Exploration of high dimensional biomedical datasets
with low-distortion embeddings

The emergence of high-dimensional datasets in biology and medicine creates
a need for developing methods of analysis that can discover the structure
and intrinsic organization of these datasets.  We have developed new
methods for constructing, in an unsupervised way, a new low dimensional
parametrization of the dataset. Our approach relies on the assumption that
the signals of interest can be described by a small number of parameters,
in comparison to the large number of degrees of freedom of the original
dataset. We illustrate our approach with several brain imaging datasets
(EEG and fMRI).