Itekhnoloji yamva nje yeMicrosoft eza kwiAzure AI ichaza imifanekiso kunye nabantu


Abaphandi beMicrosoft benze inkqubo yobukrelekrele eyenziweyo enokuvelisa amagama emifanekiso, kwiimeko ezininzi, ichaneke ngakumbi kuneenkcazo zabantu. Le mpumelelo iphawula isiganeko esibalulekileyo ekuzibopheleleni kukaMicrosoft ekwenzeni iimveliso neenkonzo ziquke kwaye zifikeleleke kubo bonke abasebenzisi.

"Inkcazo yomfanekiso yenye yemisebenzi ephambili yombono wekhompyutheni, eyenza uluhlu olubanzi lweenkonzo," kusho uXuedong Huang (Xuedong Huang), ugxa weMicrosoft wezobugcisa kunye negosa eliyintloko letekhnoloji ye-Azure AI yeeNkonzo zoQoqosho eRedmond, eWashington.

Imodeli entsha ngoku iyafumaneka kubathengi ngeComputer Vision kwi Iinkonzo zeNgcaciso ze-Azure, eyinxalenye ye-Azure AI, kwaye ivumela abaphuhlisi ukuba basebenzise obu buchule bokuphucula ukufumaneka kweenkonzo zabo. Ikwafakwe kwi-app ye-Seing AI kwaye iya kusungulwa kamva kulo nyaka kwi-Microsoft Word kunye ne-Outlook yeWindows kunye ne-Mac, kunye ne-PowerPoint ye-Windows, i-Mac kunye newebhu.

Inkcazo ezenzekelayo inceda abasebenzisi ukuba bafikelele kumxholo obalulekileyo wawo nawuphi na umfanekiso, nokuba yifoto ebuyiselwe kwisiphumo sokukhangela okanye umboniso womboniso.

"Ukusetyenziswa kweenkcazo ezichaza umxholo wemifanekiso (ebizwa ngokuba yi-alt text) kumaphepha ewebhu kunye namaxwebhu kubaluleke kakhulu kubantu abangaboniyo okanye abanombono ophantsi," kusho uSaqib Sheikh (Saqib Sheikh), umphathi wesoftware kwiqela leMicrosoft AI Platforms eRedmond.

Umzekelo, iqela lakhe lisebenzisa inkcazo ephuculweyo yomfanekiso kwi-app yabantu abangaboniyo nabangaboniyo Ukubona i-AI, eqaphela ukuba ikhamera ifota ntoni ize ithethe ngayo. I-app isebenzisa ii-captions ezenziweyo ukuchaza iifoto, kubandakanywa nenethiwekhi yoluntu.

Ngokufanelekileyo, wonke umntu kufuneka afake i-alt kuyo yonke imifanekiso ekumaxwebhu, kwi-intanethi, nakwimidiya yoluntu, njengoko oku kuvumela abantu abangaboniyo ukuba bafikelele kumxholo kwaye bathathe inxaxheba kwincoko. Kodwa, yeha, abantu abakwenzi oku, ”utshilo uSheikh. "Nangona kunjalo, kukho ii -apps ezininzi ezisebenzisa inkcazo yesithombe ukongeza enye isicatshulwa xa ingekho."
  
Itekhnoloji yamva nje yeMicrosoft eza kwiAzure AI ichaza imifanekiso kunye nabantu

ULijuan Wang, umphathi omkhulu wophando kwilebhu yakwaMicrosoft Redmond, ukhokele iqela lophando elithe lafumana iziphumo ezinjengomntu kunye nezingcono. Ifoto: Dan DeLong.

Inkcazo yezinto ezintsha

"Inkcazo yomfanekiso yenye yemisebenzi ephambili yombono wekhompyutheni, efuna inkqubo yengqondo yokufakelwa ukuqonda nokuchaza umxholo oyintloko okanye isenzo esimelelwe emfanekisweni," kuchaza uLijuan Wang (Lijuan Wang), umphathi omkhulu wophando kwilebhu yakwaMicrosoft Redmond.

β€œKufuneka uqonde ukuba kuqhubeka ntoni na, ufumanise ukuba buyintoni na ubudlelwane phakathi kwezinto nezenzo, emva koko ushwankathele kwaye uchaze konke ngesivakalisi ngolwimi oluqondwa ngabantu,” utshilo.

UWang ukhokele iqela lophando elilinganiselayo iinocaps (inkcazo-ntetho yenoveli ngokomlinganiselo, inkcazo enkulu yezinto ezitsha) iphumelele iziphumo ezithelekiseka nezabantu kwaye yagqitha kuzo. Olu vavanyo luvavanya ukuba iinkqubo ze-AI zivelisa njani iinkcazo zezinto ezibonakalisiweyo ezingeyonxalenye yedatha apho imodeli yaqeqeshwa khona.

Ngokuqhelekileyo, iisistim zenkcazo yemifanekiso ziqeqeshwe kwiiseti zedatha eziqulethe imifanekiso ehamba kunye neenkcazo ezibhaliweyo zale mifanekiso, oko kukuthi, kwiiseti zemifanekiso ebhaliweyo.

"Uvavanyo lwe-nocaps lubonisa indlela inkqubo enokuthi ichaze ngayo izinto ezintsha ezingafumanekiyo kwidatha yoqeqesho," kusho uWang.

Ukucombulula le ngxaki, iqela leMicrosoft liqeqeshe kwangaphambili imodeli ye-AI enkulu kwidathasethi enkulu enemifanekiso eneempawu zamagama, nganye yazo idibaniswe nento ethile emfanekisweni.

Kwasebenza kakuhle kakhulu ukwenza iiseti zemifanekiso enethegi zamagama endaweni yeenkcazelo ezipheleleyo, nto leyo eyavumela iqela likaWang ukuba londle idatha eninzi kwimodeli yabo. Le ndlela yokwenza inike imodeli oko iqela likubiza ngokuba sisigama esibonakalayo.

Njengoko u-Huang wachazayo, isigama esibonakalayo sokufundisa kwangaphambili siyafana nokulungiselela abantwana ukuba bafunde: Okokuqala, incwadi yemifanekiso isetyenziswa apho amagama ngamanye adityaniswa nemifanekiso, umzekelo, phantsi kwefoto yeapile ithi "apile" kwaye phantsi kwefoto yekati igama elithi "ikati".

β€œOlu qeqesho lwangaphambili lunesichazi-magama esibonakalayo yeyona nto ifunekayo ukuze kuqeqeshwe le nkqubo. Le yindlela esizama ngayo ukuphuhlisa uhlobo lwenkumbulo yemoto, ”utshilo uHuang.

Imodeli eqeqeshwe kwangaphambili ilungiswa kusetyenziswa i-dataset equka imifanekiso ene-captioned. Kweli nqanaba loqeqesho, imodeli ifunda ukwenza izivakalisi. Ukuba umfanekiso ubonakala uqulethe izinto ezintsha, inkqubo ye-AI isebenzisa isichazi-magama esibonakalayo ukwenza iinkcazo ezichanekileyo.

"Ukujongana nezinto ezintsha ngexesha lokuvavanya, inkqubo idibanisa oko yakufunda ngexesha loqeqesho lwangaphambili kwaye ngexesha lophuhliso olulandelayo," kusho uWang.
Ngokweziphumo uphandoXa ivavanywa kwiimvavanyo ze-nocaps, inkqubo ye-AI ivelise iinkcazo ezinentsingiselo nezichanekileyo kunokuba abantu benze imifanekiso efanayo.

Inguqu ekhawulezileyo ukuya kwindawo yokusebenza 

Phakathi kwezinye izinto, inkqubo entsha yenkcazo yomfanekiso iphindwe kabini ilungile njengemodeli esetyenziswa kwiimveliso kunye neenkonzo zikaMicrosoft ukusukela ngo-2015, ngokutsho kolunye uphawu loshishino.

Ngenxa yeenzuzo eziya kufunyanwa ngabo bonke abasebenzisi beemveliso kunye neenkonzo zikaMicrosoft kolu phuculo, uHuang uye wakhawulezisa ukudityaniswa kwemodeli entsha kwindawo yedesktop yeAzure.

β€œSithatha le teknoloji ye-AI siyisa e-Azure njengeqonga lokusebenzela uluhlu olubanzi lwabathengi,” utshilo. β€œKwaye oku kuyimpumelelo hayi kuphando kuphela. Ixesha elithathiweyo ukubandakanya le nkqubela phambili kwindawo yemveliso ye-Azure nayo yaba yinkqubela phambili. ”

U-Huang wongeze ukuba ukuzuza iziphumo ezinje ngomntu kuyaqhubeka nomkhwa osele usekwe kwiinkqubo zengqondo zeMicrosoft.

Kule minyaka mihlanu idlulileyo, siye safumana iziphumo zomgangatho wabantu kwiindawo ezintlanu eziphambili: ukuqondwa kwentetho, ukuguqulelwa ngomatshini, ukuphendula imibuzo, ukufundwa koomatshini kunye nokuqonda okubhaliweyo, kwaye ngo-2020, ngaphandle kwe-COVID-19, inkcazo yemifanekiso "utshilo uJuan.

Ngesihloko

Thelekisa iziphumo zenkcazo yemifanekiso eyanikwa yinkqubo ngaphambili kwaye ngoku isebenzisa i-AI

Itekhnoloji yamva nje yeMicrosoft eza kwiAzure AI ichaza imifanekiso kunye nabantu

Ifoto evela kwithala leencwadi leGetty. Inkcazo yangaphambili: Ukuvalwa kwendoda epheka inja eshushu kwibhodi yokusika. Inkcazo entsha: Indoda yenza isonka.

Itekhnoloji yamva nje yeMicrosoft eza kwiAzure AI ichaza imifanekiso kunye nabantu

Ifoto esuka kwithala leencwadi leGetty. Inkcazo yangaphambili: Indoda ihlala ekutshoneni kwelanga. Inkcazo entsha: I-Bonfire elunxwemeni.

Itekhnoloji yamva nje yeMicrosoft eza kwiAzure AI ichaza imifanekiso kunye nabantu

Ifoto evela kwithala leencwadi leGetty. Inkcazo yangaphambili: Indoda enxibe ihempe eluhlaza. Inkcazo entsha: Abantu abaninzi abanxibe iimaski zotyando.

Itekhnoloji yamva nje yeMicrosoft eza kwiAzure AI ichaza imifanekiso kunye nabantu

Ifoto esuka kwithala leencwadi leGetty. Inkcazo yangaphambili: indoda ekwibhodi yokutyibiliza iphaphazela eludongeni. Inkcazo entsha: Umdlali we-baseball ubamba ibhola.

umthombo: www.habr.com

Yongeza izimvo