Iminyaka eminingi, ukufunda okujulile bekulokhu kuyizihloko zezindaba kwezobuchwepheshe. Futhi, kulula ukuqonda ukuthi kungani.
Leli gatsha lezobunhloli bokwenziwa liguqula imikhakha kusukela kwezokunakekelwa kwezempilo kuye kwezamabhange kuya kwezokuthutha, okwenza intuthuko ebingacatshangwa ngaphambilini.
Ukufunda okujulile kwakhelwe phezu kwesethi yama-algorithms ayinkimbinkimbi afunda ukukhipha nokubikezela amaphethini ayinkimbinkimbi kusuka kumthamo omkhulu wedatha.
Sizobheka ama-algorithms wokufunda ajulile angu-15 angcono kakhulu kulokhu okuthunyelwe, kusukela ku-Convolutional Neural Networks kuya kuma-Generative Adversarial Networks kuya kumanethiwekhi enkumbulo yesikhathi esifushane.
Lokhu okuthunyelwe kuzonikeza imininingwane ebalulekile yokuthi ngabe uyi-a oqalayo noma uchwepheshe wokufunda okujulile.
1. Transformer Networks
Amanethiwekhi eTransformer ashintshile umbono wekhompyutha kanye nezinhlelo zokusebenza zokucubungula ulimi lwemvelo (NLP). Bahlaziya idatha engenayo futhi basebenzise izinqubo zokunaka ukuze bathwebule ubudlelwano bebanga elide. Lokhu kuwenza asheshe kunamamodeli ajwayelekile alandelanayo.
Amanethiwekhi e-Transformer achazwe okokuqala encwadini ethi “Ukunakwa Yikho Konke Okudingayo” kaVaswani et al.
Aqukethe isifaki khodi nesikhikhoda (2017). Imodeli ye-transformer ibonise ukusebenza ezinhlobonhlobo zezinhlelo zokusebenza ze-NLP, kuhlanganise ukuhlaziywa kwemizwa, ukuhlukaniswa kombhalo, nokuhumusha ngomshini.
Amamodeli asuselwa ku-Transformer angasetshenziswa futhi embonweni wekhompyutha ezinhlelweni zokusebenza. Bangakwazi ukuqaphela into kanye namagama-ncazo wesithombe.
2. Amanethiwekhi Enkumbulo Yesikhathi Esifushane (LSTMs)
Amanethiwekhi Enkumbulo Yesikhathi Esifushane Eside (LSTMs) awuhlobo lwe inethiwekhi ye-neural eyakhelwe ngokukhethekile ukuphatha okokufaka okulandelanayo. Babizwa ngokuthi "isikhathi esifushane eside" ngoba bangakwazi ukukhumbula ulwazi lwakudala ngenkathi bekhohlwa ulwazi olungadingekile.
Ama-LSTM asebenza “ngamasango” athile alawula ukuhamba kolwazi ngaphakathi kwenethiwekhi. Kuye ngokuthi ulwazi lubhekwa njengolubalulekile noma cha, la masango angalungenisa noma aluvimbele.
Le nqubo yenza ama-LSTM akhumbule noma akhohlwe ulwazi lwezinyathelo zesikhathi esidlule, olubalulekile emisebenzini efana nokuqaphela inkulumo, ukucutshungulwa kolimi lwemvelo, nokuqagela kochungechunge lwesikhathi.
Ama-LSTM azuzisa kakhulu kunoma yisiphi isimo lapho unedatha elandelanayo okufanele ihlolwe noma ibikezelwe. Zivame ukusetshenziswa kusofthiwe yokuqaphela izwi ukuguqula amagama akhulunyiwe abe umbhalo, noma ku stock market ukuhlaziywa kokubikezela izintengo zesikhathi esizayo ngokusekelwe kudatha yangaphambilini.
3. I-Self Organising Maps (SOMs)
Ama-SOM awuhlobo lokwenziwa inethiwekhi ye-neural engafunda futhi imele idatha eyinkimbinkimbi endaweni ene-dimensional ephansi. Indlela isebenza ngokuguqula idatha yokufaka enobukhulu obuphezulu ibe igridi enezinhlangothi ezimbili, ngeyunithi ngayinye noma i-neuron emele ingxenye ehlukile yesikhala sokufakwayo.
Ama-neurons axhunyaniswe ndawonye futhi akhe isakhiwo se-topological, esiwavumela ukuthi afunde futhi alungise idatha yokufaka. Ngakho-ke, i-SOM isuselwe ekufundeni okungagadiwe.
I-algorithm ayidingi idatha enelebula ukufunda ku. Kunalokho, isebenzisa izici zezibalo zedatha yokufaka ukuze kutholwe amaphethini nokuhlobana phakathi kokuhluka.
Ngesikhathi sokuqeqeshwa, ama-neurons ancintisana ukuze abe inkomba engcono kakhulu yedatha yokufaka. Futhi, bazihlela ngokwabo babe isakhiwo esinengqondo. Ama-SOM anezinhlobonhlobo zezinhlelo zokusebenza, okuhlanganisa ukubonwa kwesithombe nenkulumo, ukumbiwa kwedatha, nokuqashelwa kwephethini.
Ziwusizo for ukubona ngeso lengqondo idatha eyinkimbinkimbi, ukuhlanganisa amaphuzu edatha ahlobene, nokuthola okungavamile noma okungaphandle.
4. Ukufunda Okujulile Kokuqiniswa
Deep Ukuqinisa Ukufunda wuhlobo lokufunda komshini lapho i-ejenti iqeqeshelwa ukwenza izinqumo ngokusekelwe ohlelweni lokuklomelisa. Isebenza ngokuvumela i-ejenti ukuthi ihlanganyele nendawo eyizungezile futhi ifunde ngokuzama nangephutha.
I-ejenti iklonyeliswa ngaso sonke isenzo esiyenzayo, futhi inhloso yayo ukufunda indlela yokuthuthukisa izinzuzo zayo ngokuhamba kwesikhathi. Lokhu kungase kusetshenziselwe ukufundisa abenzeli ukudlala imidlalo, ukushayela izimoto, ngisho nokuphatha amarobhothi.
I-Q-Learning iyindlela eyaziwa kakhulu Yokuqiniswa Okujulile Yokufunda. Isebenza ngokuhlola ukubaluleka kokwenza isenzo esithile esimweni esithile nokubuyekeza lokho kulinganisa njengoba i-ejenti isebenzisana nemvelo.
Umenzeli ube esesebenzisa lezi zilinganiso ukuze anqume ukuthi yisiphi isenzo esingase siphumele emvuzweni omkhulu kakhulu. I-Q-Learning isetshenziselwe ukufundisa abenzeli ukudlala imidlalo ye-Atari, kanye nokuthuthukisa ukusetshenziswa kwamandla ezikhungweni zedatha.
I-Deep Q-Networks ingenye indlela yokufunda edumile yokuqinisa i-Deep (DQN). Ama-DQN afana ne-Q-Learning ngoba alinganisela amanani esenzo esebenzisa inethiwekhi ye-neural ejulile kunetafula.
Lokhu kubenza bakwazi ukubhekana nezilungiselelo ezinkulu, eziyinkimbinkimbi ngezenzo eziningi ezihlukile. Ama-DQN asetshenziselwe ukuqeqesha ama-agent ukuthi adlale imidlalo efana ne-Go ne-Dota 2, kanye nokudala amarobhothi angakwazi ukufunda ukuhamba.
5. AmaNethiwekhi Emizwa Ejwayelekile (ama-RNN)
Ama-RNN awuhlobo lwenethiwekhi ye-neural engacubungula idatha elandelanayo kuyilapho igcina isimo sangaphakathi. Kubheke njengokufana nomuntu ofunda incwadi, lapho igama ngalinye ligaywa ngokuqondene nalawo afika ngaphambi kwayo.
Ngakho-ke ama-RNN alungele imisebenzi efana nokuqaphela inkulumo, ukuhumusha ulimi, ngisho nokubikezela igama elilandelayo emshweni.
Ama-RNN asebenza ngokusebenzisa amaluphu empendulo ukuze axhume okukhiphayo kwesinyathelo ngasinye sokubuyela emuva kokokufaka kwesinyathelo esilandelayo. Lokhu kuvumela inethiwekhi ukuthi isebenzise ulwazi lwesinyathelo sesikhathi sangaphambili ukuze yazise ukuqagela kwayo ngezinyathelo zesikhathi esizayo. Ngeshwa, lokhu kusho nokuthi ama-RNN asengozini yodaba olushabalalayo lwe-gradient, lapho ama-gradient asetshenziselwa ukuqeqeshwa eba mancane kakhulu futhi inethiwekhi idonsa kanzima ukufunda ubudlelwano besikhathi eside.
Ngaphandle kwalokhu kuvinjelwa okusobala, ama-RNN athole ukusetshenziswa ezinhlobonhlobo zezinhlelo zokusebenza. Lezi zinhlelo zokusebenza zihlanganisa ukucutshungulwa kolimi lwemvelo, ukubonwa kwenkulumo, kanye nokukhiqizwa komculo.
ukuhumusha nge-Google, isibonelo, isebenzisa isistimu esekelwe ku-RNN ukuze ihumushe kuzo zonke izilimi, kuyilapho uSiri, umsizi obonakalayo, esebenzisa isistimu esekelwe ku-RNN ukuze athole izwi. Ama-RNN nawo asetshenziselwe ukubikezela izintengo zesitoko nokudala umbhalo ongokoqobo kanye nemifanekiso.
6. Amanethiwekhi we-Capsule
I-Capsule Networks iwuhlobo olusha lwedizayini yenethiwekhi ye-neural engahlonza amaphethini nokuhlotshaniswa kwedatha ngempumelelo kakhudlwana. Bahlela ama-neurons abe "amaphilisi" ahlanganisa izici ezithile zokufakwayo.
Ngale ndlela bangenza izibikezelo ezinembe kakhudlwana. I-Capsule Networks ikhipha izakhiwo eziyinkimbinkimbi kancane kancane kudatha yokufaka ngokusebenzisa izingqimba eziningi zamaphilisi.
Indlela ye-Capsule Networks ibenza bakwazi ukufunda ukumelwa okulandelanayo kokufakiwe okunikeziwe. Bangakwazi ukufaka ikhodi ngokufanelekile ukuxhumana kwendawo phakathi kwezinto ezingaphakathi kwesithombe ngokuxhumana phakathi kwamaphilisi.
Ukuhlonza into, ukuhlukaniswa kwesithombe, nokucutshungulwa kolimi lwemvelo konke kuyizinhlelo ze-Capsule Networks.
Ama-Capsule Networks anamandla okuqashwa kuwo ukushayela okuzimele ubuchwepheshe. Basiza uhlelo ekuboneni nasekuhlukaniseni phakathi kwezinto ezifana nezimoto, abantu, nezimpawu zomgwaqo. Lezi zinhlelo zingagwema ukungqubuzana ngokwenza izibikezelo ezinembayo mayelana nokuziphatha kwezinto endaweni yazo.
7. Ama-Autoencoder ahlukahlukene (VAEs)
Ama-VAE awuhlobo lwethuluzi lokufunda elijulile elisetshenziselwa ukufunda okungagadiwe. Ngokubhala ngekhodi idatha endaweni enobukhulu obuphansi bese beyikhipha ngekhodi ibuyele kufomethi yoqobo, bangase bafunde ukubona amaphethini kudatha.
Bafana nomthakathi okwazi ukuguqula unogwaja abe isigqoko bese ebuyela ekubeni unogwaja! Ama-VAE anenzuzo ekukhiqizeni okubonakalayo okungokoqobo noma umculo. Futhi, zingasetshenziselwa ukukhiqiza idatha entsha eqhathaniseka nedatha yasekuqaleni.
Ama-VAE afana ne-codebreaker eyimfihlo. Bangathola umnyombo ukwakheka kwedatha ngokuyihlephula ibe izingcezu ezilula, kufana nokuthi iphazili ihlukaniswa kanjani. Bangase basebenzise lolo lwazi ukuze bakhe idatha entsha efana neyokuqala ngemva kokuhlela izingxenye.
Lokhu kungaba usizo ekucindezeleni amafayela amakhulu noma ukukhiqiza izithombe ezintsha noma umculo ngesitayela esithile. Ama-VAE angaphinda akhiqize okuqukethwe okusha, njengezindaba zezindaba noma amagama omculo.
8. I-Generative Adversarial Networks (GANs)
Ama-GAN (Generative Adversarial Networks) awuhlobo lwesistimu yokufunda ejulile ekhiqiza idatha entsha efana neyokuqala. Basebenza ngokuqeqesha amanethiwekhi amabili: i-generator kanye nenethiwekhi yokucwasa.
Ijeneretha ikhiqiza idatha entsha eqhathaniseka neyokuqala.
Futhi, umbandlululi uzama ukuhlukanisa phakathi kwedatha yoqobo nedaliwe. Lawa manethiwekhi amabili aqeqeshwe ngokuhambisana, ijeneretha izama ukukhohlisa umbandlululi futhi umcwasi ezama ukuhlonza kahle idatha yangempela.
Cabangela ama-GAN njengesiphambano phakathi komgunyathi nomseshi. Ijeneretha isebenza ngendlela efanayo nomgunyathi, ikhiqiza umsebenzi wobuciko omusha ofana nowokuqala.
Umbandlululi usebenza njengomseshi, ezama ukuhlukanisa phakathi kwemisebenzi yobuciko yangempela kanye nomgunyathi. Lawa manethiwekhi womabili aqeqeshwe ngokuhambisana, kanti ijeneretha iyathuthuka ekwenzeni amanga azwakalayo futhi umbandlululi uyathuthuka ekuwazini.
Ama-GAN anokusetshenziswa okuningana, kusukela ekukhiqizeni izithombe ezingokoqobo zabantu noma izilwane kuya ekudaleni umculo omusha noma ukubhala. Angase futhi asetshenziselwe ukukhulisa idatha, okuhlanganisa ukuhlanganisa idatha ekhiqiziwe nedatha yangempela ukuze kwakhiwe idathasethi enkulu yamamodeli okufunda omshini wokuqeqesha.
9. I-Deep Q-Networks (DQNs)
I-Deep Q-Networks (DQNs) iwuhlobo lwe-algorithm yokufunda yokuqinisa ukwenza izinqumo. Basebenza ngokufunda umsebenzi we-Q obikezela umvuzo olindelekile wokwenza isenzo esithile esimweni esithile.
I-Q-function ifundiswa ngokuzama nangephutha, nge-algorithm ezama izenzo ezahlukahlukene nokufunda emiphumeleni.
Kucabange njenge-a umdlalo we video umlingiswa ozama ngezenzo ezahlukahlukene futhi ethola ukuthi yiziphi eziholela empumelelweni! Ama-DQN aqeqesha umsebenzi we-Q esebenzisa inethiwekhi ye-neural ejulile, ewenza abe amathuluzi asebenzayo emisebenzi enzima yokwenza izinqumo.
Baze banqoba ompetha babantu emidlalweni efana neGo and chess, kanye nakumarobhothi kanye nezimoto ezizishayelayo. Ngakho-ke, kukho konke, ama-DQN asebenza ngokufunda kokuhlangenwe nakho ukuze athuthukise amakhono awo okwenza izinqumo ngokuhamba kwesikhathi.
10. I-Radial Basis Function Networks (RBFNs)
I-Radial Basis Function Networks (RBFNs) iwuhlobo lwenethiwekhi ye-neural esetshenziselwa ukulinganisa imisebenzi nokwenza imisebenzi yokuhlukanisa. Asebenza ngokuguqula idatha yokufaka ibe yindawo enobukhulu obuphezulu kusetshenziswa iqoqo lemisebenzi yesisekelo se-radial.
Okukhiphayo kwenethiwekhi kuyinhlanganisela yomugqa yemisebenzi eyisisekelo, futhi umsebenzi ngamunye wesisekelo se-radial umele indawo emaphakathi endaweni yokokufaka.
Ama-RBFN asebenza kahle kakhulu ezimweni ezinokusebenzelana kokufakwayo okuyinkimbinkimbi, futhi angase afundiswe kusetshenziswa amasu anhlobonhlobo, okuhlanganisa ukufunda okugadiwe nokungagadiwe. Zisetshenziselwe noma yini kusukela ekuqaguleni kwezezimali kuya ekubonweni kwesithombe nenkulumo kuya ekuxilongweni kwezokwelapha.
Cabangela ama-RBFN njengohlelo lwe-GPS olusebenzisa uchungechunge lwamaphoyinti okubamba ihange ukuze uthole indlela yawo endaweni eyinselele. Okukhiphayo kwenethiwekhi kuyinhlanganisela yamaphoyinti okubamba ihange, amele imisebenzi yesisekelo se-radial.
Singaphequlula olwazini oluyinkimbinkimbi futhi sikhiqize izibikezelo ezinembayo mayelana nokuthi isimo sizoba kanjani ngokusebenzisa ama-RBFN.
11. I-Multilayer Perceptrons (MLPs)
Uhlobo olujwayelekile lwenethiwekhi ye-neural ebizwa ngokuthi i-multilayer perceptron (MLP) isetshenziselwa imisebenzi yokufunda egadiwe njengokuhlela nokuhlehla. Asebenza ngokunqwabelanisa izingqimba ezimbalwa zamanodi axhunyiwe, noma ama-neurons, isendlalelo ngasinye sishintsha idatha engenayo ngokungaqondile.
Ku-MLP, i-neuron ngayinye ithola okokufaka kuma-neuron kungqimba engezansi bese ithumela isignali kuma-neurons kungqimba olungenhla. Okukhiphayo ngakunye kwe-neuron kunqunywa kusetshenziswa umsebenzi wokwenza kusebenze, okunikeza inethiwekhi ukungasebenzi komugqa.
Ziyakwazi ukufunda izethulo eziyinkimbinkimbi zedatha yokufaka njengoba zingaba nezendlalelo ezimbalwa ezifihliwe.
Ama-MLP asetshenziswe emisebenzini ehlukahlukene, njengokuhlaziya imizwa, ukutholwa kokukhwabanisa, nokuqashelwa kwezwi nesithombe. Ama-MLP angahle aqhathaniswe neqembu labaphenyi abasebenza ndawonye ukuze baphule icala elinzima.
Ndawonye, bangahlanganisa amaqiniso futhi baxazulule ubugebengu naphezu kweqiniso lokuthi ngamunye unendawo ethile ekhethekile.
12. I-Convolutional Neural Networks (CNNs)
Izithombe namavidiyo acutshungulwa kusetshenziswa amanethiwekhi e-convolutional neural (CNNs), uhlobo lwenethiwekhi ye-neural. Asebenza ngokusebenzisa isethi yezihlungi ezifundekayo, noma ama-kernels, ukuze kukhishwe izici ezibalulekile kudatha yokufaka.
Izihlungi zitshuza phezu kwesithombe sokufaka, zenza izinguquko ukuze zakhe imephu yesici ethwebula izici ezibalulekile zesithombe.
Njengoba ama-CNN ekwazi ukufunda ukumelwa okulandelanayo kwezici zezithombe, awusizo ikakhulukazi ezimeni ezihlanganisa imiqulu emikhulu yedatha ebonakalayo. Izinhlelo zokusebenza ezimbalwa zizisebenzisile, njengokutholwa kwento, ukuhlukaniswa kwezithombe, nokutholwa kobuso.
Cabangela ama-CNN njengomdwebi osebenzisa amabhulashi ambalwa ukuze enze ubuciko obuhle. Ibhulashi ngalinye liwuhlamvu, futhi umdwebi angase akhe isithombe esiyinkimbinkimbi, esingokoqobo ngokuxuba izinhlamvu eziningi. Singakhipha izici ezibalulekile ezithombeni futhi sizisebenzise ukubikezela ngokunembile okuqukethwe yisithombe ngokusebenzisa ama-CNN.
13. I-Deep Belief Networks (DBNs)
Ama-DBN awuhlobo lwenethiwekhi ye-neural esetshenziselwa imisebenzi yokufunda engagadiwe njengokunciphisa ubukhulu nokufunda izici. Asebenza ngokunqwabelanisa izendlalelo ezimbalwa zeMishini Ye-Boltzmann Ekhawulelwe (ama-RBM), okungamanethiwekhi emizwa anezendlalelo ezimbili akwazi ukufunda ukuhlanganisa kabusha idatha yokufaka.
Ama-DBN azuzisa kakhulu ezindabeni zedatha enobukhulu obuphakeme ngoba angafunda ukumelela okuhlangene nokuphumelelayo kokokufaka. Zisetshenziselwe noma yini kusukela ekwazini izwi kuya ekuhlukaniseni izithombe kuya ekutholakaleni kwezidakamizwa.
Isibonelo, abacwaningi basebenzise i-DBN ukuze balinganisele ukuhambisana okubophezelayo kwamakhandidethi emithi kumamukeli we-estrogen. I-DBN yaqeqeshwa eqoqweni lezimpawu zamakhemikhali kanye nezibopho ezibophezelayo, futhi yakwazi ukubikezela ngokunembile ukuhlobana okubophezelayo kwamakhandidethi amanoveli ezidakamizwa.
Lokhu kugqamisa ukusetshenziswa kwama-DBN ekuthuthukisweni kwezidakamizwa nezinye izinhlelo zokusebenza zedatha ezinobukhulu obuphezulu.
14. Ama-autoencoder
Ama-autoencoder amanethiwekhi e-neural asetshenziselwa imisebenzi yokufunda engagadiwe. Zihloselwe ukwakha kabusha idatha yokufaka, okusho ukuthi zizofunda ukubethela imininingwane ibe isethulo esihlangene bese iphinda ihlukaniselwe kokufakwayo kokuqala.
Ama-autoencoder asebenza kahle kakhulu ekucindezelweni kwedatha, ukususwa komsindo, nokutholwa okudidayo. Angaphinda asetshenziselwe ukufunda kwesici, lapho ukumelwa okuhlangene kwe-autoencoder kufakwa emsebenzini wokufunda ogadiwe.
Cabangela ama-autoencoder njengabafundi abathatha amanothi ekilasini. Umfundi ulalela inkulumo futhi abhale amaphuzu afaneleka kakhulu ngendlela emfushane nephumelelayo.
Kamuva, umfundi angase atadishe futhi akhumbule isifundo esebenzisa amaphuzu akhe. I-autoencoder, ngakolunye uhlangothi, ibhala ngekhodi idatha yokufaka ibe isethulo esihlangene esingase kamuva sisetshenziselwe izinjongo ezihlukile njengokutholwa okudidayo noma ukuminyanisa idatha.
15. Imishini ye-Boltzmann Ekhawulelwe (ama-RBM)
Ama-RBM (Imishini Ye-Boltzmann Ekhawulelwe) awuhlobo lwenethiwekhi ye-neural ekhiqizayo esetshenziselwa imisebenzi yokufunda engagadiwe. Akhiwe ungqimba olubonakalayo kanye nongqimba olufihliwe, olunama-neurons kungqimba ngalunye, axhunyiwe kodwa hhayi ngaphakathi kwesendlalelo esifanayo.
Ama-RBM aqeqeshwa kusetshenziswa indlela eyaziwa ngokuthi ukuhlukana okuhlukile, okuhlanganisa ukushintsha izisindo phakathi kwezendlalelo ezibonakalayo nezifihliwe ukuze kuthuthukiswe amathuba edatha yokuqeqeshwa. Ama-RBM angase adale idatha entsha ngemva kokuqeqeshwa ngamasampula ekusabalaliseni okufundiwe.
Ukubonwa kwesithombe nenkulumo, ukuhlunga ngokuhlanganyela, nokutholwa okudidayo konke kuyizinhlelo zokusebenza ezisebenzise ama-RBM. Ziphinde zasetshenziswa ezinhlelweni zokuncoma ukwenza izincomo ezihambisanayo ngokufunda amaphethini ekuziphatheni komsebenzisi.
Ama-RBM nawo asetshenziswe ekufundeni kwesici ukuze kwakheke ukumelela okuhlangene nokuphumelelayo kwedatha enobukhulu obuphezulu.
Ukusonga kanye Nentuthuko Ethembisayo Esekuphakameni
Izindlela zokufunda ezijulile, njengeConvolutional Neural Networks (CNNs) kanye neRecurrent Neural Networks (RNNs), ziphakathi kwezindlela ezithuthuke kakhulu zobuhlakani bokwenziwa. Ama-CNN aguqule ukubonwa kwesithombe nomsindo, kuyilapho ama-RNN athuthuke kakhulu ekucutshungulweni kolimi lwemvelo nokuhlaziywa kwedatha okulandelanayo.
Isinyathelo esilandelayo ekuthuthukisweni kwalezi zindlela kungenzeka sigxile ekuthuthukiseni ukusebenza kahle kwazo kanye nokulinganisa, ukuzivumela ukuthi zihlaziye amasethi edatha amakhulu futhi ayinkimbinkimbi, kanye nokuthuthukisa ukutolika kwawo kanye nekhono lokufunda kudatha enelebula elincane.
Ukufunda okujulile kunethuba lokuvumela impumelelo emikhakheni efana nokunakekelwa kwezempilo, ezezimali, nezinhlelo ezizimele njengoba ithuthuka.
shiya impendulo