Removing duplicate or near-duplicate examples by comparing their vector representations in embedding space.