AI yana da ikon inganta inganci a sassa daban-daban kamar kasuwanci da kiwon lafiya. Koyaya, rashin bayanin yana hana mu dogara ga yin amfani da shi don yanke shawara.
Ya kamata mu amince da hukuncin algorithm?
Yana da mahimmanci ga masu yanke shawara a kowace masana'antu don fahimtar iyakoki da yuwuwar son zuciya samfurin koyo na inji. Don tabbatar da cewa waɗannan samfuran suna yin aiki kamar yadda aka yi niyya, fitar da kowane tsarin AI ya kamata a bayyana shi ga ɗan adam.
A cikin wannan labarin, zamuyi magana game da mahimmancin bayyanawa a cikin AI. Za mu ba da taƙaitaccen bayani game da nau'ikan hanyoyin da ake amfani da su don samun bayani daga ƙirar koyon injin.
Menene Bayyana AI?
Mai iya bayyanawa wucin gadi hankali ko XAI yana nufin dabaru da hanyoyin da ake amfani da su don ba da damar mutane su fahimci yadda ƙirar koyon inji ke isa ga takamaiman kayan aiki.
Shahararrun mutane da yawa mashin ilmin lissafi aiki kamar "akwatin baki". A cikin ilimin injin, Black box algorithms koma zuwa ƙirar ML inda ba zai yiwu a tabbatar da yadda wani takamaiman shigarwa ke kaiwa ga takamaiman fitarwa ba. Ko da mai haɓaka AI ba zai iya yin cikakken bayanin yadda algorithm ke aiki ba.
Misali, ana amfani da algorithms mai zurfi na koyo neural networks don gano alamu daga ton na bayanai. Kodayake masu bincike na AI da masu haɓakawa sun fahimci yadda hanyoyin sadarwa na jijiyoyi ke aiki daga mahangar fasaha, ba ma za su iya yin cikakken bayanin yadda hanyar sadarwa ta jijiyoyi ta fito da wani sakamako na musamman ba.
Wasu cibiyoyin sadarwa na jijiyoyi suna ɗaukar miliyoyin sigogi waɗanda duk suke aiki tare don dawo da sakamako na ƙarshe.
A cikin yanayin da yanke shawara ke da mahimmanci, rashin yin bayani zai iya zama matsala.
Me Yasa Bayyana Mahimmanci
Bayyanawa yana ba da haske kan yadda samfuran ke yanke shawara. Kasuwancin da ke shirin daidaita AI don yanke shawara dole ne su tantance ko AI ta yi amfani da shigar da ta dace don isa ga mafi kyawun yanke shawara.
Samfuran da ba a bayyana su ba matsala ce a masana'antu da yawa. Misali, idan kamfani zai yi amfani da algorithm don yanke shawarar daukar aiki, zai zama mafi kyawun sha'awar kowa don samun fayyace yadda algorithm ke yanke shawarar kin mai nema.
Wani filin inda zurfin ilmantarwa Ana amfani da algorithms akai-akai yana cikin kiwon lafiya. A cikin lokuta inda algorithms ke ƙoƙarin gano yiwuwar alamun ciwon daji, yana da mahimmanci ga likitoci su fahimci yadda samfurin ya isa wani ganewar asali. Ana buƙatar wasu matakin bayanin don ƙwararru don cin gajiyar AI kuma ba su bi shi a makance ba
Bayanin Algorithms AI Mai Bayyanawa
Mai bayyanawa AI Algorithms ya fada cikin manyan rukuni biyu: ƙirar fassara da bayani da kuma bayanin hoc.
Samfuran da za a iya fassara kansu
Samfuran da za a iya fassara kansu Algorithms ne wanda mutum zai iya karantawa da fassara kai tsaye. A wannan yanayin, samfurin kanta shine bayani.
Wasu samfurori na yau da kullun da ake iya fassara kansu sun haɗa da bishiyar yanke shawara da ƙirar koma baya.
Misali, bari mu yi la’akari da tsarin koma baya na layi wanda ke hasashen farashin gida. Juyin layi na layi yana nufin cewa tare da wasu ƙima x, za mu iya hango ƙimar ƙimar mu ta y ta amfani da takamaiman aikin layi f.
A ce samfurin mu yana amfani da girman yawa a matsayin babban shigarwa don ƙayyade farashin gida. Yin amfani da koma bayan layi, mun sami damar fito da aikin y = 5000 * x inda x shine adadin ƙafar murabba'i ko girman yawa.
Wannan samfurin mutum ne wanda za'a iya karantawa kuma yana da cikakken haske.
Bayanin Bayan Hoc
Bayanin bayan-hoc rukuni ne na algorithms da dabaru waɗanda za a iya amfani da su don ƙara bayyanawa ga sauran algorithms.
Yawancin fasahar bayani bayan hoc ba sa buƙatar fahimtar yadda algorithm ke aiki. Mai amfani kawai yana buƙatar ƙididdige shigarwar da sakamakon fitarwa na algorithm na manufa.
An kara raba waɗannan bayanai zuwa nau'i biyu: bayanin gida da bayanin duniya.
Bayanin gida yana nufin yin bayanin ɓangaren abubuwan shigarwa. Misali, da aka ba da fitowar ta musamman, bayanin gida zai iya nuna waɗanne sigogi ne suka ba da gudummawar yanke shawarar.
Bayanin duniya yana nufin samar da bayanan bayan-hoc na dukkan algorithm. Irin wannan bayanin yawanci ya fi wuya a yi. Algorithms suna da rikitarwa kuma ƙila a sami sigogi marasa ƙima waɗanda ke da mahimmanci wajen cimma sakamako na ƙarshe.
Misalai na Algorithms Bayanin Gida
Daga cikin fasahohin da yawa da aka yi amfani da su don cimma XAI, algorithms da aka yi amfani da su don bayanin gida shine abin da yawancin masu bincike ke mayar da hankali a kai.
A cikin wannan sashe, za mu kalli wasu shahararrun algorithms bayanin gida da yadda kowannensu yake aiki.
LIME
LIME (Model Fassarar Gida-Mai Bayanin Agnostic) Algorithm ne wanda zai iya bayyana hasashen duk wani algorithm na koyon injin.
Kamar yadda sunan ke nunawa, LIME samfurin-agnostic ne. Wannan yana nufin LIME na iya aiki don kowane nau'in samfuri. Samfurin kuma ana iya fassara shi a cikin gida, ma'ana za mu iya bayyana samfurin ta amfani da sakamako na gida maimakon bayyana duka samfurin.
Ko da samfurin da ake bayyanawa akwatin baƙar fata ne, LIME yana ƙirƙirar ƙirar layi na gida a kusa da maki kusa da wani matsayi.
Lime yana ba da samfurin layi wanda ke ƙayyadaddun ƙirar a cikin kusancin tsinkaya amma ba lallai ba ne a duniya.
Kuna iya ƙarin koyo game da wannan algorithm ta ziyartar wannan ma'ajiyar buɗaɗɗen tushe.
SHAFI
Shapley Additive Explanations (SHAP) hanya ce ta bayyana hasashen mutum ɗaya. Don fahimtar yadda SHAP ke aiki, dole ne mu bayyana menene ƙimar Shapley.
Ƙimar Shapley ra'ayi ne a ka'idar wasan da ta ƙunshi sanya "daraja" ga kowane ɗan wasa a wasan. Ana rarraba wannan ta yadda ƙimar da aka ba kowane ɗan wasa ta dogara ne akan gudummawar ɗan wasan ga wasan.
Ta yaya za mu nema ka'idar wasan don koyon injin model?
A ɗauka cewa kowane fasali a cikin ƙirar mu shine "mai kunnawa" kuma "wasan" shine aikin da ke fitar da tsinkaya.
Hanyar SHAP ta ƙirƙira samfurin layi mai nauyi wanda ke ba da ƙimar Shapley zuwa fasali daban-daban. Siffofin da manyan dabi'un Shapley suna da tasiri mafi girma akan sakamakon samfurin yayin da siffofi tare da ƙananan ƙimar Shapley ba su da tasiri.
Kammalawa
Bayanin AI yana da mahimmanci ba kawai don tabbatar da daidaito da lissafin tsarin AI ba, har ma don gina dogaro ga fasahar AI gabaɗaya.
Har yanzu akwai yalwar bincike da za a yi a fannin AI, amma akwai wasu hanyoyi masu ban sha'awa waɗanda za su iya taimaka mana mu fahimci hadadden tsarin AI na baki wanda aka riga aka yi amfani da shi a yau.
Tare da ƙarin bincike da ci gaba, za mu iya fatan gina tsarin AI wanda ya fi dacewa da sauƙin fahimta. A halin yanzu, 'yan kasuwa da masana a fannoni kamar kiwon lafiya yakamata su san iyakokin bayanin AI.
Leave a Reply