Talk:Felix Gers

Request edit on 21 May 2017

I created this version of the article on 13 May 2017‎: https://en.wikipedia.org/w/index.php?title=Felix_Gers&oldid=780225512 . User:Justlettersandnumbers edited it, deleting independent reliable sources on the heavy use of long short-term memory (LSTM) by very famous companies such as Google. True, the references do not mention Gers directly, but they do mention LSTM, demonstrating the notability of this work. I read Wikipedia:Conflict_of_interest. To avoid any potential COI, I'd like to ask User:Justlettersandnumbers (or others) to check this. Could you please undo the changes (or edit this article accordingly)? Thank you! Slowfun (talk) 15:31, 21 May 2017 (UTC)

Slowfun, our article on Gers will need to be based on independent reliable sources that discuss him in depth and in detail (I haven't managed to find any). Articles that are about fields that he has worked in, but that do not not mention him at all, not only do not add to our knowledge of his work, they also tend to suggest that his contribution to those fields is not of central importance. I'm afraid he is of only marginal notability by Wikipedia's rather bizarre standards (any idiot who's kicked a ball for money can have a page, but academics need to meet stringent criteria). I've not nominated the page for deletion because he is a professor in a European university, and that is one of the criteria; but unless some coverage of him can be found, the article is at some risk of deletion nevertheless. Justlettersandnumbers (talk) 14:55, 29 May 2017 (UTC)

Conflict of interest

At least one major contributor to this article appears to have a close personal or professional connection to the topic, and thus to have a conflict of interest. Conflict-of-interest editors are strongly discouraged from editing the article directly, but are always welcome to propose changes on the talk page (i.e., here). You can attract the attention of other editors by putting {{request edit}} (exactly so, with the curly parentheses) at the beginning of your request, or by clicking the link on the lowest yellow notice above. Requests that are not supported by independent reliable sources are unlikely to be accepted.

Please also note that our Terms of Use state that "you must disclose your employer, client, and affiliation with respect to any contribution for which you receive, or expect to receive, compensation." An editor who contributes as part of his or her paid employment is required to disclose that fact. Justlettersandnumbers (talk) 22:32, 29 May 2017 (UTC)

quality of Greff et al. source

"… of which Schmidhuber is one, and all are at the Istituto Dalle Molle di studi sull'Intelligenza Artificiale – not an INDEPENDENT source" -- Justlettersandnumbers. The point is that often Schmidhuber and Hochreiter are given the main credit for LSTMs, whereas in reality they didn't introduce the forget gate, which is crucial to the success of LSTMs. So I don't see how the fact that Schmidhuber is a coauthor of a paper arguing that the forget gate is critical could count against the integrity of the paper when if anything he would have an incentive to minimize its importance. [Edit: but I overlooked the fact that Schmidhuber is a coauthor on the Gers paper.] But in any case, the source is peer-reviewed, and moreover they are not just stating their opinions; they did experiments that showed that removing the forget gate had a calamitous effect on the success of the model. So I don't see what the issue is here. — Preceding unsigned comment added by 178.214.193.186 (talk) 12:53, 30 May 2017 (UTC)

Here is a further independent source: http://proceedings.mlr.press/v37/jozefowicz15.pdf. It has no shared co-authors with the Gers et al. paper, and it emphasizes in a number of places the importance of the forget gate and also of some other observations in that paper. Given that Gers is the primary author on this paper that introduces critical aspects of LSTMs, which are an extremely important technology, I don't see how he could fail to be notable enough to warrant a short Wikipedia page.