Exhaustive Visual Search for Information in Multidimensional Datasets
Overview of this DFG-aided joint research project between the Carolo-Wilhelmina-University
Braunschweig (computer graphics lab) and the Otto-von-Guericke University Magdeburg
(visual computing lab):
For analyzing characteristic information, like clusters, bifurcations, correlations or structure in
general, of several multidimensional datasets there are a lot of appropriate and useful
techniques, mostly based on statistic approaches, which known as data mining. Coeval
exists an enormous knowledge about the question how you can find information in 2D
images automatically, well-established as images processing.
The target of this research project is to use the knowledge from the images processing
and make a visual search to speed up the normally data mining-based analysis of
typical informations from a multidimensional dataset. We'll establish the term visual data
mining (vdm) respectively visual analytics (va) for that.
This will be done in a two step process: First, we generate systematically visualizations
from the dataset. Second, we analyze these visualizations goal oriented with image
Although it's basic research, three different applications are appreciable for the time being.
Supporting Non-Automatic Visual Search:
As pre-process we could choice a couple of best
visualizations from a multidimensional dataset, which will be present for a visual user-based
search of pairwise-correlated dataset-tokens, instead of present all – mostly thousands of – possible visualizations (and that would overburden the user).
Automatic Information Analysis:
We could make a very fast automatic information search in
datasets as well, which will be everywhere useful where huge amounts of data with lots of
tokens are generated (e.g. physic-research, automotive industry, climate simulation, and so
We could save automatically only these visualizations of a typical multidimensional
dataset which contains the informations we're looking for, so that we could delete all the other
visualizations and the dataset itself.