Multiple Changepoints Detection via Deep Learning
时间：2016-11-17 14:00 ~ 15:00
Changepoints are extremely important features to consider when homogenizing time series and analyzing its trends and variations. This paper introduces a new changepoint detection method using supervised deep learning algorithms. Deep learning is a branch of machine learning and uses a cascade of multiple layers of nonlinear processing units for feature extraction and transformation. The advantage of using deep neural networks is that the method does not need to assume any specific probability model or correlation structure of the time series to be tested. In this study we focus on the appropriate design of deep neural?networks for the task of changepoint detection by specifying the multiple layers and developing the training procedure. The method is applied in the analysis of three time series: an annual precipitation series from New Bedford, MA, a century of monthly temperatures from Tuscaloosa, AL, and the North Atlantic basin ropical cyclone record.
About the Speaker:
Yuan Xue is Associate Professor in the School of Statistics at the University of International Business and Economics. He obtained his Ph.D. in Statistics from the University of Georgia in 2012. His research interests include high-dimensional data analysis and data mining.