UMHLAHLANDLELA Wobuchwepheshe

I-Stochastic Weight Average

I-Stochastic Weight Averaging (SWA) ithatha isilinganiso esilula sezisindo zemodeli emaphuzwini ambalwa sekwephuzile ekuqeqesheni esikhundleni sokugcina isifinyezo sokugcina.

Uhlolojikelele

I-Stochastic Weight Averaging iyibhulokhi yokwakha yobuchwepheshe ethinta ikhwalithi yemodeli, izindleko zengqalasizinda, ukubambezeleka, nokuthembeka esikalini.

I-Deep Dive

Yethulwe ngu-Izmailov, uWilson kanye nozakwabo ngo-2018, i-SWA isebenzisa umbono wokuthi i-SGD enezinga lokufunda elingashintshi noma elijikelezayo ayiguquki ibe iphuzu elilodwa - igxumagxuma izungeza unqenqema lwesigodi esibanzi, esiyisicaba. Kunokuba ikhethe enye yalezo zindawo zokuma ezinomsindo, i-SWA isebenzisa izinga lokufunda eliphezulu (ngokuvamile elihlala njalo noma eliwumjikelezo) wezinkathi zokugcina futhi ilinganisela izisindo ezivakashelayo, ngokuvamile yonke inkathi. Izisindo ezimaphakathi zihlala eduze nendawo emaphakathi yendawo eyisicaba. Ngenxa yokuthi izibalo ze-batch-normalization zibalelwa ezisindweni ezithile, i-SWA idinga ukudlula okukodwa okungaphezulu kwedatha ukuze ibale kabusha izindlela ezisebenzayo ze-BN nokuhluka kwemodeli emaphakathi. Izindleko zimahhala, futhi izinzuzo zokunemba ziyahambisana kuzo zonke izihlukanisi zezithombe nangale kwalokho.

I-Technical Insight

I-SWA igcina isilinganiso esisebenzayo w_SWA = (n·w_SWA + w_i)/(n+1) esibuyekezwayo somjikelezo ngamunye, kuyilapho imodeli ye-SGD ebukhoma iqhubeka ihlola ngezinga lokufunda elikhulu uma kuqhathaniswa. Isilinganiso esikhaleni sesisindo silinganisa iqoqo endaweni yokusebenza kodwa kubiza imodeli eyodwa ngencazelo, hhayi eminingi. Indlela eyinhloko ukuthi i-flat minima iqinile ekuphazamisekeni kwesisindo, ngakho izindawo zokuqeqeshwa/ukuhlolwa kokulahlekelwa zihlala ziqondile, kunciphisa igebe elivamile.

I-Mastering Stochastic Weight Average

I-Stochastic Weight Averaging (SWA) ithatha isilinganiso esilula sezisindo zemodeli emaphuzwini ambalwa sekwephuzile ekuqeqesheni esikhundleni sokugcina isifinyezo sokugcina. Leli qhinga elishibhile livamise ukubeka imodeli endaweni eyisicaba, ebanzi yokwakheka kwezwe yokulahlekelwa, okuvamise ukuvela kangcono kudatha engabonakali. I-Stochastic Weight Averaging iyibhulokhi yokwakha yobuchwepheshe ethinta ikhwalithi yemodeli, izindleko zengqalasizinda, ukubambezeleka, nokuthembeka esikalini. Ukuze wakhe ukuqonda okujulile, phatha i-Stochastic Weight Averaging njengemodeli yokusebenza, hhayi isici esisodwa: chaza imiphumela efiselekayo, ucacise ukucabanga, futhi uhlukanise lokho uhlelo olungakwenza ngokwethembeka kulokho okusadinga ukwahlulela kochwepheshe.

Empeleni, amaqembu aqinile asebenzisa i-Stochastic Weight Averaging athuthukisa izakhiwo, idatha, nokukhetha kwengqalasizinda ngokumelene nokuthembeka nezindleko. Babhala imibandela yempumelelo ecacile, ukuhlola okuqhathaniswa nedatha engokoqobo nokugeleza komsebenzi, futhi baphindaphinde ngokusekelwe kumaphethini okuhluleka aqashiwe esikhundleni sokuwina kwebhentshimakhi yesikhathi esisodwa. Yilapho ukuqonda kwethiyori kuguquka kube amandla ahlala njalo kuwo wonke umkhiqizo, inqubomgomo, kanye nokusebenza.

Izinqumo zezakhiwo ziqhuba ukusebenza kanye nezindleko zokusebenza iminyaka. Ngesikhathi esifanayo, Ukuthuthukisa ibhentshimakhi eyodwa kungafihla ubuthakathaka obubanzi besistimu. Indlela eqine kakhulu iwukuhlanganisa isivinini sokuhlola nesiyalo sokuphatha: qhuba abashayeli bezindiza, bamba ubufakazi, ushicilele amalogi ezinqumo, futhi ubuyekeze izivikelo ngokuqhubekayo njengoba imodeli yokuziphatha, okulindelwe ngabasebenzisi, kanye nezimfuneko zokulawula zishintsha.

I-Strategic Impact

Izinqumo zezakhiwo ziqhuba ukusebenza kanye nezindleko zokusebenza iminyaka.

Izinqumo zezakhiwo ziqhuba ukusebenza kanye nezindleko zokusebenza iminyaka. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.

Imfundo yobuchwepheshe isiza amaqembu ukuthi akhethe isitaki esifanele, hhayi nje esisha.

Imfundo yobuchwepheshe isiza amaqembu ukuthi akhethe isitaki esifanele, hhayi nje esisha. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.

Izinketho ezingcono zobunjiniyela zinciphisa izehlakalo ezinokwethenjelwa ekukhiqizeni.

Izinketho ezingcono zobunjiniyela zinciphisa izehlakalo ezinokwethenjelwa ekukhiqizeni. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.

Ikusasa le-Stochastic Weight Average

I-SWA iveze okuhlukile okufana ne-SWA-Gaussian (SWAG) yokungaqiniseki okushibhile kwe-Bayesi, futhi umbono olinganiselayo manje usekela amaqhinga e-Exponential Moving Average asetshenziswa kabanzi kumamodeli asakazwayo, ukufunda ukuzigada, kanye nokuziqeqesha kwamamodeli amakhulu. Lindela ukulinganiswa kwesisindo ukuze kuhlale 'kuyisidlo sasemini' esizenzakalelayo ekuqeqesheni amaresiphi, ngocwaningo olunweba ekuhlanganiseni amamodeli aqeqeshwe ngokuzimela (amamodeli amasobho) kanye nokuthuthukisa ukulinganisa ngokuhambisana nokunemba okungaphekiwe.

Ukuqaliswa Komhlaba Wangempela

Ukuthuthukisa ukunemba kokuhlolwa kwezihlukanisi zezithombe ze-ResNet ne-DenseNet ku-CIFAR ne-ImageNet ngaphandle kwezindleko ezengeziwe.

I-SWAG (SWA-Gaussian) ikhiqiza izilinganiso zokungaqiniseki ezilinganiselwe zokuqagela okuzwelayo zokuphepha kusukela ekugijimeni kokuqeqeshwa okukodwa.

I-EMA-yezisindo iqinisa inethiwekhi yesampula kumajeneretha wesithombe esisabalalisiwe njenge-Stable Diffusion.

Ukwakha 'amasobho emodeli' ngokulinganisa izindawo zokuhlola ezishunwe kahle ukuze kuthuthukiswe ukuqina ngaphandle kokuqeqeshwa kabusha.

Amaphethini Okusebenzisa

I-Stochastic Weight Average in practice

Ukuthuthukisa ukunemba kokuhlolwa kwezihlukanisi zezithombe ze-ResNet ne-DenseNet ku-CIFAR ne-ImageNet ngaphandle kwezindleko ezengeziwe.

Ukuthuthukisa ukunemba kokuhlolwa kwezihlukanisi zezithombe ze-ResNet ne-DenseNet ku-CIFAR ne-ImageNet ngaphandle kwezindleko ezengeziwe zokucatshangelwa Amathimba ngokuvamile athola imiphumela engcono uma echaza izinga eliphezulu ngaphambili, egcina indlela yokukhuphuka yomuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

I-Stochastic Weight Average in practice

I-SWAG (SWA-Gaussian) ikhiqiza izilinganiso zokungaqiniseki ezilinganiselwe zokuqagela okuzwelayo zokuphepha kusukela ekugijimeni kokuqeqeshwa okukodwa.

I-SWAG (SWA-Gaussian) ikhiqiza izilinganiso zokungaqiniseki ezilinganisiwe zokuqagela okuzwelayo zokuphepha kusukela ekugijimeni okukodwa kokuqeqeshwa Amaqembu ngokuvamile athola imiphumela engcono uma echaza imikhawulo yekhwalithi ngaphambili, egcina indlela yokukhuphuka kwabantu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

I-Stochastic Weight Average in practice

I-EMA-yezisindo iqinisa inethiwekhi yesampula kumajeneretha wesithombe esisabalalisiwe njenge-Stable Diffusion.

I-EMA-yezisindo ezinzisa inethiwekhi yamasampula kumajeneretha ezithombe ezisabalalisiwe njenge-Stable Diffusion Teams ngokuvamile athola imiphumela engcono uma echaza izilinganiso zekhwalithi ngaphambili, egcina indlela yokukhuphuka yomuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

I-Stochastic Weight Average in practice

Ukwakha 'amasobho emodeli' ngokulinganisa izindawo zokuhlola ezishunwe kahle ukuze kuthuthukiswe ukuqina ngaphandle kokuqeqeshwa kabusha.

Ukwakha 'amasobho emodeli' ngokuhlola izindawo zokuhlola ezicushwe kahle eziningi ukuze kuthuthukiswe ukuqina ngaphandle kokuqeqesha kabusha Amathimba ngokuvamile athola imiphumela engcono uma echaza izinga eliphezulu ngaphambili, egcina indlela yokukhuphuka komuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

Izingozi & Guardrails

Ukuthuthukisa ibhentshimakhi eyodwa kungafihla ubuthakathaka obubanzi besistimu.

Izindleko zengqalasizinda nezokulungisa zivame ukubukelwa phansi.

Izikhala zokuphepha nokubonakala zingakhula njengoba izinhlelo ziba nzima kakhulu.

Ukuqalisa Umhlahlandlela

Chaza ukubambezeleka, ikhwalithi, nezindleko ezihlosiwe ngaphambi kokuqaliswa.

Chaza ukubambezeleka, ikhwalithi, nezindleko ezihlosiwe ngaphambi kokuqaliswa. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

Ibhentshimakhi ngaphansi komthwalo wangempela nezimo zedatha.

Ibhentshimakhi ngaphansi komthwalo wangempela nezimo zedatha. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

Ukuqapha amathuluzi amaphutha, ukukhukhuleka, nomthelela wabasebenzisi.

Ukuqapha amathuluzi amaphutha, ukukhukhuleka, nomthelela wabasebenzisi. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

Lungiselela izindlela zokuhlehlisa nezigameko ngaphambi kokukala.

Lungiselela izindlela zokuhlehlisa nezigameko ngaphambi kokukala. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

Qhubeka Uhlole

I-AI Benchmarks

Sebenzisa ukuhlola kahle uma uqhathanisa izinketho zobuchwepheshe.

Funda Umhlahlandlela

Ukuqinisa Ukufunda

Ngena ujule kumasu okuqeqeshwa kobuchwepheshe.

Funda Umhlahlandlela