Yaraynta Fiican-ka warqabka
Yaraynta Ogsoonaanta-Awareer (SAM) waa hab wanaajin oo aan raadin khasaare hoose laakiin khasaare hoose ee dhammaan xaafad miisaan - ugu yaraan siman.
Dulmar
Flatter minima tend to generalize better, so SAM often improves test accuracy and robustness without changing the model architecture.
quusid qoto dheer
Tababarka caadiga ah wuxuu yareeyaa lumitaanka hal dhibic oo miisaan ah, laakiin laba xal oo leh isla khasaaraha tababarka ayaa u dhaqmi kara si aad u kala duwan: 'fiiqan' ugu yaraan wuxuu ku fadhiyaa dooxada cidhiidhiga ah halkaas oo culeysyada yaryar ay kor u qaadaan khasaaraha, halka ugu yar ee 'flat' uu u dulqaadanayo dhibka oo caadi ahaan si fiican u soo koobaya xogta aan la arki karin. SAM, oo ay soo bandhigeen Google cilmi-baarayaal 2020, ayaa tan si cad u dhigaya. Tallaabo kasta waxa ay marka hore helaysaa qaska miisaanka u dhow (gudahood radius rho ah) kaas oo kor u qaada khasaaraha - deriska kiiskiisa ugu xun - ka dibna cusbooneysiiya miisaankii asalka ahaa si loo yareeyo khasaaraha meeshaas qallafsan. Hadafkan min-max wuxuu u riixayaa hagaajinta xagga gobollada isku midka ah, oo si muuqata u soo bandhigaya guud ahaan ka wanaagsan xagga soocidda sawirka iyo wixii ka dambeeya.
Aragtida Farsamada
Tallaabo kasta oo SAM ah waa laba baas. Marka hore, ku xisaabi gradient ee miisaanka hadda oo qaado tallaabo 'kor u fuul' cabbir cabbir rho jihada gradient si aad u gaarto meesha ugu xun ee u dhow. Marka labaad, ku xisaabi jaan-goynta meesha qallafsan oo isticmaal si aad u cusboonaysiiso miisaankii asalka ahaa. Radius rho ayaa maamula inta ay le'eg tahay xaafad aad ka ilaalinayso. Kharashku waa qiyaastii laba baas oo hore-u-socod ah tallaabo kasta, taas oo labanlaabanaysa xisaabinta - cilladda ugu weyn ee la taaban karo.
Saamaynta Istiraatijiyadeed
Qiimaha iyo miisaaniyada
Go'aamada qaab-dhismeedku waxay horseedaan waxqabadka iyo kharashka hawlgalka sannadaha.
Go'aamo cad
Waxbarashada farsamada waxay ka caawisaa kooxaha inay doortaan xidhmo sax ah, ma aha oo kaliya kan ugu cusub.
Xakamaynta tayada
Doorashooyinka injineernimada ee wanaagsan waxay yareeyaan shilalka la isku halleyn karo ee wax soo saarka.
Mustaqbalka Yaraynta Fiican-Ogaalnimada
SAM waxa ay dhaleen qoys dabagal ah oo ujeedadoodu tahay daciifnimadooda ugu weyn, xisaabinta labanlaabantay: kala duwanaansho hufan sida ESAM, LookSAM, iyo habab dhibaya oo kaliya qayb ka mid ah miisaanka ama dabaqa SAM dhowrkii tillaaboba. La qabsiga SAM (ASAM) wuxuu dib u cabbirayaa raadiyaha si uu u noqdo miisaan-aan kala duwanayn. Cilmi baadhayaashu waxay sii wadaan inay si sax ah uga doodaan sababta flatness ay u caawiso iyo sida loo cabbiro, iyo fikradaha fiiqan ee baraaruga ayaa ku faafaya hagaajinta qaababka luqadaha waaweyn iyo hagaajinta adkeysi u wareejinta qaybinta.
Dhaqangelinta Adduunka-dhabta ah
Kobcinta Transformer Vision iyo saxnaanta ResNet ee ImageNet iyadoo la tababarayo SAM bedelkii SGD cad.
Hagaajinta adkaynta in lagu calaamadiyo buuqa, maadaama minima gurigu ay aad ugu yar tahay inay xafidaan sumadaha kharriban.
Hagaajinta qaababka luqadda hore loo tababaray ee SAM si loo helo xog guud oo wanaagsan oo ku saabsan kaydka hoose ee hoose.
Isticmaalka noocyada ESAM ama LookSAM marka qiimaha labanlaaban ee xisaabinta vanilj SAM uu aad qaali u yahay.
Khatarta & Dariiqyada Ilaalada
Hagaajinta hal bartilmaameed waxay qarin kartaa daciifnimada nidaamka ballaaran.
Kaabayaasha dhaqaalaha iyo dayactirka inta badan waa la dhayalsadaa.
Nabadgelyada iyo daldaloolada u fiirsashada ayaa kori kara marka nidaamyadu noqdaan kuwo aad u adag.
Qorshe Hawleedka Dhaqangelinta
Qeex daahida, tayada, iyo bartilmaameedyada qiimaha ka hor inta aan la hirgelin.
Benchmark marka la eego culeyska dhabta ah iyo xaaladaha xogta.
La socodka qalabka khaladaadka, leexashada, iyo saamaynta isticmaalaha.
U diyaari dib-u-noqoshada iyo dariiqyada jawaab-celinta dhacdada ka hor inta aanad miisaan.
Sii wad Sahaminta
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 Sharpness-Aware Minimization 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
Hagaha xiga
DanseNet iyo Isku xirnaanta cufan
Su'aalaha soo noqnoqda
What is Sharpness-Aware Minimization?
Yaraynta Ogsoonaanta-Awareer (SAM) waa hab wanaajin oo aan raadin khasaare hoose laakiin khasaare hoose ee dhammaan xaafad miisaan - ugu yaraan siman. Flatter minima waxay u janjeertaa inay si ka sii wanaagsan u soo koobto, sidaa darteed SAM waxay inta badan wanaajisaa saxnaanta tijaabada iyo adkeynta iyada oo aan la beddelin qaab dhismeedka moodeelka.
Waa maxay nooca ugu yar ee SAM isku dayo inuu helo?
SAM waxa ay bartimaameeysaa minima flat sababtoo ah luminta ku sii hooseysa culeyska yar ee culeyska ayaa u janjeera in ay si ka wanaagsan xogta aan la arkin u soo koobto.
Immisa baas oo hore-u-dhac ah ayay u baahan tahay tallaabada SAM ee caadiga ah?
SAM waxa ay xisaabisaa jibbaarada si ay u hesho meesha ugu xun ee u dhow, ka dibna gradient kale halkaas si loo cusboonaysiiyo - qiyaastii laba jibaar qiimaha caadiga ah.
Muxuu ka taliyaa radius hyperparameter rho gudaha SAM?
Rho wuxuu dejiyaa inta ay tillaabada 'korka' u dhaqaaqdo si loo helo deris kiiskii ugu xumaa, isagoo qeexaya inta dhul fidsan SAM uu raadinayo.
Waa maxay ta koowaad ee labada tallaabo ee SAM ee soo noqnoqon kasta?
SAM marka hore waxay xumeeyaan miisaanka ku dhex jira radius rho jihada kor u qaadaysa khasaaraha, ka dibna waxay ka soo degtaa barta asalka ah iyadoo la adeegsanayo gradient-ka halkaas lagu xisaabiyay.
Maxay SAM inta badan u wanaajisaa adkaanta si ay u calaamadiso buuqa?
Xifdinta calaamadaha buuqa badanaa waxay u baahan yihiin fiiqan, yar yar oo cidhiidhi ah; Iyagoo doorbidaya gobollo siman, SAM waxay iska caabisay in xad dhaafka ah.