ELMo Ibishushanyo mbonera
ELMo (Embeddings from Language Models) yari intambwe ya 2018 yahaye buri jambo guhagararirwa ninteruro yaryo, bityo 'banki' muri 'uruzi rwinzuzi' itandukanye na 'banki' muri 'banki yo kuzigama.' Byaranze ihinduka kuva ijambo rihagaze neza kuri NLP.
Kwibira cyane
ELMo, yatangijwe na Allen Institute for AI abashakashatsi ba AI (Peters et al., 2018), itanga ijambo ryerekana gukoresha interuro binyuze mu cyerekezo cyimbitse cy’ururimi rwa LSTM rwahuguwe kuri miliyari y'ijambo. Bitandukanye na Word2Vec cyangwa GloVe, igenera icyerekezo kimwe gihamye kuri buri jambo, ELMo ibara icyerekezo gishya kuri buri kintu cyose gishingiye kumiterere ikikije. Byibanze, ELMo ikomatanya ibice byose byimbere muri LSTM ikoresheje ibyize, uburemere bwihariye bwibikorwa kuruta gukoresha urwego rwo hejuru gusa. Ibice byo hasi bikunda gufata syntax (igice-cy-imvugo, imiterere) mugihe urwego rwo hejuru rufata ibisobanuro nijambo ryumvikana. Ongeraho ELMo kuri moderi zisanzwe byatanze inyungu nini mubikorwa bitandatu, harimo gusubiza ibibazo, gusesengura amarangamutima, no kumenyekanisha ibigo.
Ubushishozi
ELMo itondekanya LSTM ebyiri: icyitegererezo cyururimi rwimbere ruvuga ijambo ritaha hamwe nijambo ryinyuma risobanura ijambo ryabanjirije, buri hejuru yinyuguti-urwego CNN yinjiza (bityo ikora amagambo atagaragara). Kubikorwa byo hasi, ELMo isenya ibice byerekana ikoresheje softmax-isanzwe yuburemere wongeyeho scalar, byose byize mugihe cyo gutunganya neza. Ibi bivuze ko buri gikorwa gishobora guhitamo ingano ya syntactique na signal semantique ishaka kuva biLM yahagaritswe.
Ingaruka z'Ingamba
Umuvuduko n'igipimo
Ururimi rwakazi rushobora kugenda byihuse nta gutamba guhuzagurika.
Kugera no kugera
Yagura uburyo bwindimi nuburyo bwo gutumanaho.
Ibyemezo bisobanutse
Amakipe arashobora kumara umwanya munini murubanza mugihe automatike ikora gusubiramo.
Kazoza ka ELMo Ibishushanyo mbonera
Igitekerezo cyibanze cya ELMo, ibishushanyo mbonera biva mu ndimi-yerekana imvugo, byabaye ishingiro, ariko ubwubatsi bwa LSTM bwagarutsweho vuba na bwangu na moderi ishingiye kuri Transformer nka BERT mu mpera za 2018, isoma interuro zose zingana kandi zingana neza. Muri iki gihe, ELMo ifite akamaro kanini mu mateka no mu burezi, nubwo inyuguti-CNN yinjiza hamwe n'ibitekerezo biremereye biracyafite uruhare mu bikorwa byihariye byo gushira mu ndimi nkeya kandi zikungahaye ku miterere.
Gushyira mu bikorwa Isi
Kunoza uburyo bwo kumenyekanisha ibigo bigomba kumenya niba 'Washington' bivuga umuntu, leta, cyangwa umujyi ukurikije amagambo akikije
Kongera isesengura ry'amarangamutima ufata ko 'urwaye' bisobanura nabi muri 'Ndumva ndwaye' ariko ni byiza mu magambo 'arwaye'
Gutezimbere ibibazo-bisubiza sisitemu kurwego rwa SQuAD mugaburira imiterere-yerekana ibimenyetso byerekanwa mubasomyi
Gutandukanya ijambo ibyumviro mubisobanuro byimashini bityo amagambo menshi cyane nka 'igihingwa' asobanura neza imiterere yatanzwe
Ingaruka & Kurinda
Ibintu bifatika bishobora kwinjiza bucece raporo, gushyigikira imigendekere, cyangwa ibisubizo byubushakashatsi.
Kwihuta byihuse birashobora gukora ibisubizo bidahuye mubisabwa bisa.
Ibyanditswe byumvikana birashobora kugaragara niba kugenzura kugenzura ari ntege.
Igishushanyo mbonera
Sobanura imiterere isohoka, amajwi, hamwe nubuziranenge mbere yo gutangira.
Ibisubizo byibanze hamwe nisoko yizewe igihe cyose ukuri kwingirakamaro.
Komeza kugenzura abantu kugenzura ibisubizo byinshi.
Kurikirana uburyo bwo kunanirwa no kongera imyitozo cyangwa akazi gahoraho.
Komeza Ubushakashatsi
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
Ubuyobozi bukurikira
Ijambo
Ibibazo bikunze kubazwa
What is ELMo Contextual Embeddings?
ELMo (Embeddings from Language Models) yari intambwe ya 2018 yahaye buri jambo guhagararirwa ninteruro yaryo, bityo 'banki' muri 'uruzi rwinzuzi' itandukanye na 'banki' muri 'banki yo kuzigama.' Byaranze ihinduka kuva ijambo rihagaze neza kuri NLP.
Ni irihe tandukaniro nyamukuru riri hagati ya ELMo no gushiramo mbere nka Word2Vec?
ELMo itanga ibisobanuro bifatika: icyerekezo cyijambo rihinduka rishingiye ku nteruro ikikije, bitandukanye na Word2Vec imwe rukumbi ihamye kuri buri jambo.
Ni ubuhe bwubatsi bwubaka ELMo ikoresha mugusoma inyandiko?
ELMo yubatswe kumurongo wimbitse LSTM yatojwe nkicyitegererezo cyururimi, gutunganya inshuro nyinshi.
Nigute ELMo ihuza ibice byayo imbere kumurimo wo hasi?
ELMo yiga imirimo yihariye ya softmax-isanzwe yuburemere kugirango ihuze ibice byose bya biLM, ureke buri gikorwa gishimangira syntax cyangwa semantique nkuko bikenewe.
Niki ELMo ikoresha nkibice byinjiza kugirango ikore amagambo atagaragara?
ELMo yubaka ijambo ryinjiza riva mumiterere-urwego CNN, bityo irashobora kwerekana amagambo itigeze ibona mugihe cyamahugurwa.
Hafi ya ELMo yatangijwe ryari kandi nande?
ELMo yasohowe muri 2018 na Peters n'abandi. mu kigo cya Allen Institute for AI (AI2).