One application of compressive sampling is the task of rebuilding an image for which only a few values are known (sampled). Typically litterature works in the L1-norm (see L1Magic).

The experiment I did was to use a NetFlix Prize technique, namely Simon Funk's SVD to rebuild 2D images. So obviously this is on the L2-norm instead of L1 but should give a sufficient approximation just for experimentation anyway. So here instead of rebuilding the image, we rebuild the Singular Value Decomposition (SVD) of a sparse image. For this experiment 75% of the points were thrown away and the Simon Funk's SVD was trained on the remaining 25% of the points, to rebuild the original image.