Visualization of Unsteady Multivariate Data Using Information Theory

Heike Janicke

(Institut fur Informatik, Universitat Leipzig)

Visualizations are well suited to communicate a large amount of complex data. With increasing resolution in the spatial and temporal domain, simple imaging techniques meet their limits, as it is quite difficult to display multiple variables in 3D or analyze long video sequences. Techniques from information theory can help to compress the original data and extract the essential. In the presentation two techniques based on ideas from computational mechanics will be presented: Local statistical complexity identifies regions in the data-set whose local dynamics diverge from the dynamics in the rest of the data-set and can be used to identify extraordinary events. A precise analysis of these structures is provided by the second method - epsilon-machines. Epsilon-machines reconstruct a finite state machine from the given data and can be visualized as directed graphs. These graphs allow for a steady two-dimensional visualization of unsteady data. Both techniques will be illustrated using examples from fluid and weather simulations.
Tuesday 23rd September 2008, 14:00
Robert Recorde Room
Department of Computer Science