Wikipedia:Articles for deletion/Angela Fan

The following discussion is an archived debate of the proposed deletion of the article below. Please do not modify it. Subsequent comments should be made on the appropriate discussion page (such as the article's talk page or in a deletion review). No further edits should be made to this page.

The result was delete. czar 21:58, 16 July 2022 (UTC)[reply]

Angela Fan (edit | talk | history | protect | delete | links | watch | logs | views) – (View log | edits since nomination)

(Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL)

Does not meet WP:SCHOLAR. Ari T. Benchaim (talk) 01:11, 2 July 2022 (UTC)[reply]

• Note: This discussion has been included in the list of Academics and educators-related deletion discussions. Ari T. Benchaim (talk) 01:11, 2 July 2022 (UTC)[reply]
• Keep. Heavy cites, even if you count only first-author work (which, judging from author order, is relevant), are enough I think for WP:PROF#C1. —David Eppstein (talk) 06:23, 2 July 2022 (UTC)[reply]
• Note: This discussion has been included in the list of Women-related deletion discussions. Spiderone^{(Talk to Spider)} 09:32, 2 July 2022 (UTC)[reply]

*Weak keep Seems to meet WP:NPROF based on citations, if possibly slightly WP:TOOSOON. -Kj cheetham (talk) 10:39, 2 July 2022 (UTC)[reply]

• Delete: Does not yet meet WP:PROF. I think that it is WP:TOOSOON. Gusfriend (talk) 04:51, 6 July 2022 (UTC)[reply]
• Comment: Hi, I added new information. Meta AI open sourced on 2022-07-06 the No Language Left Behind (NLLB) model which allows to translate 200 languages and is used by the Wikimedia Foundation.^[1] Angela Fan is the Corresponding Author and one of the 6 research and engineering leaders on the paper.^[2] The model has already been written about by CNET with Fan quoted just after Mark Zuckerberg.^[3] simon (talk) 22:40, 6 July 2022 (UTC)[reply]

Relisted to generate a more thorough discussion and clearer consensus.
Please add new comments below this notice. Thanks, North America¹⁰⁰⁰ 03:39, 9 July 2022 (UTC)[reply]

• Comment: To help with WP:SIGCOV, additional reliable WP:SECONDARY sources since 2022-07-07: New Scientist^[1], CNRS journal^[2] El Mundo^[3]. simon (talk) 14:47, 12 July 2022 (UTC)[reply]

• 1 is behind a paywall. Can you quote the parts about her? 2 is a passing mention and not very independent. 3 is also behind a paywall. Can you quote the pars about her please? CT55555 (talk) 12:22, 13 July 2022 (UTC)[reply]

  Extracts from #1: "Facebook’s owner Meta has created an artificial intelligence model that can translate 204 written languages and has released it under an open source licence so that anyone can use or improve the software." "The model, called No Language Left Behind, supports dozens more text-based languages than Google Translate, which currently works for 133, and Microsoft Translator, which caters for 110." [The model] "doesn’t yet match Wikipedia, which has articles in 327 languages." "Angela Fan at Meta says the company will continue to add other languages. “A lot of those languages are not spoken by a lot of people, and most of them don’t have written form,” she says. “And so even though there are several thousand languages in the world, we estimate that only a few hundred really have standard writing systems. And so we focus on those first. This is just the starting point.” "No Language Left Behind was developed using Meta’s new AI-specific supercomputer, called the AI Research SuperCluster (RSC). The machine is operational, but is still being added to, and when complete it will consist of 16,000 processors. Meta says at that point, it will be the fastest AI-optimised supercomputer in the world, performing at nearly 5 exaflops – meaning it can carry out 5 billion billion operations per second. Fan says that although the AI model can run on less sophisticated hardware, the supercomputer’s power was vital for quickly training and testing iterations of the model." "It can be difficult to qualitatively compare the translated text from AI models, so Meta has also created an update to its existing translation benchmark, called FLORES-200, that evaluates the result of translating more than 40,000 standardised sections of text. The company claims that No Language Left Behind is 44 per cent better than Microsoft’s equivalent DeltaLM research model using this benchmark, and marginally better than Google Translate." simon (talk) 13:55, 13 July 2022 (UTC)[reply]

  Something that contains two quotes from her is not independent (we need reporting that is written about her, without involvement from her) or significant coverage (her being the subject of some paragraphs, she is not the subject of this writing). CT55555 (talk) 14:07, 13 July 2022 (UTC)[reply]

  Extracts from #3: "Ayer miércoles presentaron su primer gran éxito, NLLB-200, un modelo de IA capaz de hablar 200 idiomas, entre los que se encuentran algunos minoritarios oriundos de África, Europa y Asia. De hecho, el sistema está preparado para realizar 25.000 traducciones diarias en todas las apps de Meta, según destaca Zuckerberg." "Por su parte, el propio Zuckerberg aseguraba en una publicación de Facebook que esta tecnología no solo se usará en distintos productos y servicios de Meta como la propia red social e Instagram, sino que se ha implementado en sitios como Wikipedia." "'Miles de millones de personas en todo el mundo no tienen acceso a una tecnología o un servicio de traducción que realmente funcione bien para su idioma', explica Angela Fan, científica investigadora de Meta AI, en un video producido por la compañía. 'Realmente esperamos que la tecnología que estamos desarrollando haga que el metaverso sea inclusivo por diseño', añade la experta." "De igual modo supera a otros traductores actuales como el de Google que solo cuenta con 133 idiomas, además de así mismo, doblando el número de idiomas con los que trabajaba hasta ahora Meta." "'Para dar una idea de la escala del programa, el modelo de 200 idiomas analiza más de 50.000 millones de parámetros. Lo hemos entrenado usando el Research SuperCluster, uno de los superordenadores más rápidos del mundo', subraya Zuckerberg en una publicación colgada ayer en su cuenta de Facebook." simon (talk) 14:07, 13 July 2022 (UTC)[reply]

  My comments for 3 are the same as 1. Sorry. CT55555 (talk) 14:08, 13 July 2022 (UTC)[reply]

  Also important to note that Fan is the corresponding author and one of the 6 lead engineer on the NNLB paper. "Angela Fan†,‡" "‡. Corresponding Author." "†. Research and engineering leadership, equal contribution, alphabetical order" simon (talk) 14:10, 13 July 2022 (UTC)[reply]

• Comment. (Note, I came here because I was invited by the article's creator via a message onto my talk page. While I like to think of myself as a neutral AfD participant, I expect that some see me leaning more towards inclusion. I have a stated interested in BLP articles about women, so presumably this is not canvassing, but I declare it here just in case). I actually was minded to vote delete, I see an absence of significant independent coverage and an unconvincing claim for notability. Then I saw Kj cheetham and David Eppstein !vote keep based on WP:NPROF. So I checked google scholar and see indeed very high citations. But honestly, I'm struggling to accept Google scholar results here, I know they are imperfect and when the subject works at Facebook as an AI expert, I can imagine how easy it might be to manipulate scholar results as an AI expert. So I will be ready to vote !keep based on WP:NPROF if anyone can convince me that the citations are bona fide, but I think when there is doubt, Google scholar is not the optimal source. So can anyone point to reliable citation sources? CT55555 (talk) 14:22, 13 July 2022 (UTC)[reply]
  • Pinging @JoelleJay: who might be able to give better analysis on citations for the field. -Kj cheetham (talk) 14:50, 13 July 2022 (UTC)[reply]

    Thanks for the ping! I'll work on it. I do want to say that this is a field where a 2020 PhD can have ~700 citations on Scopus with top papers coming in at 201, 180, 148, and 115 cites, so we should DEFINITELY not be taking GS citations at face value. JoelleJay (talk) 07:00, 14 July 2022 (UTC)[reply]

    @Kj cheetham

    Ok, so I looked at Fan's 56 coauthors who have >15 papers (this was the median number of papers among all her coauthors). Here are their Scopus citation metrics:

    Total citations: average: 3660, median: 1690, Fan: 1751

    Total papers: 68, 46, 26

    h-index: 20, 18, 12

    Top five papers: 1st: 828, 236, 683; 2nd: 446, 198, 339; 3rd: 290, 164, 230; 4th: 240, 127, 180; 5th: 175, 79, 100

    Publication in this field is unusual: not only are most research papers tied to industry rather than academia (making it difficult to do an "average professor test"), but working at Facebook/Meta AI, Google, Microsoft, etc. predicts very high citations very quickly. Normally when I see a coauthor with only 7 papers but the citations go 104, 55, 53 49, 28, 13, 8 I assume Scopus accidentally split off those papers from their primary profile, so I'll search their name directly. This usually reveals one or more other profiles of the same person (with one seemingly "central" profile where most of their citations are collected), and I'll manually merge them and recalculate their metrics. However, among Fan's colleagues, it's apparently the norm to have a paper count very close to one's h-index only a few years after getting a PhD. It's also reasonably common to have multiple >100-citation papers working as a tech with just a master's or even bachelor's. Someone can get ~3500 citations and 10 100+-citation papers in under 8 years, with 5 of them in just the last four years working at Meta AI.

    All of this is to say Fan's citation profile is very good for her experience level. Netting ~1800 citations and 5 100+ papers in six years is definitely exceptional for a 2019 PhD. But what about compared to someone who got a PhD in 2015 (TC: 4693; P: 29; h: 21; T5: 810, 638, 499, 366, 339), or 2010 (14800; 52; 32; 4454, 3016, 1186, 948, 507), or 2009 (16798; 56; 35; 3959, 1975, 1508, 1401, 735), or 2000 (42674; 152; 63; 6434, 4841, 3483, 3016, 2355)? Does she truly stand out as having significantly greater impact than the average professor (or senior researcher)? She is certainly on a terrific trajectory, but I think it probably be too soon to say she is currently notable in her field (when looking only at her citation profile). JoelleJay (talk) 00:07, 16 July 2022 (UTC)[reply]

    Thank you for that, it's very interesting! I am going to change my !vote from weak keep to weak delete based on your analysis, as I think is a bit WP:TOOSOON. -Kj cheetham (talk) 11:40, 16 July 2022 (UTC)[reply]

  • P.S. I'm not sure how an AI expert would manipulate citations. -Kj cheetham (talk) 14:52, 13 July 2022 (UTC)[reply]
• Weak delete. Does not appear to meet WP:GNG and I've not been persuaded that she meets the academic notability requirements either. CT55555 (talk) 11:48, 16 July 2022 (UTC)[reply]

The above discussion is preserved as an archive of the debate. Please do not modify it. Subsequent comments should be made on the appropriate discussion page (such as the article's talk page or in a deletion review). No further edits should be made to this page.