Okudingayo ukwazi mayelana nokuziphendukela kwe-intelligence yokufakelwa
Ukufunda okujulile kuyindlela enamandla yokufunda umshini (ML) eyakha izakhiwo eziyinkimbinkimbi zezibalo ezibizwa ngokuthi amanethiwekhi we-neural usebenzisa inqwaba yedatha (ulwazi).
Incazelo yokufunda okujulile
Ukufunda okujulile kuyindlela yokusebenzisa i-ML ngokusebenzisa izakhi eziningi zamanethiwekhi we-neural ukucubungula izinhlobo eziyinkimbinkimbi zedatha. Ngezinye izikhathi kuthiwa ukufunda okuphakeme, ukufunda okujulile kusebenzisa izinhlobo ezahlukene zamanethiwekhi omzimba we-neural ukuze ufunde izici (ezibizwa nangokuthi izethulo) futhi uzifake ezinhlakeni ezinkulu zedatha eluhlaza, engabhaliwe (idatha engakhiwa). Esinye sezibonakaliso zokuqala zokufunda ngokujulile kwakuhlelo oluthathe ngokuphumelelayo izithombe zamakati eziphuma kumaqoqo amavidiyo we-YouTube.
Izibonelo Ezijulile Zokufunda Ekuphileni Kwansuku zonke
Ukufunda okujulile akusetshenziswanga kuphela ekuqaphelisweni komfanekiso, kodwa futhi nokuhumusha ulimi, ukutholakala kwenkohliso, nokuhlaziya idatha eqoqwe izinkampani ngamakhasimende abo. Isibonelo, i-Netflix isebenzisa ukufunda okujulile ukuhlaziya imikhuba yakho yokubukela nokubikezela ukuthi yimaphi amamabhayisikobho namafilimu othanda ukuwabuka. Yile ndlela uNetflix azi ngayo ukubeka amafilimu ezenzo nezithombe zemvelo kusigcawu sakho sokusikisela. I-Amazon isebenzisa ukufunda okujulile ukuhlaziya ukuthenga kwakho okwedlule nezinto ozisanda kusesha ukuze udale iziphakamiso zama-albhamu omculo wezwe omusha ongase ube nesithakazelo kuwo nokuthi usemakethe ye-pair of gray tennis and yellow tennis izicathulo. Njengokufunda okujulile kunikeza ukuqonda okuningiliziwe kusuka kwedatha engakhiwe futhi engafani, izinkampani zingalindela kangcono izidingo zamakhasimende abo ngenkathi wena, ikhasimende ngalinye lithola insizakalo yamakhasimende enziwe ngokwengeziwe.
Ama-Neural Networks Asebenzayo Nokufunda Okujulile
Ukuze senze ukufunda okujulile kulula ukuyiqonda, ake sihlole ukufaniswa kwethu nenethiwekhi ye-neural yokufakelwa (ANN). Ukuze ufunde ngokujulile, cabanga ukuthi isakhiwo sethu sezakhiwo ezingu-15 sinebhilidi lomuzi nezinye izakhiwo ezinhlanu zehhovisi. Kunezakhiwo ezintathu ezihlangothini ngalunye zomgwaqo. Isakhiwo sethu sakha i-A futhi sihlanganyela eceleni komgwaqo njengoba izakhiwo B kanye noC. Ngaphakathi nomgwaqo kusukela ekwakhiweni A kuyakha 1, futhi ngaphesheya kwesakhiwo B yakha 2, njalonjalo. Isakhiwo ngasinye sinombhalo ohlukile wezansi, senziwe ngezinto ezihlukene futhi sinesitayela esithile esakhiwe ngabanye. Noma kunjalo, isakhiwo ngasinye sisalokhu senziwe ngamapulangwe (ama-layers) ahlukene (ama-nodes) -njalo isakhiwo ngasinye yi-ANN ehlukile.
Cabanga ukuthi iphakethe ledijithali lifika ekwakheni A, equkethe izinhlobo eziningi zolwazi oluvela emithonjeni eminingi efana nedatha esekelwe ematheksthini, imifudlana yevidiyo, ukusakazwa komsindo, izingcingo zocingo, amagagasi omsakazo nezithombe-noma kunjalo, kufinyelela endaweni eyodwa eyodwa enkulu futhi ayibhaliwe noma ihlelwe nganoma iyiphi indlela enengqondo (idatha engakhiwe). Lolu lwazi luthunyelwa ngegumbi ngalinye ngokulandelana kusukela ngomhlaka 1 kuya ku-15 th wokucubungula. Ngemuva kokuba i-jumble yolwazi ifinyelela ku-floor yeshumi nantathu (ukukhishwa), ithunyelwa kwisitezi esingu- 1 (okokufaka) kwesakhiwo 3 kanye nomphumela wokugcina wesakhiwo A. Ukwakha 3 ufunda futhi uhlanganisa umphumela othunyelwe ukwakha A kanye bese uqhuba ulwazi lwe-jumble ngokusebenzisa igumbi ngalinye ngendlela efanayo. Uma ulwazi lufika esiteji esiphezulu sesakhiwo sesi-3, sithunyelwa kusuka lapho kanye nemiphumela yaleso sakhiwo ekwakheni 1. Ukwakha 1 ufunda kusuka futhi ufaka imiphumela kusuka ekwakheni 3 ngaphambi kokucubungula phansi-phansi. Ukwakhiwa 1 kudlulisa ulwazi nemiphumela ngendlela efanayo yokwakhiwa kweC, okusebenzayo nokuthumela ekwakheni 2, okusebenzayo nokuthumela ekwakhiweni B.
I-ANN ngayinye (ukwakha) kusibonelo sethu isesha isici esihlukile kumininingwane engakhiwe (i-jumble of information) futhi sidlulisela imiphumela esakhiweni esilandelayo. Isakhiwo esilandelayo sihlanganisa (sifunda) ukukhipha (imiphumela) kusuka kwangaphambili. Njengoba idatha isetshenzwa yi-ANN ngayinye (isakhiwo), ihlelwa futhi ibhalwe (ihlukaniswe) ngesici esithile ukuze uma idatha ifinyelela ekugcineni kokuphuma (esiteji esiphezulu) se-ANN yokugcina (ukwakha), ihlukaniswa futhi ibhalwe phansi (ehlelwe ngaphezulu).
Intelligence Intelligence, Learning Machine, nokufunda Okujulile
Ukufunda okujulile kufanelana kanjani nesithombe esiphezulu se -intelligence yokufakelwa (AI) ne-ML? Ukufunda okujulile kwandisa amandla e-ML futhi kwandisa ububanzi bemisebenzi e-AI ekwazi ukwenza. Ngenxa yokuthi ukufunda okujulile kuncike ekusetshenzisweni kwamanetha e-neural nokubona izinto ngaphakathi kwamasethingi wedatha esikhundleni sezinhlelo zokusebenza ezilula zomsebenzi, ungathola futhi usebenzise imininingwane kusuka kumininingwane engakhiwe (eluhlaza) ngaphandle kwesidingo somhleli wokubhala ilebuli kuqala-isikhathi -kuqeda umsebenzi ongafaka amaphutha. Ukufunda okujulile kusiza amakhompiyutha athole kangcono futhi angcono ekusebenziseni idatha ukusiza zombili izinkampani kanye nabantu ngabanye.