Multidimensional Scaling

Multidimensional scaling (MDS) takes input a $n\times n$ matrix of distances $D$ where $D_{ij}$ denotes the target distance between entity $i$ and entity $j$. It produces an $n\times p$ matrix $X$ where the $i$th row is the position in p-dimensional latent space. MDS transforms the pairwise distance matrix D into a similarity matrix ${\tilde {D}}$ using linear transformations

External Link