Dropout na Stochastic Regularization
Dropout bụ aghụghọ a na-emezigharị nke na-agbanyụ obere neurons n'enweghị ihe ọ bụla n'oge usoro ọzụzụ ọ bụla, na-amanye netwọk iji wuo ihe nnọchianya siri ike.
Nchịkọta
It became one of the most influential techniques for fighting overfitting in deep learning.
Ime miri emi
N'ịbụ nke ndị otu Hinton webatara gburugburu 2012, nkwụsị nkwụsị na-ekwu maka adịghị ike nke nnukwu netwọk: neurons nwere ike ịmekọrịta, na-amụta iji dozie mmejọ ibe ha n'ụzọ na-arụ ọrụ naanị na data ọzụzụ. Na ngafe ọ bụla na-aga n'ihu n'oge ọzụzụ, nkwụsịtụ na-edobe mmepụta neuron ọ bụla ka ọ bụrụ efu yana ụfọdụ ihe nwere ike ime p (na-abụkarị 0.5 na nnukwu akwa akwa). N'ihi na neuron ọ bụla nwere ike ịla n'iyi, netwọkụ enweghị ike ịdabere na mmekọ na-esighị ike ma gbasaa ozi bara uru n'ọtụtụ nkeji. Nke a na-eme dị ka ọzụzụ nnukwu mkpokọta netwọkụ dị gịrịgịrị na-ekekọrịta ibu. N'oge ule, a na-agbanyụ nkwụsị ma jiri netwọk zuru ezu mee ihe, na-arụ ọrụ na-arụ ọrụ nke mere na mmepụta a na-atụ anya dakọtara na ọzụzụ. Nsonaazụ a na-adịkarị mma n'ozuzu ya ma ọ bụrụ na ọ na-efu ọzụzụ dị ogologo ogologo oge.
Nghọta nka nka
N'oge ọzụzụ, a na-edobe ngalaba nke ọ bụla nwere ike (1 minus p) site na nkpuchi ọnụọgụ abụọ na-enweghị usoro, yabụ, a na-enyocha obere netwọkụ dị iche iche nke ọ bụla. Usoro ọgbara ọhụrụ na-eji mwepu tụgharịrị: A na-ekewa ọrụ ndị dị ndụ site na (1 minus p) n'oge ụgbọ oloko, yabụ na ọ dịghị mkpa ọ bụla mkpali na ntinye. Nke a randomness injecting mkpọtụ na-akụda imekọ ihe ọnụ na ihe dị ka nkezi n'elu ihe exponential ọnụ ọgụgụ nke ibu-arọ networks, a ọnụ ala ụdị ensembling.
Mmetụta atụmatụ
Mkpebi doro anya
Ọ na-enyere gị aka ikewapụta nkwupụta ọrụ aka doro anya na asụsụ ahịa.
Ọnụ ego na mmefu ego
Ị nwere ike ịjụ ajụjụ mmejuputa iwu ka mma tupu itinye ego ma ọ bụ oge.
Team na usoro ọrụ
Ndị otu nwere nghọta na-eme ka ngwaahịa, amụma na mkpebi mmụta ka mma.
Ọdịnihu nke Dropout na Regularization Stochastic
Na netwọk ọhụụ convolutional, ogbe normalization achụpụla nke ukwuu ọkọlọtọ dropout, ma variants na-eme nke ọma n'ebe ọzọ: transformers na-etinye dropout na nlebara anya na nri-n'ígwé n'ihu, na DropPath (stochastic omimi) tụfuo dum ihe fọdụrụ blocks. A na-eji nkwụsị nke Monte Carlo, nke na-eme ka nkwụsị akwụkwọ na-arụsi ọrụ ike na ntinye aka, na-atụle ejighị n'aka ihe nlereanya. Na-atụ anya nhazigharị stochastic ka ọ ga-abụ ngwa ọrụ na-agbanwe agbanwe, emebere n'otu ụkpụrụ ụlọ karịa otu uzommeputa edoziri.
Mmejuputa n'ezie n'ụwa
Na-agbakwụnye oyi akwa Dropout nwere p gburugburu 0.5 n'etiti oke oyiri nke onyonyo ma ọ bụ nhazi ederede na PyTorch ma ọ bụ Keras
Ụdị ihe ngbanwe na-etinye nkwụsịtụ na nha nlebara anya na ntinye aka n'ihu n'oge ọzụzụ ọzụzụ
Ọpụpụ Monte Carlo, ebe nkwụsị na-anọ na ntinye aka iji wepụta atụmatụ ejighị n'aka maka amụma ahụike ma ọ bụ nchekwa dị mkpa.
Stochastic omimi (DropPath) na-awụpụ ihe mgbochi na-enweghị usoro iji hazie netwọkụ miri emi dị ka ResNets na ndị ntụgharị ọhụụ.
Ihe ize ndụ & okporo ụzọ nche
Otu dị iche iche nwere ike iji otu okwu ahụ mee ihe n'ụzọ dị iche, yabụ kọwapụta oge n'oge.
Ihe nrịbama nwere ike ịdị ike ebe arụmọrụ ụwa na-adaghị adaba.
Ileghara ogo data na atụmatụ nyocha anya na-emepụtakarị nsonaazụ na-adịghị mma.
Map mmejuputa
Malite na nkọwa asụsụ dị larịị nke nsonaazụ ịchọrọ.
Họrọ otu metrik ịga nke ọma na otu ọnọdụ ọdịda tupu nnwale.
Gbaa obere onye na-anya ụgbọ elu nwere data nnọchite anya, ọ bụghị ihe ngosi ngosi na-egbu maramara.
Detuo ebe Dropout na Stochastic Regularization na-enyere aka yana ebe ụzọ dị mfe ka mma.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Stochastic gradient mgbada nwere oge
Ajụjụ a na-ajụkarị
What is Dropout and Stochastic Regularization?
Dropout bụ aghụghọ a na-emezigharị nke na-agbanyụ obere neurons n'enweghị ihe ọ bụla n'oge usoro ọzụzụ ọ bụla, na-amanye netwọk iji wuo ihe nnọchianya siri ike. Ọ ghọrọ otu n'ime usoro kachasị emetụta maka ịlụ ọgụ imebiga ihe ókè na mmụta miri emi.
Kedu ihe nkwụsị na-eme n'oge ọzụzụ?
Mwepu n'enweghị ihe ọ bụla na-eme ka neuron ọ bụla nwee ohere p n'oge ọzụzụ, ya mere, a na-eji netwọọdụ dị mkpa dị iche iche eme ihe nke ọ bụla.
Kedu ihe kpatara nkwụsị nkwụsị na-eme ka nchịkọta zuru ezu?
Site n'iwepụ nkeji na-enweghị usoro, nkwụsị nkwụsị na-akwụsị neurons ịmepụta mmekọrịta na-esighị ike nke na-arụ ọrụ na data ọzụzụ, na-eme ka ike sie ike.
Kedu ihe na-eme nkwụsị n'oge ule (ntụgharị)?
N'ịtụle netwọkụ zuru ezu na-agba ọsọ yana ndị nwere nkwarụ na-apụ apụ, na arụ ọrụ na-abawanye ka nsonaazụ a tụrụ anya dabara na nkesa ọzụzụ.
Ọnụego nkwụsị nke p = 0.5 na oyi akwa pụtara ihe n'oge ọzụzụ?
Na p = 0.5 nke ọ bụla neuron nwere ohere 50 pasent nke ịbụ efu efu, ya mere, na nkezi, ihe dị ka ọkara na-adaba n'otu ụzọ gafere.
Kedu ihe a na-akọwakarị nkwụsị nkwụsị dị ka ihe ruru?
Ịtụle obere netwọkụ dị iche iche nke ọ bụla na-adaba na nkezi karịa ọnụọgụ netwọọdụ dị arọ, ụdị nchịkọta ọnụ ala.