Encoder-Decoder Architectures
Encoder-decoder architectures sun raba samfuri zuwa rabi biyu: ɗaya wanda ke karantawa da matsawa shigarwa cikin ingantaccen wakilci na ciki, da kuma wanda ke haifar da fitarwa daga gare ta.
Dubawa
This design powers translation, summarization, and any task where the input and output are different sequences.
Zurfafa nutsewa
Samfurin mai rikodin rikodin rikodin yana aiwatar da matsala a matakai biyu. Mai rikodin yana karanta gabaɗayan jerin shigarwar (ce, jimlar Turanci) kuma ya juya ta zuwa saitin mahallin mahallin da ke ɗaukar ma'ana. Sa'an nan mai ƙaddamarwa yana samar da jerin abubuwan fitarwa (a ce, Faransanci) alama ɗaya a lokaci ɗaya, yana waiwaya kan abubuwan da ya fitar a baya da kuma wakilcin mai rikodin. Asalin Transformer na 2017 shine mai rikodin rikodin rikodin da aka gina don fassara. Samfura kamar T5 da BART suna amfani da wannan siffa kuma suna tsara kowane ɗawainiya azaman rubutu-in, rubutu-fita. Rarraba yana da ƙarfi saboda mai rikodin yana iya ganin gabaɗayan shigarwar a lokaci ɗaya ( mahallin bidirectional), yayin da mai ƙaddamarwa ke haifar da hagu-zuwa-dama. Wannan ya sa ƙira ta zama dacewa ta yanayi don matsalolin jeri-zuwa-jeri inda tsayin fitarwa da abun ciki ya bambanta da shigarwar.
Fahimtar Fasaha
Encoder yana amfani da hankalin kai-bi-directional, don haka kowace alamar shigarwa tana halartar kowace alamar lokaci guda. Mai ƙididdigewa yana da autoregressive kuma yana amfani da abin rufe fuska, ma'ana kowane matsayi zai iya ganin matsayi na farko kawai don adana tsararru. Haɗin su yana da jan hankali: mai gyara yadudduka yana tambayar ɓoyayyun jihohin ƙarshe na ɓoyayyiyar. Wannan rabuwa yana ƙyale mai rikodin ya gina cikakkiyar fahimta mai zaman kanta na oda yayin da mai yankewa ke yin alama ɗaya a lokaci guda.
Dabarun Tasiri
Gudu da sikelin
Gudun aikin harshe na iya tafiya da sauri ba tare da sadaukar da daidaito ba.
Shiga ku isa
Yana faɗaɗa damar shiga cikin harsuna da salon sadarwa.
Shawarwari masu haske
Ƙungiyoyi za su iya ciyar da ƙarin lokaci akan hukunci yayin da aiki da kai ke sarrafa maimaitawa.
Makomar Encoder-Decoder Architectures
Samfuran mai ƙididdigewa kawai kamar GPT yanzu sun mamaye taɗi na gama-gari saboda ma'auni guda ɗaya yana ɗaukar ayyuka da yawa ta hanyar faɗakarwa. Amma ƙirar ƙira-ƙira-ƙira ta ci gaba inda fahimtar shigarwa da tsararrun fitarwa suka bambanta da gaske: fahimtar magana (Whisper), taƙaitaccen bayani, da tsarin multimodal masu haɗa mahaɗar hangen nesa tare da mai gyara rubutu. Yi tsammanin tsarin gine-gine masu haɗaka waɗanda ke aron fahimtar mahallin mahaɗar don dawo da ƙasa yayin kiyaye sassauƙar dikodi, musamman kamar yadda ƙirar ke haɗa rubutu, sauti, da hotuna.
Aiwatar da Gaskiyar Duniya
Google Fassara da DeepL suna amfani da masu canza ma'anar rikodin rikodin don taswirar jumla a cikin harshe ɗaya zuwa wani.
OpenAI's Whisper yana sanya bayanan sifofi na sauti kuma yana yanke su zuwa rubutun da aka fassara ko aka fassara.
T5 da BART ikon taƙaitawa, tattara dogayen labarai cikin taƙaitaccen taƙaitaccen bayani.
Tsare-tsaren taken hoto suna haɗa mahaɗar hangen nesa tare da mai gyara rubutu don kwatanta hotuna cikin kalmomi.
Hatsari & Tsare-tsare
Abubuwan da aka ruɗe suna iya shigar da rahotanni cikin nutsuwa, kwararar tallafi, ko abubuwan bincike.
Hankali na gaggawa na iya ƙirƙirar sakamako mara daidaituwa a cikin buƙatun iri ɗaya.
Za a iya fallasa bayanan rubutu mai ma'ana idan ikon samun dama yana da rauni.
Taswirar Hanya
Ƙayyade tsarin fitarwa, sautin, da ma'auni masu inganci kafin fitowa.
Amsa a ƙasa tare da amintattun tushe a duk lokacin da daidaito ya shafi mahimmanci.
Ajiye wurin binciken ɗan adam don abubuwan da ake samu masu girma.
Bibiyar tsarin gazawar kuma sake horar da tsokaci ko tafiyar aiki akai-akai.
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 Encoder-Decoder Architectures 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
Cross-Encoders vs Bi-Encoders
Tambayoyin da ake yawan yi
What is Encoder-Decoder Architectures?
Encoder-decoder architectures sun raba samfuri zuwa rabi biyu: ɗaya wanda ke karantawa da matsawa shigarwa cikin ingantaccen wakilci na ciki, da kuma wanda ke haifar da fitarwa daga gare ta. Wannan ƙira tana ba da ikon fassara, taƙaitawa, da kowane ɗawainiya inda shigarwar da fitarwa ke da jeri daban-daban.
Menene babban aikin mai rikodi a cikin ƙirar rikodi-dikodi?
Encoder yana cinye gabaɗayan jerin shigarwar kuma yana samar da mahallin mahallin da ke ɗaukar ma'anarsa, waɗanda mai yankewa ke amfani da shi.
Me yasa dikodar ke amfani da abin rufe fuska (dalilin) hankalin kai?
Masking yana tabbatar da kowane matsayi na fitarwa kawai yana zuwa matsayi na farko, yana adana odar tsarar autoregressive hagu-zuwa-dama.
Wace hanya ce ke haɗa na'ura zuwa wakilcin mai rikodin?
Tsare-tsare-tsare yana ƙyale kowane yanki mai dikodi ya bincika ɓoyayyun jihohin mai rikodin, yana haɗa fahimtar shigarwar zuwa tsara abubuwan fitarwa.
Wanne daga cikin waɗannan ƙirar ƙirar ƙirar encoder-decoder (jeri-zuwa-jeri) gine?
T5 (da BART) ƙirar ƙira-ƙira ce ta gargajiya waɗanda ke tsara ayyuka azaman rubutu-zuwa-rubutu. BERT encoder-kawai kuma GPT-3 dikodi ne kawai.
Me ya sa ƙirar encoder-decoder ya dace da yanayin fassara?
Fassara taswirorin taswirori guda ɗaya zuwa wani daban, don haka karanta tushen gaba ɗaya (encoder) da ƙirƙirar sabon manufa (decoder) ya dace daidai.