LAION da Buɗe Datasets
LAION wata ƙungiyar sa-kai ce ta Jamus wacce ta fito da manyan buɗaɗɗen bayanan rubutu-rubutu, wanda aka fi sani da LAION-5B, wanda ya haɓaka horar da buɗaɗɗen ƙirar ƙira kamar Stable Diffusion.
Dubawa
It matters because it made web-scale multimodal data freely available to researchers outside large corporations.
Zurfafa nutsewa
LAION (Large-Spec-Artificial Intelligence Open Network) ƙungiyar sa-kai ce ta Jamus wacce aka kafa a cikin 2021 don ba da dimokraɗiyya bincike na koyon inji ta hanyar fitar da manyan bayanan buɗaɗɗen bayanai. Fitaccen sanannen fitowarta, LAION-5B, ya ƙunshi kusan nau'ikan rubutun hoto biliyan 5.85 da aka tace daga bayanan gidan yanar gizo na gama gari ta amfani da ƙirar CLIP na OpenAI don kiyaye nau'i-nau'i a inda taken da hoton suka daidaita. Mahimmanci, LAION ba ya daukar nauyin hotunan da kansu; yana rarraba URLs da metadata, don haka masu amfani suna zazzage hotuna daga tushen yanar gizo na asali. Waɗannan bayanan bayanan sun kasance kayan aiki don horar da Stable Diffusion da sauran buɗaɗɗen ƙirar rubutu-zuwa hoto. LAION ta fuskanci babban bincike: a cikin 2023 masu bincike sun sami hanyoyin haɗin kai ga hotunan cin zarafi ba bisa ƙa'ida ba a cikin bayanan, wanda ya sa LAION ta sauke shi, ta tsaftace shi, da sake fitar da sigar mafi aminci, yana nuna haɗarin lalata sikelin yanar gizo mara tacewa.
Fahimtar Fasaha
LAION-5B an gina ta ta hanyar duban Common Crawl don alamun hoton HTML tare da alt-rubutu, sannan ta amfani da CLIP don ƙididdige kamance tsakanin kowane hoto da taken sa. An yi watsi da nau'i-nau'i da ke ƙasa da madaidaicin madaidaicin cosine, don haka kawai madaidaicin madaidaicin rubutun-rubutu ya rage. An raba saitin bayanan ta harshe kuma ya haɗa da CLIP da aka riga aka ƙidaya, yana ba da damar binciken kamanni cikin sauri. Saboda URLs kawai aka adana, ruɓewar hanyar haɗin gwiwa a hankali yana lalata haɓakawa akan lokaci.
Dabarun Tasiri
Dabarun mai siyarwa
Taswirorin hanyoyin tallace-tallace suna yin tasiri ga abubuwan da ƙungiyar ku za ta iya ginawa na gaba.
Kudin da kasafin kuɗi
Sharuɗɗan kasuwanci da zaɓuɓɓukan turawa suna shafar farashi da haɗari na dogon lokaci.
Haɗari da aminci
Ƙwararrun kamfani suna siffanta ɓangarorin samfur, yanayin aminci, da buɗewa.
Makomar LAION da Buɗe Datasets
Buɗe saitunan bayanai na multimodal za su fuskanci matsin lamba a kusa da haƙƙin mallaka, yarda, da abun ciki mai cutarwa, turawa zuwa mafi ƙarfi tacewa, tarin lasisi-sane, da fita rajista. Sake sakin LAION na tsaftataccen saitin bayanai yana nuna alamar canji zuwa duba lafiya azaman matakin tsoho. Yi tsammanin ƙarin bayanan roba ko lasisi, ƙa'idodin tabbatarwa, da gano kayan aiki. Tashin hankali tsakanin buɗaɗɗen damar yin amfani da ƙananan dakunan gwaje-gwaje da haƙƙin doka da ƙa'ida na bayanan da aka goge na yanar gizo zai ayyana mataki na gaba na ginin bayanan.
Aiwatar da Gaskiyar Duniya
Horar da samfuran buɗaɗɗen rubutu-zuwa hoto kamar Stable Diffusion akan biliyoyin nau'ikan taken hoto
Maido da rubutu-style na CLIP na gini da ƙima da tsarin rarrabuwar sifili.
Binciken son zuciya na saitin bayanai, amincin abun ciki, da tabbatar da bayanan a sikelin yanar gizo
Tace juzu'i ta harshe, ƙudiri, ko ma'auni mai kyau don ƙirƙirar keɓaɓɓen saitin bayanai masu kyau
Hatsari & Tsare-tsare
Sanarwar ƙaddamarwa na iya ƙetare kwanciyar hankali a cikin ayyukan samarwa na gaske.
Farashin API ko sauye-sauyen manufofi na iya karya zato cikin dare.
Dogaro mai siyarwa guda ɗaya yana ƙara kulle-kulle da farashin ƙaura.
Taswirar Hanya
Kimanta masu samarwa ta amfani da ayyukan ku da saitin bayanai.
Yi bitar sirri, tsaro, da sharuɗɗan doka kafin haɗin kai.
Kula da tsarin koma baya a cikin samfura ko masu siyarwa.
Saka idanu bayanin kula don haka canje-canjen taswirar hanya kada suyi mamakin ƙungiyoyi.
Ci gaba da Bincike
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the LAION and Open Datasets quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Jagora na gaba
Bude Source AI
Tambayoyin da ake yawan yi
What is LAION and Open Datasets?
LAION wata ƙungiyar sa-kai ce ta Jamus wacce ta fito da manyan buɗaɗɗen bayanan rubutu-rubutu, wanda aka fi sani da LAION-5B, wanda ya haɓaka horar da buɗaɗɗen ƙirar ƙira kamar Stable Diffusion. Yana da mahimmanci saboda ya samar da bayanan multimodal na yanar gizo kyauta ga masu bincike a wajen manyan kamfanoni.
Menene LAION da farko ke rarrabawa a cikin bayanan rubutun sa na hoto?
LAION ba ya daukar nauyin hotuna; yana rarraba URLs da metadata, don haka masu amfani suna ɗaukar hotuna daga tushen gidan yanar gizon su na asali.
Kimanin nau'ikan rubutu nawa nawa ne a cikin LAION-5B?
LAION-5B ya ƙunshi kusan nau'i-nau'i-nau'i-nau'i na hoto biliyan 5.85, don haka '5B' a cikin sunanta.
Wane samfurin LAION yayi amfani dashi don tace nau'ikan rubutu-rubutu don ingancin jeri?
LAION ta yi amfani da CLIP na OpenAI don auna kamanceceniya da taken hoto da watsar da nau'i-nau'i mara kyau.
Wanne fitaccen ƙirar ƙirar ƙira aka horar da ta amfani da bayanan LAION?
Stable Diffusion, buɗaɗɗen samfurin rubutu-zuwa-hoto, an horar da shi akan bayanan rubutun-hoton LAION.
Wace babbar matsala ce masu bincike suka gano a LAION-5B a cikin 2023?
Masu bincike sun gano hanyoyin haɗin kai zuwa kayan cin zarafin yara ba bisa ƙa'ida ba, suna jagorantar LAION don saukar da bayanan bayanan, tsaftace shi, da sake fitar da sigar mafi aminci.