ELMo Nau'in Yanayi
ELMo (Embeddings from Language Models) wani ci gaba ne na 2018 wanda ya ba kowace kalma wakilcin da aka tsara ta hanyar jimla, don haka 'banki' a cikin 'bankin kogi' ya bambanta da 'banki' a cikin 'bankin ajiya'. Ya nuna alamar canji daga madaidaicin kalmomin vectors zuwa NLP mai sane da mahallin.
Zurfafa nutsewa
ELMo, wanda Allen Institute for AI masu bincike ya gabatar (Peters et al., 2018), yana samar da wakilcin kalmomi ta hanyar gudanar da jumla ta hanyar ƙirar harshe mai zurfi na LSTM mai zurfi wanda aka horar da shi a kan wani biliyan-kalmar corpus. Ba kamar Word2Vec ko GloVe ba, waɗanda ke ba da ƙayyadaddun vector guda ɗaya a kowace kalma, ELMo tana ƙididdige sabon vector don kowane abin da ya faru dangane da mahallin kewaye. Mahimmanci, ELMo yana haɗa duk yadudduka na LSTM na ciki ta hanyar koyo, takamaiman ma'auni na ɗawainiya maimakon amfani da saman saman kawai. Ƙananan yadudduka suna yin kama da syntax (bangaren-magana, tsari) yayin da manyan yadudduka ke ɗaukar ma'ana da ma'anar kalma. Ƙara ELMo zuwa samfuran da ake da su ya haifar da babban fa'ida a cikin ayyuka masu ƙima guda shida, gami da amsa tambaya, nazarin ra'ayi, da sanin mahaɗan mai suna.
Fahimtar Fasaha
ELMo ta tattara LSTMs guda biyu: samfurin harshe na gaba yana tsinkayar kalma ta gaba da baya wanda ke tsinkayar kalmar da ta gabata, kowanne sama da abubuwan shigar CNN na matakin hali (don haka yana sarrafa kalmomin da ba a gani). Don aiki na ƙasa, ELMo yana rushe wakilcin Layer ta amfani da ma'aunin nauyi na softmax-na al'ada tare da scalar, duk an koya yayin daidaitawa. Wannan yana nufin kowane ɗawainiya zai iya yanke shawarar adadin syntactic da siginar tamani da yake so daga biLM da aka rigaya daskararre.
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 ELMo Contextual Embddings
Babban ra'ayin ELMo, wakilcin mahallin daga horon ƙirar harshe, ya zama tushen tushe, amma tsarin gine-ginen LSTM na yau da kullun ya kasance cikin sauri ya rufe shi da samfuran masu canji kamar BERT a ƙarshen 2018, waɗanda ke karanta jimlar jimloli a layi ɗaya da sikeli mafi kyau. A yau ELMo galibi yana da mahimmancin tarihi da ilimi, kodayake sarrafa shigarwar hali-CNN da ra'ayoyi masu nauyi har yanzu suna tasiri na musamman aikin haɗawa a cikin ƙananan albarkatu da wadatattun harsuna.
Aiwatar da Gaskiyar Duniya
Haɓaka tsarin tantance mahaɗan mai suna waɗanda dole ne su faɗi ko 'Washington' na nufin mutum, jiha, ko birni dangane da kalmomin da ke kewaye.
Ƙarfafa nazarin ra'ayi ta hanyar ɗaukar cewa 'mara lafiya' yana nufin mara kyau a cikin 'Ina jin rashin lafiya' amma tabbatacce a cikin ɓangarorin 'wannan ba shi da lafiya'
Haɓaka tsarin amsa tambayoyi akan ma'aunin SQuAD ta hanyar ciyar da abubuwan da suka dace a cikin mai karatu.
Rarraba ma'anar kalma a cikin fassarar inji don haka kalmomi masu yawa kamar 'tsika' suna fassara daidai abin da aka bayar
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 ELMo Contextual Embeddings 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
Haɗin Kalma
Tambayoyin da ake yawan yi
What is ELMo Contextual Embeddings?
ELMo (Embeddings from Language Models) wani ci gaba ne na 2018 wanda ya ba kowace kalma wakilcin da aka tsara ta hanyar jimla, don haka 'banki' a cikin 'bankin kogi' ya bambanta da 'banki' a cikin 'bankin ajiya'. Ya nuna alamar canji daga madaidaicin kalmomin vectors zuwa NLP mai sane da mahallin.
Menene babban bambanci tsakanin ELMo da abubuwan da aka shigar a baya kamar Word2Vec?
ELMo yana samar da abubuwan haɗawa da mahallin mahallin: vector don kalma yana canzawa dangane da jimlar da ke kewaye, sabanin madaidaicin vector guda ɗaya na Word2Vec a kowace kalma.
Wane gine-ginen jijiya ELMo ke amfani da shi don karanta rubutu?
An gina ELMo a kan LSTM mai zurfi mai zurfi da aka horar da shi azaman ƙirar harshe, tsarin sarrafawa akai-akai.
Ta yaya ELMo ke haɗa yadudduka na ciki don aiki na ƙasa?
ELMo yana koyon takamaiman ma'auni na softmax-na al'ada na ɗawainiya don haɗa duk yadudduka na biLM, barin kowane ɗawainiya ya jaddada ma'auni ko ma'ana kamar yadda ake buƙata.
Menene ELMo ke amfani dashi azaman raka'o'in shigarwarta don sarrafa kalmomin da ba a gani?
ELMo tana gina abubuwan shigar da kalmomi daga CNN-matakin hali, don haka yana iya wakiltar kalmomin da bai taɓa gani ba yayin horo.
Kusan yaushe aka gabatar da ELMo kuma ta wa?
An buga ELMo a cikin 2018 ta Peters et al. a Cibiyar Allen don AI (AI2).