\^^M{
  \frametitle{Conclusion}

  Some machine learning applications involve an $M \times N$ array $A$ of
  data, were $N$ counts the number of data instances, and $M$ measures
  the size of each data item.
  \vskip 0.1in
  The array $A$ can be exactly represented by the full SVD, $A = U * D * V'$
  \vskip 0.1in
  Often, the data items have a great deal of similarity and correlation.
  Then the entries of the $D$ matrix will vary greatly in size.
  \vskip 0.1in
  Then the information in $A$ can be well approximated by a low rank
  approximation using only the first $K$ columns of the SVD factors.
}