B1766
Title: Admixture analysis of multi-site multivariate time series
Authors: Chi Tim Ng - Hang Seng University of Hong Kong (Hong Kong) [presenting]
Yan Wu - Hang Seng University of Hong Kong (Hong Kong)
Abstract: By introducing the ideas of admixture analysis and latent Dirichlet distributions, statistical models and methods are developed for extracting the information from the multisite high-dimensional time series data about the hidden driving forces that cannot be observed directly. Though the concepts of admixture have been employed by researchers in the context of population genetics and text mining, this is the first research that extends these ideas to multi-site high-dimensional time series analysis. The admixture components in the novel model can then be used to describe the so-called hidden driving forces. With the extra time ingredient, the time of appearance and disappearance of a driving force is further investigated. This cannot be done directly with existing time series clustering methods and factor analysis methods. [The work described in this presentation was fully supported by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No., e.g. UGC/FDS14/P04/22).]
