Analysis of Rapidly Evolving Multivariate Oscillations

• Activity Number: 147 - High-Dimensional Time Series Analysis and Its Applications
• Type: Invited
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Sofia C Olhede* and Adam Sykulski and Arthur Guillaumin and Jonathan Lilly and Jeffrey Earley
• Companies: University College London and Lancaster and UCL and NWRA and NWRA
• Keywords: time series; multivariate; oscillation

Abstract

It is classic to decompose a time series into oscillations, smooth variability, as well as an irregular component. When we observe more than one such time series, the description of an oscillation becomes more complex, as an underlying geometry will underpin our description. We shall discuss how to describe this phenomena, and show its utility when modelling and inferring observations from climate time series, namely the global drifter programme. We shall also discuss what happens if the generating mechanism of the time series is evolving very rapidly, violating classic assumptions, and what additional assumptions are needed to still be able to make inferences.