I think this is a <i>very</i> elaborate joke, though I was always surprised by some linear algebra algorithms that had a 'cooking recipe' feeling (looking at you svd) of 'loop enough time to get better results'...
When your operations are statistical in nature, as Approximate Computing is, so then these operations themselves are different sampling methods.<p>If the for loop is calculating a sum, but that value only needs to be over a threshold, there is no reason to add the whole thing, you need a proper sample size and distribution and that it what Loop Perforation is giving you in this case.