Dropout da Stochastic Regularization
Dropout dabara ce ta daidaitawa wacce ke kashe juzu'i na neuron ba tare da izini ba yayin kowane matakin horo, tilasta wa hanyar sadarwar gina sabbin abubuwa masu ƙarfi.
Dubawa
It became one of the most influential techniques for fighting overfitting in deep learning.
Zurfafa nutsewa
Ƙungiya ta Hinton ta gabatar a kusa da 2012, dropout yana magance babban rauni na manyan cibiyoyin sadarwa: neurons na iya daidaitawa, koyan gyara kuskuren juna ta hanyoyin da kawai ke aiki akan bayanan horo. A kan kowane fasinja na gaba yayin horo, ficewa ba da gangan yana saita fitowar kowane neuron zuwa sifili tare da wasu yuwuwar p (sau da yawa 0.5 a cikin yadudduka masu yawa). Saboda duk wani neuron na iya ɓacewa, hanyar sadarwar ba za ta iya dogaro da ƙawance masu rauni ba kuma dole ne ta yada bayanai masu amfani a cikin raka'a da yawa. Wannan yana aiki kamar horar da ɗimbin ɗimbin cibiyoyin cibiyoyin sadarwa waɗanda ke raba nauyi. A lokacin gwaji ana kashe ficewar kuma ana amfani da cikakkiyar hanyar sadarwa, tare da daidaita matakan kunnawa don haka abin da ake sa ran ya dace da horo. Sakamakon yawanci shine mafi kyawun haɓakawa akan farashi na ɗan ɗan lokaci horo.
Fahimtar Fasaha
Yayin horarwa ana kiyaye kowane rukunin tare da yuwuwar (1 debe p) ta hanyar abin rufe fuska na binary, don haka ana yin samfuri daban-daban ƙananan cibiyoyin sadarwa kowane tsari. Tsarin zamani yana amfani da jujjuyawar jujjuyawar: ana rarraba abubuwan kunnawa masu rai ta hanyar (1 debe p) a lokacin jirgin ƙasa, don haka ba a buƙatar sikeli idan aka kwatanta. Wannan bazuwar yana shigar da amo wanda ke hana daidaitawa tare da kusan maƙasudi sama da adadi mai ƙima na ƙananan hanyoyin sadarwa masu nauyi, nau'in haɗaɗɗiyar arha.
Dabarun Tasiri
Shawarwari masu haske
Yana taimaka muku keɓance bayyanannen da'awar fasaha daga harshen talla.
Kudin da kasafin kuɗi
Kuna iya yin mafi kyawun tambayoyin aiwatarwa kafin kashe kuɗi ko lokaci.
Ƙungiya da aikin aiki
Ƙungiyoyin da ke da fahimtar juna suna yin mafi kyawun samfura, manufofi, da yanke shawara na koyo.
Makomar Dropout da Ka'ida ta Stochastic
A cikin hanyoyin sadarwa na hangen nesa na jujjuyawar, daidaitawar batch ya fi mayar da madaidaicin madaidaicin raguwa, amma bambance-bambancen suna bunƙasa a wani wuri: na'urori masu canzawa suna amfani da faduwa zuwa hankali da shimfidar ciyarwa, kuma DropPath (zurfin stochastic) yana sauke duka tubalan. Ficewar Monte Carlo, wanda ke ci gaba da barin barin aiki bisa ga ra'ayi, ana amfani da shi don kimanta rashin tabbas. Yi tsammanin daidaitawa na stochastic ya kasance mai sassauƙan kayan aiki, wanda ya dace da kowane gine-gine maimakon tsayayyen girke-girke guda ɗaya.
Aiwatar da Gaskiyar Duniya
Ƙara Layer Dropout tare da p a kusa da 0.5 tsakanin manyan yadudduka na hoto ko rubutun rubutu a cikin PyTorch ko Keras
Samfuran masu canza canji da ke amfani da faduwa zuwa ma'aunin hankali da kunnawa ciyarwa gaba yayin horo
Ficewar Monte Carlo, inda barin barin ya tsaya akan ƙima don samar da ƙididdiga marasa tabbas don tsinkayar lafiya ko aminci.
Zurfin Stochastic (DropPath) ba da gangan ba yana tsallake ragowar tubalan don daidaita hanyoyin sadarwa masu zurfi kamar ResNets da masu canza hangen nesa.
Hatsari & Tsare-tsare
Ƙungiyoyi daban-daban na iya amfani da kalmar iri ɗaya daban, don haka ayyana iyaka da wuri.
Alamomi na iya yin kama da ƙarfi yayin da aikin zahirin duniya bai yi daidai ba.
Yin watsi da ingancin bayanai da tsare-tsaren kimantawa galibi yana haifar da sakamako mara ƙarfi.
Taswirar Hanya
Fara da ma'anar harshe a sarari na sakamakon da kuke buƙata.
Zaɓi ma'aunin nasara ɗaya da yanayin gazawa ɗaya kafin gwaji.
Gudun ƙaramin matukin jirgi tare da bayanan wakilci, ba saitin demo da aka goge ba.
Daftarin aiki inda Dropout da Stochastic Regularization ke taimakawa kuma inda mafi sauƙi hanyoyin sun fi kyau.
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 Dropout and Stochastic Regularization 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
Saukowar Gradient Stochastic tare da Momentum
Tambayoyin da ake yawan yi
What is Dropout and Stochastic Regularization?
Dropout dabara ce ta daidaitawa wacce ke kashe juzu'i na neuron ba tare da izini ba yayin kowane matakin horo, tilasta wa hanyar sadarwar gina sabbin abubuwa masu ƙarfi. Ya zama ɗaya daga cikin mafi tasiri dabarun yaƙi da wuce gona da iri a cikin zurfin koyo.
Menene barin barin aiki yayin horo?
Dropout ba da gangan ba kowane neuron tare da yuwuwar p yayin horo, don haka ana amfani da hanyar sadarwa daban-daban na kowane mataki.
Me yasa ficewa ke inganta gama gari?
Ta hanyar cire raka'a ba da gangan ba, barin barin na'urar yana hana neurons ƙirƙirar haɗin gwiwa mara ƙarfi wanda ke aiki akan bayanan horo kawai, haɓaka ƙarfi.
Me zai faru da ficewa a lokacin gwaji (fito)?
Bisa la'akari da cikakken hanyar sadarwa tana gudana tare da nakasassu, kuma ana daidaita ayyukan kunnawa don haka abubuwan da ake sa ran su dace da rarraba horo.
Adadin raguwa na p = 0.5 a cikin Layer yana nufin kusan menene lokacin horo?
Tare da p = 0.5 kowane neuron yana da kashi 50 cikin dari na damar zama sifili, don haka a matsakaici kusan rabin ana raguwa ta hanyar wucewa ta gaba.
Menene aka fi bayyana ficewa a matsayin kusanta?
Samfuran wata hanyar sadarwa daban-daban kowane mataki ya kai matsakaicin matsakaicin adadin hanyoyin sadarwa masu nauyi, nau'i na haɗakarwa mai arha.