Mmetụta mgbada okpukpu abụọ
Mmụba okpukpu abụọ bụ nhụbanya na-atụ n'anya na ka ihe nlereanya na-ebuwanye ibu, nnwale nnwale na-ebu ụzọ na-akawanye njọ n'akụkụ 'ọnụ ụzọ mbata' mana ọ na-akawanye mma ọzọ - na-emegide azụmaahịa akwụkwọ ọgụgụ ochie.
Nchịkọta
It matters because it helps explain why enormous, overparameterized neural networks generalize well instead of overfitting.
Ime miri emi
Ọnụọgụ oge gboo na-akuzi usoro nke nwere ụdị U: dịka mgbagwoju anya nlereanya na-ebili, njehie nnwale na-ada, gbadaa ala, wee bilie ka ihe nlereanya ahụ gafere. Mmụba okpukpu abụọ, nke Belkin, Hsu, Ma na Mandal na-ewu ewu na 2019 wee mụọ ya n'ọ̀tụ̀tụ̀ site n'aka OpenAI, na-egosi na usoro ahụ nwere mgbanaka nke abụọ. Nnwale ule kacha elu nke ọma na ọnụ ụzọ interpolation - ebe ihe nlereanya nwere naanị paramita zuru oke iji dabara ebe ọzụzụ ọ bụla kpọmkwem (njehie ọzụzụ efu). Bugharịa nke ahụ n'ime usoro ihe karịrị akarị na njehie nnwale na-ada ọzọ, na-adịkarị n'okpuru ebe ụtọ oge gboo. Otu mmetụta ahụ na-egosi n'ofe nha ụdị, oge ọzụzụ ('epoch-wise' mgbada okpukpu abụọ), na nha dataset. Ọ na-eweghachi egwu ochie na 'ihe ndị ọzọ paramita na-apụtakarị karịa.'
Nghọta nka nka
Na ọnụ ụzọ interpolation enwere otu ngwọta dabara adaba na data ahụ, a na-amanye ya ka ọ bụrụ jagged na ụkpụrụ dị elu, ya mere ọ na-adịkarị mma. N'ime usoro ihe karịrị akarị, ọtụtụ ngwọta efu na-enweghị njedebe dị, na nhụsianya mmụda gradient na-eduga na nke kachasị dị nro, nke kachasị dị ala. Mmasị ahụ maka ndị interpolators dị mgbagwoju anya - ọ bụghị ọnụ ọgụgụ nke onwe ya - bụ ihe na-eme ka mgbada nke abụọ wedata njehie ule.
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 Mmetụta mgbada okpukpu abụọ
Ndị na-eme nchọpụta na-eji mgbanaka okpukpu abụọ na-emezigharị iwu ịgbatị ma họrọ mgbe ha ga-akwụsị ọzụzụ, ebe ọ bụ na 'ịzụ ogologo oge, na-akawanye njọ, mgbe ahụ ka mma' nwere ihe ọ pụtara n'ezie. Na-atụ anya tiori siri ike na-ejikọta ya na nhazi nke ọma, kernel tangent akwara na grokking. N'ezie, nkuzi - ibu na ogologo nwere ike inye aka gafere mpaghara ihe egwu - na-akwadolarị mkpebi ịzụ ụdị ntọala ka ukwuu karịa nke ọma.
Mmejuputa n'ezie n'ụwa
Na-akọwa ihe kpatara ụdị asụsụ dị ijeri ijeri 175 ji gbakọta nke ọma karịa nke ejiri nlezianya gee ntị nke ọma n'agbanyeghị ikike dị ukwuu.
Ịhọrọ ịzụ ọzụzụ gafere ebe nkwụsị nkwado na-akawanye njọ nwa oge, n'ihi na ọgbọ abụọ maara ihe oge ochie na-ebu amụma mgbake ma emechaa.
Na-achọpụta ụdị ọhụhụ nke izi ezi ya dara kpọmkwem mgbe ọnụ ọgụgụ paramita dabara nha ọzụzụ a na-ahazi, wee duzie ya n'ime oke oke.
Na-eme ka a mara mkpebi nha ihe nlere anya na AutoML ka ndị na-eme ya zere mpaghara ụzọ mbata na-esighi ike.
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 Phenomenon okpukpu abụọ na-enyere aka yana ebe ụzọ ndị dị mfe dị 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 Double Descent Phenomenon 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
Ọdịda gradient
Ajụjụ a na-ajụkarị
What is Double Descent Phenomenon?
Mmụba okpukpu abụọ bụ nhụbanya na-atụ n'anya na ka ihe nlereanya na-ebuwanye ibu, nnwale nnwale na-ebu ụzọ na-akawanye njọ n'akụkụ 'ọnụ ụzọ mbata' mana ọ na-akawanye mma ọzọ - na-emegide azụmaahịa akwụkwọ ọgụgụ ochie. Ọ dị mkpa n'ihi na ọ na-enyere aka ịkọwa ihe kpatara nnukwu netwọkụ akwara ozi karịrị akarị na-agbakọta nke ọma kama ịgafe oke.
Ebee ka njehie ule na-adịkarị elu n'usoro mgbada okpukpu abụọ?
Mmasị mperi ahụ na-apụta na ọnụ ụzọ interpolation, ebe ihe nlereanya nwere ikike zuru oke iji bute njehie ọzụzụ na efu yana otu ngwọta siri ike.
Kedu ihe na-eme iji nwalee njehie ka ihe nlereanya na-abawanye oke?
Na overparameterized ọchịchị, njehie ule na-agbada nke ugboro abụọ, nke bụ njirimara njirimara nke na-enye aha ya okpukpu abụọ.
Kedu ihe kpatara mgbada gradient na-enyere aka n'usoro ọchịchị karịrị akarị?
N'ọtụtụ ngwọta njehie efu dị, nhụsianya nke mgbada gradient na-eduga na nke kachasị dị nro, nke kachasị dị ala, nke na-eme ka ọ dịkwuo mma.
'Epoch-maara ihe' mgbada okpukpu abụọ na-ezo aka na mmetụta pụtara dị ka ọrụ nke gịnị?
Mmụba okpukpu abụọ nwere ike ime n'akụkụ oge ọzụzụ: njehie ule nwere ike ịka njọ wee kwalite ka ọzụzụ na-aga n'ihu maka oge ndị ọzọ.
Kedu echiche oge gboo na-ama aka mgba mgbada ala abụọ?
Ọ na-emegide akwụkwọ ọgụgụ U-dị ka akụkụ nke na-ekwu na mgbagwoju anya karịa otu isi ihe na-abawanye njehie ule mgbe niile site na nfefe.