Principle Component Analysis (PCA)
Given L data points
in R^n (z_l, l=1,2,...,L),
the problem is to find an n-m dimensional
hyperplane that best represents the data points. Mathematically, the
problem is to find an mxn matrix A
whose rows are orthonormal to each other and an n vector a
that minimizes sum_l || A(z_l - a) ||^2.
The following model is a robust version of the PCA model.
That is, it minimizes the sum of the Euclidean distances rather than the sum
of their squares.