Multidimensional Scaling (MDS)

given: - a set of n objects

- the dissimilarities dij between them

find: points on the plane whose distances dij are as close as possible to the dij

minimize:

? missing values: only ~20 distances/object are needed

[Kruskal 1964]