I believe the author is correct. I've read the paper he's referring to and the code in the pytorch lib, and linked Google code from sibling posts. To me it is extremely impressive to have come up with a simple change to a known algorithm to deal with a recently identified issue. There is a strange, but very human, tendency to discount discoveries which are not novel. As if somehow genius and insight can only occur once. (It is, unfortunately, especially common among academics.) So I hope the author feels pride in being right and knowing enough to come to an excellent conclusion.
