Estimating dynamic graphical models from multivariate time-series data
Alex J. Gibberd, James D. B. Nelson · 2015
We consider the problem of estimating dynamic graphical models that describe the time-evolving conditional dependency structure between a set of data-streams. The bulk of work in such graphical structure learning problems has focused in the stationary i.i.d setting. However, when one introduces dynamics to such models we are forced to make additional assumptions about how the estimated distributions may vary over time. In order to examine the effect of such assumptions we introduce two regularisation schemes that encourage piecewise constant structure within Gaussian graphical models. This article reviews previous work in the field and gives an introduction to our current research.