You end up with different weights when using different random initialization, but with modern techniques the behavior of the resulting model is not really disti

You end up with different weights when using different random initialization, but with modern techniques the behavior of the resulting model is not really distinct. Back in the image-recognition days it was like +/- 0.5% accuracy. If you imagine you're descending in a billion-parameter space, you will always have a negative gradient to follow in some dimension: local minima frequency goes down rapidly with (independent) dimension count.