A sparse graph recovery based approach for multivariate time series segmentation

Our approach for multivariate time series segmentation, called 'tGLAD', utilizes probabilistic graphical models called conditional independence (CI) graphs. These graphs represent partial correlations between nodes, which correspond to variables in the time series. By applying a graph recovery model to short intervals of the time series, we can obtain a temporal CI graph representation of the entire series. We then use a trajectory tracking algorithm to analyze the evolution of the graphs and determine a suitable segmentation. This approach has a competitive time complexity and has shown successful results in a Physical Activity Monitoring dataset.