gotta admit that while I've discovered a lot of wrong results in papers, it was always by reading the code or reimplementing the method, and never by reading the proofhttps://twitter.com/ra/status/1249711149790244866 …
-
Show this thread
-
of course I've seen a ton of incorrect or spurious proofs. it's just that these flaws have mostly been disconnected from flaws in the things being proved.
1 reply 1 retweet 13 likesShow this thread -
not at all saying that proofs are useless, but I think in a lot of cases they're more useful to the authors than to the readers
8 replies 1 retweet 31 likesShow this thread -
Replying to @johnregehr
This is true, but they are included (I think) so that the reader can check whether the author did enough to be useful. Part of this is a limitation of "peer review only happens over the part in the 12 page pdf".
2 replies 0 retweets 5 likes -
Replying to @samth @johnregehr
For a hotter take: the fact that we manage to mostly include our proofs in the reviewed PDF, but don't include our implementations, mean that our theoretical claims are mostly better-checked and more believable than our empirical ones.
3 replies 0 retweets 7 likes -
Replying to @samth @johnregehr
YEP. See: all the polygon triangulation papers I tried to implement that had impressive proofs but were useless in practice due to not considering floating point error.
1 reply 2 retweets 9 likes -
Replying to @pcwalton @johnregehr
Theoretically-nice but useless in practice is a real phenomenon, but not really the one I'm pointing out. Did the papers claim they had implementations that worked, but actually the implementations didn't work?
1 reply 0 retweets 2 likes -
Replying to @samth @johnregehr
I forget, but I don’t think so, as the most famous result in polygon triangulation is an O(n log n) algorithm that is thought to never been successfully implemented by anyone
1 reply 0 retweets 6 likes -
(Not saying that unimplementable algorithms are useless, BTW—Babai’s graph isomorphism work is impressive because of the P?=NP implications—but I have a hard time getting excited about unimplementable polygon triangulation algorithms.)
1 reply 0 retweets 0 likes -
Replying to @pcwalton @johnregehr
Right, I agree that unimplementable (or useless-in practice) algorithms are an issue. But there's also the issue of "we implemented this thing, here are the results" but actually the implementation is riddled with bugs that invalidate the results.
1 reply 0 retweets 6 likes
I don’t think I’ve seen any of those, but I have to admit I rarely bother trying to get a paper’s implementation to run.
-
-
Replying to @pcwalton @johnregehr
The question is really if you've tried to implement something that they say they implemented, and then it doesn't actually work out.
0 replies 0 retweets 2 likesThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.