统计学习之---多尺度变换 MDS (Multidimensional Scaling)

Multidimensional Scaling (MDS)

The idea is to find a set of k-dimensional continuous variables to representation the original data, which only contains the dis-similarity information. It is used in problems: for a set of dissimilarities between very pair of n terms, find a representation of the iterms in ${\rm I\!R^d}$ . It seeks to produce a lower dimensional representation of the data such that distances between points $i$ and $j$ in the representation, $\delta_{ij}$, are close to the dissimilarities between points, $d_{ij}$, for all $i, j$

The least square MDS

Scaling by majorizing a convex function
1. Start with