Presentation - MIV
Presentation - MIV
Presentation - MIV
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
collet@lsiit.u-strasbg.fr<br />
iAstro Workshop - Nice Observatory<br />
16/17 October 2003<br />
Hughe phenomenon<br />
Inherent sparsity of high dimensional spaces<br />
* in the absence of simplifying assumptions, the amount of data needed<br />
to get reasonably low variance estimators is really high<br />
* N-band observations >> N times more data but in R N space<br />
Dimensionality reduction<br />
* appropriate dimensionality of the reduced feature space<br />
* Important structure in the data actually lies in a much smaller<br />
dimensional space, and will therefore try to reduce the<br />
dimensionality before attempting the classification.<br />
This approach can be successful if the dimensionality reduction/feature<br />
extraction method loses as little relevant information as possible in the<br />
transformation from high-dimensional space to the low-dimensional one.