Uma wake wazibuza ukuthi umshini ungabubona kanjani ubuso bakho noma uqonde ukuthi uthini, impendulo itholakala ekuthakaselekeni kobuchopho bomuntu. Lezi zakhiwo eziphilayo, ezakhiwe ama-neurons axhumene adlulisa izimpawu zikagesiZisebenze njengohlelo lokudala lokho esikwaziyo namuhla njengamanethiwekhi e-neural okwenziwa, okuyi-algorithms eklanyelwe ukuxazulula amaphazili ezibalo ayinkimbinkimbi.
Ezweni lanamuhla, sesijwayele ukuhlakanipha kokwenziwa okufana nama-supercomputer kanye nokusetshenziswa kwamandla okukhulu. Kodwa-ke, kuvela umkhuba othakazelisayo kakhulu ofuna... ukwandisa ukusebenza kahle kwamandlasizibuza ukuthi sidinga ngempela yini ukucubungula izigidi zeziphindaphindo ukuze sithole impendulo elula, ngaleyo ndlela silingise ukuphuphuma kwamandla kobuchopho bethu.
Ukwakhiwa kwezakhiwo zakudala: kusukela ku-perceptron kuya ku-Deep Learning
Ukuze siqonde ukuthi siyaphi, sidinga ukwazi ukuthi sivelaphi. Inethiwekhi eyisisekelo yezinzwa ngokuvamile ihlelwa ngamazinga amathathu. Konke kuqala ku- isendlalelo sokufakaYilapho uhlelo luthola khona idatha evela ngaphandle, luyihlaziye, bese luyithumela phambili. Bese kufika izendlalelo ezifihliweokungaba yinye noma inqwaba yazo; yilapho ukucubungula kwangempela kwenzeka khona, kuhlungwa ulwazi esigabeni ngasinye.
Ekugcineni, safika e- isendlalelo sokukhiphaokusinika isinqumo sokugcina. Kuye ngokuthi sifunani, singase sibe ne-node eyodwa yempendulo ethi yebo noma cha (ukuhlukaniswa kwe-binary), noma ama-node amaningana uma inkinga iyinkimbinkimbi kakhulu futhi idinga ukukhetha phakathi kwezigaba eziningi.
Uma sikhuluma ngakho Ukufunda OkujulileSibhekisela kumanethiwekhi anezendlalelo eziningi ezifihliwe kanye nezigidi zokuxhumeka. Kulo modeli, ubudlelwano phakathi kwama-neurons buchazwa yi- inani lezinombolo elibizwa ngokuthi isisindoInkinga ukuthi la manethiwekhi ayizinqwaba zedatha namandla, adinga izigidi zezibonelo ukuze kugwenywe amaphutha.
Inhliziyo yokuqeqeshwa: Ukusakazwa Kwangaphambili Nokungemuva
Ukuze inethiwekhi ifunde, kumele iqale izame ukuqagela. Inqubo Phambili Ukusabalalisa Ngokuyisisekelo luhambo olusuka kwesobunxele uye kwesokudla: idatha iyangena, iphindaphindwa yizisindo, kufakwe ukulungiswa okubizwa ngokuthi i-bias, bese idlula kumsebenzi wokwenza kusebenze (njenge-Sigmoid noma i-ReLu) ukuze kunqunywe ukuthi i-neuron "iyavutha" noma cha.
Kodwa njengoba inethiwekhi ingazi lutho ekuqaleni, izohluleka ngendlela emangalisayo. Yilapho-ke I-Backpropagationyilapho umlingo wenzeka khona. I-algorithm manje ihamba ukusuka kwesokudla iye kwesobunxele, ibala iphutha phakathi komphumela otholiwe nomphumela wangempela. okususelwe kwezibaloInethiwekhi ilungisa izisindo ukuze inciphise lelo phutha ekuphindaphindeni okulandelayo.
Imininingwane ebalulekile kule nqubo yi- izinga lokufundaUma iphakeme kakhulu, inethiwekhi ifunda ngokushesha kodwa ayinembile; uma iphansi kakhulu, inqubo iba engapheli. Umgomo wokugcina ukufinyelela inani eliphansi lomhlaba wonke lomsebenzi wezindleko, okwenza inethiwekhi ikwazi ukuhlanganisa ulwazi ngendlela efanele.
Uguquko lwamanethiwekhi e-neural angenasisindo
Yilapho izinto ziba mnandi khona. UProfesa Lizy K. John kanye nethimba lakhe baphula isisekelo ngokuphakamisa amanethiwekhi e-neural angenasisindoNakuba i-AI evamile incike ekwenzeni izigidigidi zokuphindaphinda (umsebenzi obiza kakhulu wehadiwe), le ndlela isebenzisa amathebula okubheka axhunyiwe.
Esikhundleni sokubala umsebenzi oyinkimbinkimbi wezibalo, inethiwekhi imane ibheka impendulo egciniwe ngokusekelwe kokufakwayo okubili. Kuyisistimu efana kakhulu nendlela abantu abacabanga ngayo: asiphindaphindi izinombolo emakhanda ethu ukuze sinqume ukuthi sizohamba ngakwesobunxele noma kwesokudla, kodwa kunalokho sicubungula izimpawu zokwenza kusebenze.
Imiphumela imangalisa nje. Emisebenzini efana nokuqapha kwezokwelapha (i-ECG noma i-EEG) noma ukuqashelwa kwamagama angukhiye, lawa manethiwekhi angaba okuncane ngokuphindwe kayinkulungwane futhi okusebenzayo kakhudlwanaCabanga ukusuka kumodeli engu-17-megabyte uye kwenye yama-kilobyte angu-14 kuphela. Lokhu kuvumela i-AI ukuthi iphile ngqo ngaphakathi kwenzwa ngokusebenzisa i- I-Edge Computing ngaphandle kwesidingo sokuthumela idatha efwini, okuyi-a Iphuzu elisekela ubumfihlo umsebenzisi.
Kubheke ku-AI eqhubekayo nelula
Akukhona nje kuphela ukususa izinsimbi; kukhona nomkhuba... amanethiwekhi e-neural alulaLokhu kwenzelwe ukudla amandla amancane kakhulu kanye nenkumbulo ngangokunokwenzeka, kusetshenziswa amasu anjengokuthi ukuqeqeshwa okuncane noma ukuthenaokuhlanganisa ukuqeda ukuxhumana okungafaki isandla emphumeleni wokugcina.
Le ndlela ibalulekile emikhakheni efana nokwakha ukuphathwa kwamandla noma ukupholisa isikhungo sedatha, lapho ukusetshenziswa kahle kungasho khona ukonga amandla kufika ku-30%Umgomo ucacile: ukuthi i-AI ingabi yinkinga yesimo sezulu, kodwa ibe yingxenye yesisombululo, ivumele amadivayisi ashibhile, njenge Ama-microcontroller e-ARM Cortex, yenza imisebenzi ehlakaniphile ngesikhathi sangempela.
Kusukela kuma-transformer anika amandla ama-chatbot esimanje kuya kumanethiwekhi e-neural agxile esithombeni (ama-CNN), indlela eya ekusebenzeni kahle iholela nciphisa ubunzima bekhompyuthaIntuthuko esuka ku-perceptron yakudala iye kwi-AI ekhiqizayo ibe nkulu kakhulu kudatha namandla, kodwa ubuhlakani bangempela manje bufuna ukwenza okuningi ngokuncane kakhulu.
Ukuguquka kusuka kuzakhiwo eziminyene, ezisindayo kuya ezinhlelweni ezisekelwe ekusesheni okubili kanye namamodeli aqoshiwe kwenza ubuhlakani bokwenziwa busebenze kumadivayisi amancane. Ngokuhlanganisa ukusebenza kahle kwamandla kanye nokwenza lula izibalo, ukucubungula idatha kuyashesha kakhulu futhi kuzinzile, kuvula indlela yokubala kwasendaweni, okuyimfihlo okungasathembeli kuma-GPU amakhulu ukuthi kusebenze kahle.
