Kwiga Ibikorwa byinshi
Kwiga-imirimo myinshi itoza icyitegererezo cyo gukora imirimo myinshi icyarimwe, gusangira ibyerekanwe imbere muri bo.
Incamake
By learning shared structure, each task helps the others, often improving accuracy and data efficiency over training separate models.
Kwibira cyane
Aho kubaka icyitegererezo cyihariye kuri buri gikorwa, kwiga-imirimo myinshi (MTL) ikoresha umugongo usangiwe amashami mumitwe yihariye. Urusobe rwo kwiyobora rwonyine, kurugero, rushobora gusangira kodegisi iyerekwa hanyuma igabanyamo imitwe yo kumenya imodoka, gutandukanya umuhanda, no kugereranya ubujyakuzimu. Ibice bisangiwe byiga ibintu rusange bifite akamaro mumirimo, mugihe buri mutwe wihariye. Ibi bikora nkuburyo bwo kubogama no kubogama: ibimenyetso biva kumurimo umwe bibuza guhagararirwa hamwe, kugabanya gukabya no kunoza rusange, cyane cyane iyo imirimo imwe n'imwe ifite amakuru make. Ikibazo nyamukuru nukuringaniza imirimo - niba umunzani wabo wigihombo cyangwa gradients bivuguruzanya, umurimo umwe urashobora kuganza abandi bakababara, ikibazo cyitwa transfert mbi. Ubuhanga nko kugabanya ibiro, uburemere bushingiye ku gushidikanya, no kubaga buhoro buhoro bigamije gukomeza imirimo ikorana aho guhatana.
Ubushishozi
Intego rusange ni igiteranyo kiremereye cyo gutakaza igikorwa, L = Σ wᵢ Lᵢ, no guhitamo uburemere wᵢ ni ngombwa kuko imirimo itandukanye mubipimo no mubibazo. Kugabana ibipimo bigoye (umutiba rusange, imitwe itandukanye) nuburyo bworoshye kandi busanzwe; kugabana byoroshye bituma moderi zitandukanye zifatanije. Amakimbirane ya gradients mumirimo arashobora guhagarika, kuburyo rero nkuburemere budashidikanywaho (kwiga wᵢ mu buryo bwikora) cyangwa PCGrad (kwerekana ibice bivuguruzanya) bifasha imirimo guhugura hamwe.
Ingaruka z'Ingamba
Igiciro na bije
Ibyemezo byubwubatsi bitwara imikorere nigiciro cyimikorere kumyaka.
Ibyemezo bisobanutse
Ubuhanga bwa tekinike bufasha amakipe guhitamo umurongo ukwiye, ntabwo ari shyashya gusa.
Kugenzura ubuziranenge
Guhitamo neza bya injeniyeri bigabanya ibintu byizewe mubikorwa.
Ejo hazaza h'imyigire myinshi
Kwiga-imirimo myinshi ishimangira icyerekezo cyerekezo rusange. Ururimi runini rwururimi rusanzwe rufite imirimo myinshi - umuyoboro umwe ukora ibisobanuro, incamake, code, na Q&A - kandi sisitemu ya multimodal yongerera ibi inyandiko, amashusho, n'amajwi. Tegereza gukura kwimyubakire ihuriweho hamwe nubuyobozi buhuza ibikorwa byinshi muburyo bumwe, hiyongereyeho uburyo bwikora bwikora-buringaniza no guhuza (nkuko bivanze-by-impuguke) bityo rero kongera imirimo ntibisobanura kongera moderi zitandukanye.
Gushyira mu bikorwa Isi
Kwiyitirira gutwara ibinyabiziga bisangiye icyerekezo kimwe kodegisi yo gutahura ibintu, gutandukanya umurongo, no kugereranya ubujyakuzimu.
Ururimi runini rwerekana ibisobanuro, incamake, amarangamutima, hamwe no gusubiza ibibazo hamwe numuyoboro umwe usangiwe.
Sisitemu yo gutanga ibyifuzo hamwe hamwe guhanura gukanda, kureba igihe, no kugura kugirango uhuze ibikorwa byabakoresha.
Uburyo bwo gufata amashusho yubuvuzi bwerekana icyarimwe ikibyimba, kigabanya imbibi zacyo, kandi kigashyira ubwoko bwacyo muri scan imwe.
Ingaruka & Kurinda
Gutezimbere igipimo kimwe gishobora guhisha intege nke za sisitemu.
Ibikorwa Remezo no kubungabunga akenshi usanga bidahabwa agaciro.
Icyuho cyumutekano no kwitegereza birashobora kwiyongera uko sisitemu igenda igorana.
Igishushanyo mbonera
Sobanura ubukererwe, ubuziranenge, nigiciro cyibiciro mbere yo kubishyira mubikorwa.
Ibipimo byerekana umutwaro ufatika hamwe namakuru yimiterere.
Gukurikirana ibikoresho kubikosa, drift, ningaruka zabakoresha.
Tegura inzira yo gusubiza ibyabaye mbere yo gupima.
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 Multi-Task Learning 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
Kugabana Parameter Ikomeye muri Multi-Task Networks
Ibibazo bikunze kubazwa
What is Multi-Task Learning?
Kwiga-imirimo myinshi itoza icyitegererezo cyo gukora imirimo myinshi icyarimwe, gusangira ibyerekanwe imbere muri bo. Mu kwiga imiterere isangiwe, buri gikorwa gifasha abandi, akenshi kunezeza ukuri no gukora neza amakuru kurenza amahugurwa atandukanye.
Niki gitekerezo cyibanze cyo kwiga imirimo myinshi?
MTL itoza icyitegererezo kimwe kumirimo myinshi kuburyo ibice bisangiwe bifata imiterere ifite akamaro muribyose.
Mubisanzwe bigoye-gusangira urusobe rwa MTL, ni iki gisangiwe niki gitandukanye?
Kugabana ibipimo bikomeye bikoresha umutiba rusange kubintu rusange hamwe numutwe utandukanye wihariye kuri buri gikorwa.
Ni ukubera iki kwiga imirimo myinshi bishobora guteza imbere rusange?
Kwiga imirimo myinshi ibuza ibintu bisangiwe, ikora nkibisanzwe akenshi bifasha amakuru make cyane cyane.
Niki 'kwimura nabi' mukwiga imirimo myinshi?
Amakimbirane ya gradients cyangwa igihombo cyiganje arashobora gutera umurimo umwe gutesha agaciro abandi, bitandukanye no kwimura bifasha.
Nigute igihombo muri rusange gisanzwe muburyo bwo kwiga imirimo myinshi?
Intego mubisanzwe Σ wᵢ Lᵢ, kandi guhitamo uburemere nibyingenzi kuva imirimo itandukanye mubipimo no mubibazo.