Iyini i-Algorithm ye-AI? [Ividiyo Nombuzo]

Impendulo emfushane: I-algorithm ye-AI yindlela ikhompyutha eyisebenzisayo ukufunda amaphethini kusuka kudatha, bese yenza izibikezelo noma izinqumo isebenzisa imodeli eqeqeshiwe. Ayinawo umqondo oqinile othi “uma-ke”: iyaguquguquka njengoba ihlangana nezibonelo kanye nempendulo. Uma idatha ishintsha noma ithatha ukucwaswa, isengakhiqiza amaphutha aqinisekile.

Izinto ezibalulekile okufanele uzicabangele:

Izincazelo: Hlukanisa iresiphi yokufunda (i-algorithm) kumuntu oqaphile oqeqeshiwe (imodeli).

Umjikelezo Wokuphila: Phatha ukuqeqeshwa kanye nokucabanga njengokuhlukile; ukwehluleka kuvame ukuvela ngemva kokusetshenziswa.

Ukuziphendulela: Nquma ukuthi ubani obuyekeza amaphutha nokuthi kwenzekani uma uhlelo luphambuka.

Ukumelana nokusebenzisa kabi: Qaphela ukuvuza, ukukhetha okuzenzakalelayo, kanye nokudlala imidlalo ye-metric okungakhulisa imiphumela.

Ukuhlolwa: Landelela imithombo yedatha, izilungiselelo, kanye nokuhlolwa ukuze izinqumo zihlale zingaphikiswana kamuva.

Izihloko ongase uthande ukuzifunda ngemva kwalesi:

🔗 Iyini i-AI ethics
Izimiso ze-AI enomthwalo wemfanelo: ukulunga, ukucaca, ukuzibophezela, kanye nokuphepha.

🔗 Kuyini ukubandlulula kwe-AI
Indlela idatha enobandlululo ephazamisa ngayo imiphumela ye-AI nokuthi ingalungiswa kanjani.

🔗 Iyini i-AI scalability
Izindlela zokukhulisa izinhlelo ze-AI: idatha, ukubala, ukuthunyelwa, kanye nezinhlelo zokusebenza.

🔗 Yini echazwe yi-AI
Kungani amamodeli ahunyushwayo ebalulekile ekuthembekeni, ekulungiseni amaphutha, nasekulandeleni imithetho.

Iyini ngempela i-algorithm ye-AI? 🧠

I -algorithm ye-AI inqubo esetshenziswa yikhompyutha ukuze:

• Funda kudatha (noma impendulo)

• Yazi amaphethini

• Yenza izibikezelo noma izinqumo

• Thuthukisa ukusebenza ngolwazi [1]

Ama-algorithm ajwayelekile anjengokuthi: “Hlunga lezi zinombolo ngokulandelana okukhuphukayo.” Sula izinyathelo, umphumela ofanayo njalo.

Ama-algorithm e-AI afana kakhulu nalawa: “Nazi izibonelo eziyisigidi. Sicela uthole ukuthi liyini ‘ikati’.” Bese lakha iphethini yangaphakathi evame ukusebenza . Ngokuvamile. Ngezinye izikhathi libona umcamelo othambile bese limemeza lithi “Ikati!” ngokuzethemba okuphelele. 🐈⬛

I-Algorithm ye-AI vs Imodeli ye-AI: umehluko abantu abawubonayo 😬

Lokhu kususa ukudideka okuningi ngokushesha:

• I-algorithm ye-AI = indlela yokufunda / indlela yokuqeqesha
  ("Lena yindlela esizibuyekeza ngayo kusukela kudatha.")

• Imodeli ye-AI = into eqeqeshiwe oyisebenzisayo kokufakwayo okusha
  (“Lena yinto ebikezela manje.”) [1]

Ngakho-ke, i-algorithm ifana nenqubo yokupheka, futhi imodeli ukudla okuqediwe 🍝. Mhlawumbe isingathekiso esintengantengayo kancane, kodwa siyasebenza.

Futhi, i-algorithm efanayo ingakhiqiza amamodeli ahlukene kakhulu kuye ngokuthi:

• idatha oyiphakelayo

• izilungiselelo ozikhethayo

• uqeqesha isikhathi esingakanani

• ukuthi isethi yakho yedatha ayihlelekile kangakanani (i-spoiler: cishe ayihlelekile njalo)

Kungani i-algorithm ye-AI ibalulekile (noma ngabe awuyena "uchwepheshe") 📌

Ngisho noma ungalokothi ubhale umugqa wekhodi, ama-algorithm e-AI asakuthinta. Kuningi.

Cabanga: izihlungi zogaxekile, ukuhlolwa kokukhwabanisa, izincomo, ukuhumusha, ukwesekwa kwezithombe zezokwelapha, ukwenza ngcono umzila, kanye nokuthola amaphuzu engozi. (Hhayi ngoba i-AI "iyaphila," kodwa ngoba ukuqashelwa kwephethini ngezinga kuyigugu ezindaweni eziyisigidi ezibalulekile ngokuthula.)

Futhi uma wakha ibhizinisi, uphatha ithimba, noma uzama ukungaphazanyiswa yi-jargon, ukuqonda ukuthi iyini i -algorithm ye-AI kukusiza ukuthi ubuze imibuzo engcono:

• Khomba ukuthi yiluphi ulwazi uhlelo olufunde kulo.

• Hlola ukuthi ukucwasa kulinganiswa futhi kuncishiswa kanjani.

• Chaza ukuthi kwenzekani uma uhlelo lungalungile.

Ngoba kuzoba yiphutha ngezinye izikhathi. Lokho akukhona ukuphelelwa yithemba. Yilokho okuyiqiniso.

Indlela i-algorithm ye-AI "efunda ngayo" (ukuqeqeshwa vs ukuphetha) 🎓➡️🔮

Izinhlelo eziningi zokufunda komshini zinezigaba ezimbili ezinkulu:

1) Ukuqeqeshwa (isikhathi sokufunda)

Ngesikhathi sokuqeqeshwa, i-algorithm:

• ubona izibonelo (idatha)

• yenza izibikezelo

• kulinganisa ukuthi akulungile kangakanani

• ilungisa amapharamitha angaphakathi ukuze inciphise iphutha [1]

2) Ukuphetha (ukusebenzisa isikhathi)

Isiphetho yilapho imodeli eqeqeshiwe isetshenziswa kokufakwayo okusha:

• hlela i-imeyili entsha njengogaxekile noma cha

• bikezela isidingo ngesonto elizayo

• ilebula isithombe

• dala impendulo [1]

Ukuqeqeshwa “ukufunda.” Ukuphetha “ukuhlolwa.” Ngaphandle kokuthi ukuhlolwa akupheli futhi abantu baqhubeka nokushintsha imithetho phakathi nohlelo. 😵

Imindeni emikhulu yezitayela ze-algorithm ze-AI (enokuqonda okucacile kwesiNgisi) 🧠🔧

Ukufunda okuqondisiwe 🎯

Unikeza izibonelo ezinelebula ezifana nalezi:

• “Lokhu kungugaxekile” / “Lokhu akusikho ugaxekile”

• “Leli khasimende liphazamisekile” / “Leli khasimende lihlale”

I-algorithm ifunda ukumepha kusuka kokufakwayo → okukhiphayo. Kuvamile kakhulu. [1]

Ukufunda okungaqondiswanga 🧊

Azikho amalebula. Uhlelo lufuna isakhiwo:

• amaqembu amakhasimende afanayo

• amaphethini angavamile

• izihloko kumadokhumenti [1]

Ukufunda kokuqinisa 🕹️

Uhlelo lufunda ngokuzama nokwenza amaphutha, luqondiswa yimivuzo. (Kuhle kakhulu uma imivuzo icacile. Iyaxakaxaka uma ingekho.) [1]

Ukufunda okujulile (amanethiwekhi e-neural) 🧠⚡

Lokhu kufana kakhulu nomndeni wobuchwepheshe kunokuba kube yi-algorithm eyodwa. Kusebenzisa izethulo ezinezingqimba futhi kungafunda amaphethini ayinkimbinkimbi kakhulu, ikakhulukazi embonweni, enkulumweni, kanye nolimi. [1]

Ithebula lokuqhathanisa: imindeni ethandwayo ye-algorithm ye-AI ngokubuka nje 🧩

Akuyona "uhlu olungcono kakhulu" - kufana nemephu ngakho-ke uyeka ukuzizwa sengathi konke kuyisisombululo esikhulu se-AI.

Yini eyenza inguqulo enhle ye-algorithm ye-AI? ✅🤔

I-algorithm "enhle" ye-AI akuyona ekhangayo ngokuzenzakalelayo. Empeleni, uhlelo oluhle luvame ukuba:

• Kunembile ngokwanele emgomweni wangempela (akuphelele - kuyazuzisa)

• Iqinile (ayiwi lapho idatha ishintsha kancane)

• Kuchazeka ngokwanele (hhayi ngempela ukuthi kusobala, kodwa hhayi umgodi omnyama ophelele)

• Kuhlolwe ngokulingana nangokubandlulula (idatha egobile → imiphumela egobile)

• Isebenza kahle (akukho supercomputer yomsebenzi olula)

• Iyagcinwa (iyaqashelwa, iyabuyekezwa, iyathuthukiswa)

Ikesi elincane elisebenzayo elisheshayo (ngoba yilapho izinto zibonakala khona)

Cabanga ngemodeli ye-churn "emangalisayo" ekuhlolweni ... ngoba ngephutha ithole i-proxy "yekhasimende elivele lixhunyaniswe yithimba lokugcina." Lokho akuwona umlingo wokubikezela. Lokho ukuvuza. Kuzobukeka kuyiqhawe uze uyisebenzise, ​​​​bese uyibeka ebusweni ngokushesha. 😭

Indlela esihlulela ngayo ukuthi i-algorithm ye-AI "ilungile" 📏✅

Awugcini nje ngokukubuka ngamehlo (kahle, abanye abantu bayakubona, bese kulandela inhlekelele).

Izindlela zokuhlola ezivamile zifaka:

• Ukunemba

• Ukunemba / ukubuyisa

• Isikolo se-F1 (silinganisa ukunemba/ukukhumbula) [2]

• I-AUC-ROC (ikhwalithi yokubeka ezingeni lesigaba se-binary) [3]

• Ukulinganisa (ukuthi ukuzethemba kuyafana yini neqiniso)

Bese kuba nokuhlolwa kwangempela:

• Ingabe kuyabasiza abasebenzisi?

• Ingabe kunciphisa izindleko noma ingozi?

• Ingabe kudala izinkinga ezintsha (izixwayiso ezingamanga, ukwenqatshwa okungafanele, izindlela zokusebenza ezididayo)?

Ngezinye izikhathi imodeli “embi kancane” ephepheni ingcono ekukhiqizweni ngoba izinzile, iyachazeka, futhi kulula ukuyiqapha.

Izingibe ezivamile (okwaziwa nangokuthi amaphrojekthi e-AI ahamba kanjani buthule eceleni) ⚠️😵💫

Ngisho namaqembu aqinile afinyelela lokhu:

• Ukufaka ngokweqile (kuhle kakhulu kudatha yokuqeqesha, kubi kakhulu kudatha entsha) [1]

• Ukuvuza kwedatha (kuqeqeshwe ngolwazi ongeke ube nalo ngesikhathi sokubikezela)

• Izinkinga zobandlululo kanye nobulungisa (idatha yomlando iqukethe ukungalungi komlando)

• Ukuzulazula komqondo (umhlaba uyashintsha; imodeli ayishintshi)

• Izilinganiso ezingalungile (wenza ngcono ukunemba; abasebenzisi bakhathalela okunye)

• Uvalo lwebhokisi elimnyama (akekho ongachaza isinqumo uma sibalulekile ngokuzumayo)

Enye inkinga ecashile: ukucwasa okuzenzakalelayo - abantu bayaluthemba ngokweqile uhlelo ngoba lukhipha izincomo eziqinisekile, ezinganciphisa ukuqapha kanye nokuhlola okuzimele. Lokhu kubhalwe phansi ocwaningweni lokusekela izinqumo, kufaka phakathi izimo zezempilo. [4]

"I-AI ethembekile" akuyona into ehlekisayo - iwuhlu lokuhlola 🧾🔍

Uma uhlelo lwe-AI luthinta abantu bangempela, ufuna okungaphezu kokuthi "lunembile ngokwezinga lethu."

Uhlaka oluqinile luwukuphathwa kwengozi yomjikelezo wokuphila: uhlelo → ukwakha → ukuhlola → ukusebenzisa → ukuqapha → ukubuyekeza. Uhlaka Lokuphathwa Kwengozi lwe-AI lwe-NIST luveza izici ze-AI "ethembekile" njenge- esemthethweni nethembekile, ephephile, evikelekile neqinile, enesibopho nesobala, echazekayo neqondakalayo, ethuthukisiwe yobumfihlo, kanye nelungile (ubandlululo oluyingozi olulawulwayo). [5]

Ukuhumusha: ubuza ukuthi iyasebenza yini.
Ubuza nokuthi yehluleka yini ngokuphephile, nokuthi ungakubonisa yini lokho.

Izinto Ezibalulekile Okufanele Uzicabangele 🧾✅

Uma ungathathi lutho olunye kulokhu:

• I-algorithm ye-AI = indlela yokufunda, iresiphi yokuqeqesha

• Imodeli ye-AI = umphumela oqeqeshiwe owusebenzisayo

• I-AI enhle ayigcini nje ngokuba "ihlakaniphile" - ithembekile, iyaqashelwa, ihlolwe ukucwasa, futhi ifanele umsebenzi.

• Ikhwalithi yedatha ibaluleke kakhulu kunalokho abantu abaningi abafuna ukukuvuma

• I-algorithm engcono kakhulu ngokuvamile yileyo exazulula inkinga ngaphandle kokudala izinkinga ezintathu ezintsha 😅

Isibonelo sangempela: Ukuhlola i-algorithm yokubikezela i-churn ngaphambi kokuqaliswa 📉🧪

Isimo

Cabanga ngenkampani encane yesofthiwe yokubhalisa efuna ukubikezela ukuthi yimaphi amakhasimende angase akhansele phakathi nezinsuku ezingu-30 ezizayo.

Ithimba linemininingwane yamakhasimende yezinyanga ezingu-18: imvamisa yokungena ngemvume, amathikithi okusekela, uhlobo lohlelo, ukubambezeleka kokukhokha, ukusetshenziswa komkhiqizo, izinsuku zokuvuselela, nokuthi ngabe ikhasimende ngalinye ligcine likhanseliwe. Umhlaziyi wedatha wakha izinguqulo ezimbili zemodeli: isisekelo sokuhlehla kwempahla esilula kanye nemodeli yokukhulisa i-gradient eyinkimbinkimbi kakhulu.

Umgomo akukhona “ukuthola i-algorithm ehlakaniphe kakhulu.” Umgomo ukuthola imodeli esiza ithimba lempumelelo yamakhasimende ukuthi lixhumane nama-akhawunti afanele kusenesikhathi, ngaphandle kokuchitha ingxenye yesonto lilandela ama-alamu angamanga.

Okudingwa umsebenzi wokusebenza

Ngaphambi kokukhetha i-algorithm, ithimba lilungiselela:

• Isethi yedatha yokuqeqesha ehlanzekile enomugqa owodwa ngekhasimende ngalinye

• Ilebula ecacile: “kukhanseliwe zingakapheli izinsuku ezingu-30” yebo/cha

• Uhlu lwamakholomu olutholakala ngaphambi kosuku lokubikezela

• Isethi yokuhlolwa kokubamba kusukela ezinyangeni ezintathu zakamuva

• Inqubo elula yokubuyekeza izinto ezinhle nezimbi ezingamanga

• Umthetho wokuthi akukho maphuzu obungozi bokukhansela okuzenzakalelayo aboniswa kumakhasimende

Ukuhlola okubalulekile: susa noma yini evuza impendulo. Isibonelo, "isaphulelo esinikezwa yithimba lokugcina" akufanele sisetshenziswe uma lokho kwenzeka kuphela ngemva kokuba othile esesolwa ngokukhansela.

Isibonelo semiyalelo

Sebenzisa lo myalelo uma ucela umsizi we-AI noma umhlaziyi ukuthi abukeze ukusethwa:

Buyekeza lo mklamo wesethi yedatha yokubikezela ye-churn. Khomba noma yimaphi amakholomu angabangela ukuvuza kwedatha, noma yiziphi izici ezingase zishintshe izibikezelo ngokungafanele, kanye nanoma yiziphi izilinganiso okufanele sizilandele ngaphambi kokusetshenziswa. Imodeli izosetshenziswa yithimba lempumelelo yamakhasimende ukuze libeke phambili ukufinyelela, hhayi ukwenza izinqumo ze-akhawunti ngokuzenzakalelayo.

Indlela yokuyihlola

Hlola imodeli ngemibuzo efana nale:

• Ingabe imodeli isasebenza kudatha yezinyanga ezintathu zakamuva?

• Yimaphi amakholomu ayi-10 athonya kakhulu izibikezelo?

• Ingabe amakhasimende asebenzisa izinhlelo ezishibhile avame ukuhlatshwa amamaki ngezizathu ezingahlobene nengozi yangempela yokungakhokhi?

• Mangaki amakhasimende anezimpawu ithimba elizoba nesikhathi sokuxhumana nawo isonto ngalinye?

• Kwenzekani uma ukusetshenziswa komkhiqizo kwehla kuwo wonke umuntu ngesikhathi samaholide?

Ukuhlolwa okuhle kuyasebenziseka, hhayi nje ngokwezibalo. Uma imodeli ikhomba amakhasimende angu-600 ngesonto futhi ithimba lingaxhumana no-80 kuphela, i-algorithm ingase ibe nembambo kodwa isaklanywe kabi ukuze ihambisane nomsebenzi.

Umphumela

Umphumela obonisayo: ngokusekelwe kusethi yokuhlola yama-akhawunti amakhasimende ayi-1,000, imodeli elula yokubuyela emuva kwe-logistic ifinyelele ekukhunjweni okungu-71% kanye nokunemba okungu-42%. Imodeli yokukhulisa i-gradient ifinyelele ekukhunjweni okungu-78% kanye nokunemba okungu-48%, kodwa yadinga ukubuyekezwa okwengeziwe ngoba izici zayo eziphezulu zazihlanganisa izingozi ezimbili zokuvuza ezingaba khona.

Ngemva kokususa amakholomu avame ukuvuza, imodeli yokukhulisa i-gradient yehle kancane yaya ekukhunjulweni okungu-74% kanye nokunemba okungu-46%. Lokho kwakusewusizo: ekubuyekezweni kwamasonto onke kwama-akhawunti ayi-100 anemaki, ithimba lingalindela amakhasimende angaba ngu-46 asengozini enkulu esikhundleni sokuxhumana nama-akhawunti ngokungahleliwe.

Isikhathi esilinganiselwe: uma ukubuyekezwa kwe-akhawunti ngesandla kuthatha imizuzu eyi-6 ngekhasimende ngalinye, ukubuyekezwa kwama-akhawunti ayi-100 akhethwe ngokungahleliwe kungathatha amahora ayi-10. Ukusebenzisa imodeli ukufaka ohlwini lwezingozi ezingase ziphazamiseke kugcina isikhathi sokubuyekezwa sisemahoreni ayi-10 kodwa kwandisa inani lemizamo yokufinyelela ewusizo. Isilinganiso sokuqinisekisa silula: landelela ukuthi mangaki amakhasimende aphawuliwe axhumaniswe nawo, ukuthi mangaki asengozini yangempela, nokuthi mangaki agcine okubhalisele kwawo ngemva kokufinyelela efwini.

Yini engase ihambe kabi

Imodeli ingabukeka ingcono kunalokho eyikho ngempela uma isethi yedatha ifaka ulwazi lwesikhathi esizayo, njengokunikezwa kokugcinwa, izimpendulo zocwaningo lokukhansela, noma amanothi okusekela abhalwe ngemuva kokuba ikhasimende selivele linqume ukuhamba.

Ithimba lingaphinde liwele ekubandlululeni okuzenzakalelayo. Isilinganiso "esiyingozi kakhulu" kufanele sibangele ukubuyekezwa komuntu, hhayi i-imeyili yerobhothi ecasula amakhasimende athembekile.

Elinye iphutha ukuphishekela ukunemba kuphela. Uma amakhasimende angu-5% kuphela ekhansela, imodeli evilaphayo ebikezela ukuthi “akekho ozokhansela” ingase ibukeke inembile ngenkathi inganikezi inzuzo engokoqobo.

Ukudla okuwusizo

I-algorithm ye-AI engcono kakhulu yileyo esindayo uma ixhumana nomsebenzi obukhoma. Qala ngesisekelo, hlola ukuvuza, hlola idatha yakamuva, kala ama-alamu angamanga, bese uqinisekisa ukuthi abantu bayazi ukuthi kufanele babuze nini amaphuzu.

Imibuzo Evame Ukubuzwa

Iyini i-algorithm ye-AI ngamagama alula?

I-algorithm ye-AI iyindlela ikhompyutha eyisebenzisayo ukufunda amaphethini kusuka kudatha nokwenza izinqumo. Esikhundleni sokuthembela emithethweni engaguquki ethi “uma-ke”, iyazilungisa ngemva kokubona izibonelo eziningi noma ukuthola impendulo. Inhloso ukuthuthuka ekubikezeleni noma ekuhlukaniseni okufakwayo okusha ngokuhamba kwesikhathi. Inamandla, kodwa isengenza amaphutha aqinisekile.

Uyini umehluko phakathi kwe-algorithm ye-AI kanye nemodeli ye-AI?

I-algorithm ye-AI inqubo yokufunda noma iresiphi yokuqeqesha - indlela uhlelo oluzivuselela ngayo kusuka kudatha. Imodeli ye-AI ingumphumela oqeqeshiwe owusebenzisayo ukuze wenze izibikezelo kokufakwayo okusha. I-algorithm efanayo ye-AI ingakhiqiza amamodeli ahlukene kakhulu kuye ngedatha, isikhathi sokuqeqeshwa, kanye nezilungiselelo. Cabanga "ngenqubo yokupheka" uma kuqhathaniswa "nokudla okuqediwe."

I-algorithm ye-AI ifunda kanjani ngesikhathi sokuqeqeshwa uma kuqhathaniswa nokuphetha?

Ukuqeqeshwa yilapho i-algorithm ifunda khona: ibona izibonelo, yenza izibikezelo, ikala amaphutha, futhi ilungisa amapharamitha angaphakathi ukuze inciphise lelo phutha. Ukuqagela yilapho imodeli eqeqeshwe isetshenziswa kokufakwayo okusha, njengokuhlukanisa ugaxekile noma ukulebula isithombe. Ukuqeqeshwa kuyisigaba sokufunda; ukuqagela kuyisigaba sokusebenzisa. Izinkinga eziningi zivela kuphela ngesikhathi sokuqagela ngoba idatha entsha iziphatha ngendlela ehlukile kulokho uhlelo olukufundile.

Yiziphi izinhlobo eziyinhloko zama-algorithms e-AI (aqondiswayo, angaqondiswayo, aqiniswayo)?

Ukufunda okuqondisiwe kusebenzisa izibonelo ezinelebula ukuze kufundwe ukumepha kusuka kokufakwayo kuya kokukhiphayo, njengogaxekile vs hhayi ugaxekile. Ukufunda okungaqondisiwe akunazo amalebula futhi kubheka isakhiwo, njengeqoqo noma amaphethini angavamile. Ukufunda okuqinisiwe kufunda ngokuzama nokwenza iphutha kusetshenziswa imivuzo. Ukufunda okujulile kuwumndeni obanzi wamasu enethiwekhi yezinzwa angabamba amaphethini ayinkimbinkimbi, ikakhulukazi emisebenzini yombono kanye nolimi.

Wazi kanjani ukuthi i-algorithm ye-AI "ilungile" empilweni yangempela?

I-algorithm enhle ye-AI akuyona eyinkimbinkimbi kakhulu ngokuzenzakalelayo - yileyo ehlangabezana nomgomo ngokuthembekile. Amaqembu abheka amamethrikhi afana nokunemba, ukunemba/ukukhumbula, i-F1, i-AUC-ROC, kanye nokulinganisa, bese ehlola ukusebenza kanye nomthelela ongezansi kuzilungiselelo zokufakwa. Ukuzinza, ukuchaza, ukusebenza kahle, kanye nokugcinwa kubalulekile kakhulu ekukhiqizweni. Ngezinye izikhathi imodeli ebuthakathaka kancane ephepheni iyaphumelela ngoba kulula ukuyiqapha nokuyethemba.

Kuyini ukuvuza kwedatha, futhi kungani kuphula amaphrojekthi e-AI?

Ukuvuza kwedatha kwenzeka lapho imodeli ifunda kolwazi olungeke lutholakale ngesikhathi sokubikezela. Lokhu kungenza imiphumela ibukeke imangalisa ekuhlolweni ngenkathi yehluleka kabi ngemva kokusetshenziswa. Isibonelo esivamile ukusebenzisa ngengozi izimpawu ezibonisa izenzo ezithathwe ngemuva komphumela, njengokuxhumana kwethimba lokugcina kumodeli ye-churn. Ukuvuza kudala "ukusebenza mbumbulu" okunyamalala emsebenzini wangempela.

Kungani ama-algorithm e-AI eba mabi ngokuhamba kwesikhathi noma ngabe ayenembile lapho eqaliswa?

Idatha iyashintsha ngokuhamba kwesikhathi - amakhasimende aziphatha ngendlela ehlukile, izinqubomgomo ziyashintsha, noma imikhiqizo iyashintsha - okubangela ukuzulazula komqondo. Imodeli ihlala ifana ngaphandle kokuthi uqaphe ukusebenza futhi uyibuyekeze. Ngisho nokushintsha okuncane kunganciphisa ukunemba noma kwandise ama-alamu angamanga, ikakhulukazi uma imodeli yayibuthakathaka. Ukuhlola okuqhubekayo, ukuqeqeshwa kabusha, kanye nemikhuba yokusabalalisa ngokucophelela kuyingxenye yokugcina uhlelo lwe-AI luphilile.

Yiziphi izingibe ezivame kakhulu lapho kusetshenziswa i-algorithm ye-AI?

Ukufaka ngokweqile kuyinto enkulu: imodeli isebenza kahle kakhulu kudatha yokuqeqesha kodwa ayisebenzi kahle kudatha entsha. Izinkinga zokukhetha kanye nokungakhethi zingavela ngoba idatha yomlando ivame ukuqukatha ukungakhethi komlando. Izilinganiso ezingaqondile zingase ziqede amaphrojekthi - ukuthuthukisa ukunemba lapho abasebenzisi bekhathalela okunye. Enye ingozi ecashile ukubandlulula okuzenzakalelayo, lapho abantu bethemba ngokweqile imiphumela yemodeli eqinisekile futhi bayeke ukuhlola kabili.

Kusho ukuthini ukuthi “i-AI ethembekile” empeleni?

I-AI ethembekile ayigcini nje ngokuba "ukunemba okuphezulu" - iyindlela yokuphila: ukuhlela, ukwakha, ukuhlola, ukusebenzisa, ukuqapha, nokubuyekeza. Empeleni, ufuna izinhlelo ezisebenzayo nezithembekile, eziphephile, eziphephile, eziphendulayo, ezichazayo, eziqaphela ubumfihlo, nezihlolwe ngokukhetha. Ufuna nezindlela zokwehluleka eziqondakalayo nezilulamayo. Umqondo oyinhloko ukukwazi ukukhombisa ukuthi iyasebenza futhi yehluleka ngokuphepha, hhayi nje ukuthemba ukuthi iyasebenza.

Izinkomba

1. Onjiniyela be-Google - Isichazamazwi Sokufunda Komshini

2. ukufunda ngokunemba, ukukhumbula, isilinganiso se-F

3. i-scikit-learn - isikolo se-ROC AUC

4. UGoddard nabanye - Ukubuyekezwa okuhlelekile kokukhetha okuzenzakalelayo (umbhalo ogcwele we-PMC)

5. I-NIST - Uhlaka Lokuphathwa Kwengozi ye-AI (AI RMF 1.0) PDF

Thola i-AI Yakamuva Esitolo Esisemthethweni Somsizi we-AI

Mayelana NATHI