I discovered why Matlab does that. I will explain it.
Matlab tried to make the samples easier to train, that why he deletes the values of the vector, which are either zero or not important to train. But you get at the end, what you would have.
In the attachment, you see that I have in the output vector 60 components but the last 20 are just zeros, what why he deletes them from the output layer and put it back in the output.