How to reduce matrix dimension using SVD?

Question

Mathematics

There is a matrix data = NxM (N-features, i.e. rows. M-counts, i.e. columns). For example 100x10. Each count has one hundred attributes.
I need to downsize this matrix to 10x10 because only a few features are significant.
What I do in matlab:

[U, S, V] = svd(data);
  newData = S*V' ;                        // V-транспонированная

S-diagonal, ie:

newData I still get 100x10
What am I doing wrong?

Answer 1

You need to trim columns