉梅

报告地点:线上腾讯会议(会议ID:595-728-477)

报告时间:2026年11月17日(星期二)09:00-10:00

邀请人:刘俊

报告摘要:

Time series segmentation is an important problem in time series analysis. The aim of time series segmentation is to partition the time series into homogeneous segments by detecting breakpoints which have abrupt changes or anomalies in the time series. Time series segmentation can provide meaningful information for further data analysis, prediction and policy decision. The abrupt changes usually do not appear frequently in a time series and the breakpoints may not be dense in a normal time series. In this talk, by minimizing the negative log-likelihood function of the time series, we propose a multivariate time series segmentation model which is regularized by the L0-norm  of the time series mean change. The proposed model is solved by an alternating  process. The dynamic programming (DP) algorithm is applied to determine the breakpoints and the cross-validation (CV) method is used to determine the parameters in the proposed model. The experiments of segmentation for both synthetic and real time series show that the proposed method is efficient for multivariate time series segmentation.

主讲人简介:

黄玉梅,兰州大学数学与统计学院教授,博导。2008年在香港浸会大学获得博士学位。研究兴趣包括数值线性代数和图像复原,已发表论文30余篇,发表期刊包括SIAM Journal on Imaging Sciences, IEEE transactions on Image Processing and SIAM Journal on Scientific Computing等。以第一完成人获得2021年甘肃省自然科学二等奖。

©2019 东北师范大学数学与统计学院 版权所有

地址:吉林省长春市人民大街5268号 邮编:130024 电话:0431-85099589 传真:0431-85098237

师德师风监督举报电话、邮箱:85099577 sxdw@nenu.edu.cn