Oko kufuneka ukwazi malunga nokuziphendulela kwengqondo
Ukufunda okunzulu luhlobo olunamandla lokufunda umatshini (ML) olwakha izakhiwo zeemathematika ezinzima ezibizwa ngokuba yi- neural networks usebenzisa ubuninzi beenkcukacha (ulwazi).
Inkcazo yoNzulu
Ukufunda ngokujulileyo yindlela yokuphumeza iML usebenzisa iindlela ezininzi zokunxibelelana kwee-neural ukwenza iinkqubo ezinzima zenkcukacha. Ngamanye amaxesha ubizwa ngokuba yi-hierarchical learning, ukufunda okunzulu kusetyenziswa iintlobo ezahlukeneyo zokunxibelelanisa izixhobo ze-neural ukufunda iimpawu (ezikwabizwa ngokuba zimelo) kwaye uzifumene kwiiseti ezinkulu zeenkcukacha eziluhlaza, ezingabonakaliyo (idatha engaqinisekanga). Enye yezinto zokuqala zokubonakaliswa kokufunda okunzulu yinkqubo eyiphumelele ethatha imifanekiso yeekati ephuma kwiiseti zevidiyo ze-YouTube.
Imizekelo eFundisiweyo yoBomi bemihla ngemihla
Ukufunda ngokujulileyo akusetyenziswanga kuphela ekuqaphelisweni komfanekiso, kodwa kwakhona ukuguqulelwa kolwimi, ukufumanisa ubuqhetseba, nokuhlalutya idatha eqokelelwe ngamashishini ngamakhasimende abo. Umzekelo, i-Netflix isebenzisa ukufunda okunzulu ukuhlalutya imikhuba yokubukela kunye nokuqikelela oko kukubonisa kunye namafilimu ofuna ukuwajonga. Yiyo indlela iNetflix eyazi ngayo ukubeka iifrimu zamanyathelo kunye neencwadi zendalo kwinqongo yakho yokucebisa. I-Amazon isebenzisa ukufunda okunzulu ukuhlalutya ukuthengwa kwakho kwangoko kunye nezinto ozifunayo kutshanje ukwenza iziphakamiso zama-albhamu omculo welizwe elitsha onokuthi unomdla kuzo kwaye uhamba kwimarike yebhanki ye-grey ne-yellow izicathulo. Njengoko ukufunda okunzulu kunika ingqiqo nangakumbi kwi data engaqulunqiweyo kunye neyobugcisa, iinkampani zinokulindela kangcono iimfuno zabaxumi babo ngelixa wena, umthengi ngamnye ufumana inkonzo yamakhasimende athile.
IiNewural Networks kunye neZifundo eziPhezulu
Ukuze sifunde lula ukufunda, masibuye sihlolisise ukuthelekiswa kwethu kwe-network neural network (ANN). Ukufunda okunzulu, sicinga ukuba isakhiwo sethu sezitezi ezili-15 sinomxholo wesikhokelo sipala kunye nezinye izakhiwo zesebe. Kukho izakhiwo ezintathu kwicala ngalinye kwisitalato. Isakhiwo sethu sakha i-A kwaye sichaza kwicala elifanayo kwisitrato njengezakhiwo B kunye noC. Ngaphandle kwesitalato ukusuka kwisakhiwo A yakha 1, kwaye ngaphaya kwesakhiwo B isakhiwo 2, njalo njalo. Isakhiwo ngasinye sinomgangatho ohlukeneyo weepansi, senziwa kwizinto ezihlukeneyo kwaye sineendlela ezahlukeneyo zokwakha ezivela kwabanye. Nangona kunjalo, isakhiwo ngasinye sisalungiselelwe kwindawo ehlukeneyo (iifayile) zeeofisi (iinombolo) -so isakhiwo ngasinye si-ANN ekhethekileyo.
Khawucinge ukuba iphakheji yedijithali iyafika ekwakheni A, equle iindidi ezininzi zeenkcukacha ezivela kwimithombo emininzi efana neenkcukacha ezisekelwe kwimiyalezo, imilayezo yevidiyo, imifudlana yomsindo, iifowuni, iifowuni zomsakazo kunye neefoto-nangona kunjalo, zifikile kwelinye elikhukhuni elikhulu kunye ayibhaliwe okanye ihlelwe nayiphi na indlela enengqiqo (idatha engaqinisekanga). Ingcaciso ithunyelwa kumgangatho ngamnye ngokulandelelana ukususela ngo-1 ukuya kwe-15 ukuya ku-15. Emva kokuba i-jumble information ifike kumgangatho we-15 (i-output), ithunyelwa kwisiteji esingu- 1 (ingeniso) yesakhiwo sesi-3 kunye nesiphumo sokugqibela esakhiwa kwisakhiwo A. Isakhiwo 3 sifunda kwaye siquka isiphumo esithunyelwe ngesakhiwo A kwaye emva koko iqulethe i-information jumble ngaphakathi komgangatho ngamnye ngendlela efanayo. Xa ulwazi lufikelela kumgangatho ophezulu wesakhiwo sesi-3, ithunyelwa ukusuka apho kunye neziphumo zesakhiwo ekwakheni 1. Ukwakha 1 ufundela kwaye iquka iziphumo ezisuka kwisakhiwo 3 ngaphambi kokucwangcisa phansi-phantsi. Ukwakhiwa kwe-1 kudlula ulwazi kunye neziphumo ngendlela efanayo yokwakha iC, eyenza inkqubo kwaye ithumele kwisakhiwo sesi-2, esenza kwaye ithumele kwisakhiwo B.
Ngalunye i-ANN (isakhiwo) kumzekelo wethu sifuna inkalo eyahlukileyo kwiedatha ezingenakulungiswa (i-jumble yolwazi) kwaye idlula iziphumo kwisakhiwo esilandelayo. Isakhiwo esilandelayo sibandakanya (sifunda) umphumo (iziphumo) ukusuka kwangaphambili. Njengoko idatha iqhutywe yi-ANN (isakhiwo) nganye, ihlelwa kwaye ibhalwe (ihlelwe) ngento ethile ukuze xa idatha ifikelela kumgangatho wokugqibela (umgangatho ophezulu) we-ANN yokugqibela (isakhiwo), ichazwa kwaye ibhalwe (ehlelwe ngakumbi).
Intelligence Intelligence, Learning Machine, kunye nokuFunda okuNzulu
Ukufunda okunzulu kufanelana njani nomfanekiso jikelele weengcinga zobuchule (AI) kunye neML? Ukufunda ngokujulile kwandisa amandla eML kunye nokwandisa uluhlu lwemisebenzi e-AI inokwenza. Ngenxa yokuba ukufunda okunzulu kuxhomekeke ekusebenziseni amathenwa e-neural kunye nokuqwalasela iimpawu kwiiseti zeenkcukacha endaweni yezinto ezilula zomsebenzi, zifumane kwaye zisebenzise iinkcukacha ezivela kwidata engaqulunqiweyo (eluhlaza) ngaphandle kwemfuneko yokuba umprogram ukuba ayibhale phantsi okokuqala-ixesha - umsebenzi wokuqulunqa onokuthi ungenise iimpazamo. Ukufunda ngokujulileyo kukunceda iikhomputha ziphucule kwaye zibhetele ekusebenziseni idatha ukunceda zombini iinkampani kunye nabantu ngabanye.