Is this the same as uniform SO(N) sampling? I had to do something similar like this for some deep learning training method I was working on. I gave up