Multidimensional scaling

cosmos 1st September 2017 at 12:31am
Dimensionality reduction

Find lower dimensional embedding such that the square distances in the embedding are as close as possible to the distances derived from a dissimilarity matrix.

http://users.stat.umn.edu/~gary/classes/5401/handouts/34.mds.handout.pdf

Can solve by finding Eigenvectors of distance matrix. In case dissimilarity matrix is Euclidean distance, this is the same as Principal component analysis

https://en.wikipedia.org/wiki/Multidimensional_scaling