Finding Hidden Factors in Large Spatiotemporal Data Sets

In many fields of science, engineering, medicine and economics, large or huge data sets are routinely collected. Processing and transforming such data to intelligible form for the human user is becoming one of the most urgent problems in near future. Neural networks and related statistical machine learning methods have turned out to be promising solutions. In many cases, the data matrix has both a spatial and a temporal dimension. Removing correlations and thus reducing the dimensionality is typically the first step in the processing. After this, higher-order statistical methods such as independent component analysis can often reveal the structure of the data by finding hidden factors. This can sometimes be enhanced by semi-blind techniques such as temporal filtering in order to use prior knowledge. Examples to be covered in the talk are biomedical fMRI data and long-term climate data, both having dimensionalities in the tens of thousands. Recent results are shown on brain activations to stimuli and on climate patterns.