Ragowar hanyoyin sadarwa
Residual Networks (ResNets) hanyoyin sadarwa ne masu zurfi waɗanda ke ƙara 'tsalle haɗin kai' barin yadudduka su koyi ƙananan gyare-gyare maimakon cikakken canji.
Dubawa
This simple trick made it possible to train networks hundreds of layers deep, sparking a leap in image recognition accuracy.
Zurfafa nutsewa
Kafin ResNets, tara yadudduka da yawa sun sanya hanyoyin sadarwa suna yin muni, har ma akan bayanan horo, matsala da ake kira lalata. A cikin 2015, Microsoft masu bincike Kaiming He da abokan aiki sun gabatar da ragowar toshe: maimakon tambayar tarin yadudduka don samar da fitowar H(x) kai tsaye, sai suka bar shi ya koyi saura F(x) = H(x) - x, sannan su ƙara ainihin shigarwar x baya ta hanyar gajeriyar hanya. Idan Layer ba a buƙata ba, zai iya koyon yin kome kawai (F(x) = 0). ResNet-152 ya lashe gasar ImageNet ta 2015 tare da kuskuren sama-5 na kusan kashi 3.6, yana bugun ƙididdiga na matakin mutum, kuma gine-ginensa ya zama kashin baya don ganowa, rarrabawa, da kuma hoton likita.
Fahimtar Fasaha
Haɗin tsallakewa yana juya kowane aikin toshe zuwa y = F(x) + x. Lokacin yaɗa baya, gradient yana gudana ta hanyar gajeriyar hanyar ainihi ba ta canzawa, don haka ba zai iya ɓacewa zuwa kusa da sifili ko da a cikin ɗaruruwan yadudduka. Wannan yana kiyaye zurfafa zurfafa horarwa. Gajerun hanyoyin tantancewa ba su ƙara ƙarin sigogi ba; kawai lokacin shigarwa da girman fitarwa sun bambanta yana ɗan ƙaramin tsinkaya (1x1 convolution) yana daidaita ma'auni kafin ƙari.
Dabarun Tasiri
Gudu da sikelin
Kayayyakin AI na iya sarrafa aiki da bincike, ganowa, da ayyuka masu alama a sikelin.
Gina zaɓuɓɓuka
Ƙungiyoyin ƙirƙira za su iya samar da ra'ayoyi cikin sauri tare da ƙarancin bita da hannu.
Ƙungiya da aikin aiki
Ayyuka na iya amfani da siginar hoto da bidiyo waɗanda a baya suke da wahalar aiwatarwa.
Makomar Residual Networks
Haɗin da ya saura yanzu yana kusa da duniya: Masu canza canji, samfuran watsawa, da manyan nau'ikan harshe duk suna amfani da su don daidaita horar da tari mai zurfi. Ana ci gaba da bincike akan bambance-bambancen kamar su ResNets da aka fara kunnawa, hanyoyin haɗin gwiwar ResNeXt, da haɗa ra'ayoyin da suka rage tare da horarwa marasa daidaituwa. Yi tsammanin ainihin ƙa'idar ƙetare-haɗin za ta ci gaba da kasancewa tushen ginin gini, ko da yayin da gine-ginen da ke kewaye da su ke ƙaura daga tsattsauran ra'ayi zuwa hankali da ƙirar ƙira.
Aiwatar da Gaskiyar Duniya
ImageNet rarrabuwar kasusuwan baya (ResNet-50, ResNet-101) da aka yi amfani da su azaman abubuwan da aka riga aka horar da su don canja wurin koyo.
Gano ƙwayar cuta da rauni a cikin rediyo da hotunan cututtukan cututtuka ta amfani da maƙallan tushen ResNet
Gano abu da tsarin rarrabuwa misali kamar Faster R-CNN da Mask R-CNN waɗanda ke amfani da kashin baya na ResNet
Bututun hasashe masu tuƙi da kai waɗanda ke rarraba masu tafiya a ƙasa, motoci, da alamu daga firam ɗin kyamara
Hatsari & Tsare-tsare
Haƙƙoƙin hoto da yarda na iya zama haxarin doka idan ba a fayyace ba.
Ayyukan samfuri na iya bambanta a ko'ina cikin haske, ƙididdiga, da mahalli.
Ƙarya tabbataccen ƙila ba za a iya lura da shi ba sai dai idan an kula da ƙofofin amincewa.
Taswirar Hanya
Ƙayyade ma'auni na karɓa don daidaito, tunowa, da farashi na kuskure.
Gwada tare da bayanan da suka dace da ainihin yanayin samarwa.
Ƙara bita na ɗan adam don ƙarancin amincewa ko tsinkaya mai tasiri.
Bi diddigin ƙirar ƙira kuma sake ingantawa bayan canje-canjen kamara ko saitin bayanai.
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 Residual Networks 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
Siffar hanyoyin sadarwa na Pyramid
Tambayoyin da ake yawan yi
What is Residual Networks?
Residual Networks (ResNets) hanyoyin sadarwa ne masu zurfi waɗanda ke ƙara 'tsalle haɗin kai' barin yadudduka su koyi ƙananan gyare-gyare maimakon cikakken canji. Wannan dabara mai sauƙi ta ba da damar horar da cibiyoyin sadarwa ɗaruruwan yadudduka zurfi, yana haifar da tsalle cikin daidaiton tantance hoto.
Wace matsala saura haɗin haɗin gwiwa ya warware musamman?
Kafin ResNets, ƙara ƙarin yadudduka ya sa daidaito ya ƙasƙanta koda akan bayanan horo. Tsallake hanyoyin haɗin gwiwa sun gyara wannan ta hanyar yin yadudduka cikin sauƙi don ingantawa.
Menene ragowar toshe a haƙiƙa yana ƙididdige shi azaman fitarwa?
Sauran abubuwan toshewa y = F(x) + x, yana ƙara sauran abubuwan da aka koya zuwa shigarwar ta hanyar haɗin tsallakewa.
Me yasa tsallake haɗin gwiwa ke taimakawa gradients yayin horo?
Gajerar hanyar ainihi tana ba da hanya kai tsaye ga gradients don gudana baya baya canzawa, yana hana matsalar ɓarna-girma a cikin rijiyoyi masu zurfi.
Kusan yadudduka nawa ne samfurin ResNet mai nasara daga 2015 ya samu?
ResNet-152, tare da yadudduka 152, ya lashe gasar ImageNet na 2015, yana nuna cewa yanzu ana iya horar da cibiyoyin sadarwa mai zurfi cikin nasara.
Idan ragowar toshe ya koyi F(x) = 0, menene toshe yake yi?
Lokacin da F(x) = 0, abin da ake fitarwa shine x kawai, don haka toshe ya zama taswirar ainihi. Wannan yana sa ƙarin yadudduka mara lahani idan ba a buƙata ba.