The "Machine Learning for Time Series" group aims at developing  mathematical and computational tools to infer structure and models from high-dimensional time series data, especially for complex  physical and chemical systems. These systems are generally stochastic, nonlinear, high-dimensional, metastable and out of equilibrium, and advanced statistical techniques become increasingly necessary for analysis of the related simulation and experimental 
data. Our work is interdisciplinary at the interfaces of computational sciences, statistics, dynamic systems theory and machine learning, and focuses on the following topics:

  1. Compression and sparse representation of dynamic data. 
  2. Inverse modeling and spectral analysis of Markovian and non-Markovian processes.
  3. Cointegration and causality analysis of multiple time series.