InoteveraGaidhi rinotevera
Weight Initialization
Tekinoroji
Nhungamiro yehunyanzvi
Stochastic Weight Avhareji
Stochastic Weight Averaging (SWA) inotora avhareji yakapusa yehuremu hwemodhi kubva akati wandei mapoinzi kunonoka mukudzidziswa pane kungochengeta mufananidzo wekupedzisira.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
Uhwu hunyengeri hwakachipa hunowanzo dzika modhi munzvimbo yakapfava, yakafara nzvimbo yekurasika, iyo inowanzo wedzera zvinooneka zvirinani pane isingaonekwe data.
Kudzika Kwakadzika
Yakaunzwa naIzmailov, Wilson nevamwe vaaishanda navo muna 2018, SWA inoshandisa fungidziro yekuti SGD ine chisingaperi kana cyclical kudzidza mwero haichinji kune imwe nzvimbo - inosvetuka ichitenderedza kumucheto kwemupata wakafara, wakafuratira. Panzvimbo pekutora imwe yeidzo nzvimbo dzekumisa dzine ruzha, SWA inomhanya zvine mwero (kazhinji inogara kana kutenderera) mwero wekudzidza wenguva yekupedzisira uye avhareji huremu hwainoshanyira, kazhinji nguva yega yega. Huremu hweavhareji hunogara padyo nepakati penzvimbo yakati sandara. Nekuda kwekuti batch-normalization statistics inoverengerwa kune chaiwo huremu, SWA inoda imwe yekuwedzera kumberi kupfuura data kudzoreredza BN inomhanya nzira uye misiyano yeavhareji modhi. Mutengo wacho ndewemahara, uye kuwana kwechokwadi kunopindirana mumhando dzemifananidzo uye nekupfuura.
Kuitwa muNyika yeChokwadi
Kuwedzera bvunzo kurongeka kweResNet uye DenseNet mufananidzo classifiers paCIFAR uye ImageNet pasina imwe yekuwedzera mutengo.
SWAG (SWA-Gaussian) inogadzira fungidziro yekusavimbika yakayerwa yekufembera inotarisisa kubva kune imwechete kudzidziswa kumhanya.
EMA-ye-huremu inodzikamisa iyo sampling network mukuparadzira mifananidzo jenareta seStable Diffusion.
Kugadzira 'soups yemuenzaniso' neaverage yakawanda-yakanyatso gadziridzwa nzvimbo dzekutarisa kuti uvandudze kusimba pasina kudzidziswazve.
Ramangwana reStochastic Weight Averaging
SWA yakaburitsa akasiyana seSWA-Gaussian (SWAG) yekusagadzikana kweBayesi yakachipa, uye iyo pfungwa yeavhareji ikozvino inosimbisa Exponential Moving Average matipi anoshandiswa zvakanyanya mumhando dzekuparadzira, kudzidza kwekuzvitarisira, uye hombe-modhi yekudzidzira. Tarisira kuti huremu hurambe huri 'yemahara yekudya kwemasikati' mukudzidzira mabikirwo, netsvakiridzo inoiwedzera kusvika pakubatanidza mamodheru akadzidzira (modhi soups) uye kuvandudza kuenzanisa padivi pechokwadi mbishi.
Kuzvishandisa
Nei zvichikosha
- Mutengo uye bhajeti. Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore.
- Sarudzo dzakajeka. Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete.
- Kudzora kwemhando yepamusoro. Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira.
Chenjera
- Kugadzirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba.
- Infrastructure uye mari yekugadzirisa inowanzotarisirwa pasi.
- Chengetedzo uye kucherechedzwa mapundu anogona kukura sezvo masisitimu anowedzera kuoma.
Matanho anotevera
- Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa.
- Benchmark pasi pechokwadi mutoro uye data mamiriro.
- Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro.
- Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera.
Ramba Uchiongorora
Tsamba yemahara
Tora iyo yemazuva ese AI pfupiso
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Take the Stochastic Weight Averaging quiz
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Mibvunzo inowanzo bvunzwa
Chii chinonzi Stochastic Weight Averaging?
Stochastic Weight Averaging (SWA) inotora avhareji yakapusa yehuremu hwemodhi kubva akati wandei mapoinzi kunonoka mukudzidziswa pane kungochengeta mufananidzo wekupedzisira. Uhu hunyengeri hwakachipa hunowanzo dzika modhi munzvimbo yakapfava, yakafara nzvimbo yekurasika, iyo inowanzo wedzera zvinooneka zvirinani pane isingaonekwe data.
Chii chinonzi Stochastic Weight Averaging chaizvo?
SWA inoyera mapeji emuenzaniso (zviremu) zvakaunganidzwa panzvimbo dzinoverengeka dzekutarisa panguva yekupedzisira yekudzidziswa, kwete kufanotaura kana gradients.
Sei SWA ichida kuvandudza generalization?
Avhareji inofambisa mhinduro yakananga pakati penzvimbo yakati sandara yenzvimbo yekurasikirwa, uye flat minima generalize zvirinani nekuti chitima uye kurasikirwa kwebvunzo kunogara kwakabatana.
Nderipi rimwe danho rinodiwa neSWA kune network inoshandisa batch normalization?
Nekuti huwandu hweBN hunoenderana nehuremu chaihwo, iyo yeavhareji modhi inoda imwe yekuwedzera kupfuura data kuti idzokorore nzira dzekumhanya uye kusiyana.
Ndeapi maitiro ekudzidza-chiyero anowanzo shandiswa panguva yeavhareji yeSWA chikamu?
SWA inochengeta mwero wakakwira zvine mwero kana cyclic yekudzidza kuitira kuti SGD ienderere mberi ichiongorora nzvimbo yakafara furati ine mapoinzi anobva apinzwa pakati.
Mari yekufungidzira yeSWA inofananidzwa sei neyechinyakare ensemble yeN modhi?
SWA inodonhedza macheki akawanda kuita avhareji huremu seti, saka kufungidzira kunodhura zvakafanana nemhando imwe chete kwete N.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi