User talk:DrKlas

February 2023

Hello, DrKlas. We welcome your contributions, but it appears as if your primary purpose on Wikipedia is to add citations to research published by a small group of researchers.

Editing in this way is a violation of the policy against using Wikipedia for promotion and is a form of conflict of interest in Wikipedia – please see WP:SELFCITE and WP:MEDCOI. The editing community considers excessive self-citing to be a form of spamming on Wikipedia (WP:REFSPAM); the edits will be reviewed and the citations removed where it was not appropriate to add them.

Scientific articles should prefer secondary sources to ensure that the information added is trusted by the scientific community.

The editing community highly values expert contributors, so I do hope you will consider contributing more broadly. If you wish to contribute, please first consider citing review articles written by other researchers in your field and which are already highly cited in the literature. If you wish to cite your own research, please start a new section on the article's talk page and add {{request edit}} to ask a volunteer to review whether or not the citation should be added.

MrOllie (talk) 20:26, 10 February 2023 (UTC)[reply]

As robust machine learning is an active area of research, this page is notably outdated. I found there was no discussion of certifiable defenses against adversarial examples, which was the purpose of my additions.

I assume you are not knowledgable on this topic, but the ICML 2021 citation refers to a method for certified training that follow up work in 2023 has demonstrated to be the highest performing approach to date (of course this latter work hasn't been reviewed yet). I would be happy to add additional citations to showcase broader work on the topic, but as this topic is highly relevant to this page, removing the relevant citations provided in my edit because of their resemblance to my user name is asinine.

As for my second edit on membership inference section, the existing source comes from a blog post, while I added a highly-cited academic paper that supports the same claim. While I know many other works on membership inference, most do not specifically emphasize the role of overfitting. While I'm not a fan of pedantry, in the spirit of Wikipedia's customs (which I was not previously familiar with), I will use a request to have this citation added, because the current citation is of low quality.

Tl;dr, the additional bullet point I added to defenses is clearly relevant to the topic, but I will update the revision to include a wider selection of citations. DrKlas (talk) 20:58, 10 February 2023 (UTC)[reply]

To be clear, you make such requests by using the article's associated talk page, not by repeating the same edit with minor variations. MrOllie (talk) 21:55, 10 February 2023 (UTC)[reply]

From my response: "Tl;dr, the additional bullet point I added to defenses is clearly relevant to the topic, but I will update the revision to include a wider selection of citations." This is exactly what I did. Perhaps since you are not familiar with the subject matter you didn't read my revisions closely to see that there are works from several authors, as well as meaningful content additions.

From the policies you sent me: "Using material you have written or published is allowed within reason, but only if it is relevant, conforms to the content policies, including WP:SELFPUB, and is not excessive." The citations given are clearly relevant, and it's hard to see how two citations to distinct works by the same authors is considered "excessive." Cite-spamming is described in the material you linked to as "adding numerous references to work published by yourself and none by other researchers." Clearly many researchers are being included here. You are being needlessly pedantic here. You are assuming that I am an author of the work I am citing, and therefore insisting that I should follow some convoluted process to make relevant additions to this page. DrKlas (talk) 22:10, 10 February 2023 (UTC)[reply]

Perhaps you are not familiar with our guidelines on conflict of interest, which I linked above. They point out that you should not write about your own work, as that tends to lead to promotional language 'State-of-the-art certification methods' and undue weight. This is not needless pedantry, this is protecting a valuable common resource from exploitation. You are assuming that I am an author of the work I am citing If you're not, your username is impersonating that person, which would be grounds for blocking. Are you one of the authors? - MrOllie (talk) 22:23, 10 February 2023 (UTC)[reply]

I'm not confirming or denying that I am an author, but I can assure that my user name is not an impersonation. State-of-the-art is actually an objective term, meaning in this case, achieving the highest certified performance in the literature. Roughly, there are two categories of certification approaches: post hoc, and certified training. The former does not scale (for fundamental reasons) while the latter does; therefore it seems appropriate to point out the direction that has led to the best results to date. Perhaps this could be made more clear by expanding the section, but removing this entire branch of research seems counterproductive to the quality of this valuable common resource.

I would posit as well, that many of the bullet points/citations in this list (and elsewhere on this page) could even more readily be deemed "unduly weighted." Perhaps this is only because they escaped your vigilance, but if it's the case that they are considered acceptable on the grounds that, apparently, they were not added by their respective authors, then this would be a perfect example of what I would consider "needless pedantry." I understand the purpose of the policy, but the end goal shouldn't be to ensure that a specific set of people do or don't make specific additions, but to ensure the quality of the content of the additions. This is why I take issue with having the entire edit removed on the basis that a few of the citations appear to come from an author (I understand before when there was a narrower set of citations). I would be more happy to debate the merits of mentioning specific approaches and expanding the discussion of defenses on this wikipedia page to make it more complete. But I take it your complaint is a procedural one, and not a technical one---moreover, it seems likely that my latest addition (in isolation) would have appeared fine to you, if my user name hadn't borne the same resemblance to the work I cited (because I doubt you would otherwise have a strong opinion on the merits of specific certification approaches), which suggests simply exploiting anonymity would have been rewarded here. Thus, it seems you are gatekeeping experts from expanding on topics they are knowledgable about, which despite your good intentions, is detrimental to the quality of this article. I suppose I will jump through the hoops, but it will be interesting to see if the end result is the same. DrKlas (talk) 23:53, 10 February 2023 (UTC)[reply]

Promotional additions and use of primary sourcing where secondary sources are required are not 'procedural objections', they are cornerstones (WP:NPOV, WP:OR) of Wikipedia's policy. The whole reason we have conflict of interest guidelines is that COI editors are not qualified to judge their own contributions objectively. I am indeed 'gatekeeping' selfpromotional editing, since it is harmful to the encyclopedia. MrOllie (talk) 23:59, 10 February 2023 (UTC)[reply]