Kangangeminyaka, ukufunda okunzulu bekusenza iintloko kwitekhnoloji. Kwaye, kulula ukuqonda ukuba kutheni.
Eli sebe lobukrelekrele bokwenziwa liguqula amacandelo asusela kukhathalelo lwempilo ukuya kwiibhanki ukuya kwezothutho, nto leyo evumela ukuqhubela phambili obekungacingelwa ngaphambili.
Ukufunda nzulu kwakhelwe kwiseti yee-algorithms ezintsonkothileyo ezifunda ukukhupha kunye nokuqikelela iipateni ezintsonkothileyo ukusuka kumthamo omkhulu wedatha.
Siza kujonga ezona ndlela zili-15 zokufunda nzulu kwesi sithuba, ukusuka kwiConvolutional Neural Networks ukuya kwiGenerative Adversarial Networks ukuya kuthungelwano lweMemori yexesha elifutshane.
Esi sithuba siya kunika ulwazi olubalulekileyo malunga nokuba unguye na oqalayo okanye ingcaphephe ekufundeni nzulu.
1. Uthungelwano lweTransformer
Uthungelwano lweTransformer lutshintshile umbono wekhompyutha kunye nezicelo zokusetyenzwa kolwimi lwendalo (NLP). Bahlalutya idatha engenayo kwaye basebenzise iinkqubo zokuqwalaselwa ukubamba ubudlelwane obude. Oku kuzenza zikhawuleze kuneemodeli eziqhelekileyo zokulandelelana ukuya kulandelelwano.
Uthungelwano lweTransformer luchazwe okokuqala kwipapasho elithi "Ingqalelo Yiyo yonke into oyifunayo" nguVaswani et al.
Ziquka i-encoder kunye ne-decoder (2017). Imodeli ye-transformer ibonise ukusebenza kwiintlobo ezahlukeneyo zezicelo ze-NLP, kubandakanywa Uhlalutyo lweemvakalelo, ukuhlelwa kombhalo, kunye noguqulelo lomatshini.
Iimodeli ezisekwe kwiTransformer nazo zinokusetyenziswa kumbono wekhompyuter kwizicelo. Banokwenza ukuqaphela into kunye ne-captioning yomfanekiso.
2. Uthungelwano lweMemori yexesha elifutshane (LSTMs)
IiNethiwekhi zeMemori zeXesha Elifutshane (LSTMs) luhlobo lwe inethiwekhi yomnatha ngakumbi yakhelwe ukuphatha igalelo elilandelelanayo. Zibizwa ngokuba “ziixesha elifutshane elide” kuba zikhumbula ulwazi lwakudala ngelixa zilibala ulwazi olungeyomfuneko.
Ii-LSTMs zisebenza “ngamasango” athile alawula ukuhamba kolwazi ngaphakathi kuthungelwano. Kuxhomekeka ekubeni ulwazi lujongwa njengebalulekileyo okanye akunjalo, la masango anokuthi ayingenise okanye ayithintele.
Obu buchule benza ukuba ii-LSTM zikhumbule okanye zilibale ulwazi olusuka kumanyathelo exesha elidlulileyo, olubaluleke kakhulu kwimisebenzi efana nokuqonda intetho, ukusetyenzwa kolwimi lwendalo, kunye noqikelelo lothotho lwexesha.
Ii-LSTM ziluncedo kakhulu kuyo nayiphi na imeko apho unedatha elandelelanayo ekufuneka ivavanywe okanye iqikelelwe. Zihlala zisetyenziswa kwisoftware yokuqondwa kwelizwi ukuguqula amagama athethiweyo abe ngumbhalo, okanye ngaphakathi imakethi yesitokhwe uhlalutyo ukubikezela amaxabiso exesha elizayo ngokusekelwe kwidatha yangaphambili.
3. Iimephu zokuZilungiselela (ii-SOMs)
Ii-SOM luhlobo oluthile lokwenziwa inethiwekhi ye-neural enokufunda kwaye imele idatha enzima kwindawo ephantsi-dimensional. Indlela isebenza ngokuguqula idatha yegalelo eliphezulu-dimensional ibe yigridi ene-dimensional, kunye neyunithi nganye okanye i-neuron emele indawo eyahlukileyo yendawo yokufaka.
I-neurons idibaniswe kunye kwaye yenze i-topological structure, evumela ukuba bafunde kwaye balungelelanise idatha yegalelo. Ke, iSOM isekwe kwimfundo engajongwanga.
I-algorithm ayifuni idatha ephawulweyo ukufunda kwi. Endaweni yoko, isebenzisa iimpawu zezibalo zedatha yokufaka ukufumanisa iipateni kunye nokulungelelaniswa phakathi kwezinto eziguquguqukayo.
Ngexesha loqeqesho, i-neurons ikhuphisana ukuze ibe yeyona nto ibonakalisa idatha yegalelo. Kwaye, bazicwangcisa ngokwabo kwisakhiwo esinentsingiselo. Ii-SOM zinoluhlu olubanzi lwezicelo, kubandakanywa ukuqaphela umfanekiso kunye nentetho, ukuchithwa kwedatha, kunye nokuqatshelwa kwepateni.
Ziluncedo kwi ukubona idatha entsonkothileyo, ukuhlanganisa amanqaku edatha anxulumeneyo, kunye nokufumanisa izinto ezingaqhelekanga okanye ngaphandle.
4. UkuFundisa ngokuNxibelela ngokuNzulu
nzulu Ukomeleza ukuFunda luhlobo lokufunda koomatshini apho iarhente iqeqeshelwa ukwenza izigqibo ezisekelwe kwinkqubo yomvuzo. Isebenza ngokuvumela i-arhente isebenze nendawo eyingqongileyo kwaye ifunde ngolingo nangempazamo.
I-arhente ivuzwa ngayo yonke into eyenzayo, kwaye injongo yayo kukufunda indlela yokuphucula izibonelelo zayo ngokuhamba kwexesha. Oku kusenokusetyenziswa ukufundisa iiarhente ukudlala imidlalo, ukuqhuba iimoto, kunye nokulawula iirobhothi.
I-Q-Learning yindlela eyaziwayo yokuFundisa ngokuNzululwazi ngokuNzululwazi. Isebenza ngokuvavanya ixabiso lokwenza isenzo esithile kwimeko ethile kunye nokuhlaziya olo qikelelo njengoko i-arhente isebenzisana nokusingqongileyo.
I-arhente ke isebenzise olu qikelelo ukumisela ukuba leliphi inyathelo elinokukhokelela kumvuzo omkhulu. I-Q-Learning isetyenziselwe ukufundisa ama-agent ukuba adlale imidlalo ye-Atari, kunye nokuphucula ukusetyenziswa kwamandla kumaziko edatha.
I-Deep Q-Networks yenye indlela edumileyo yokuFundisa ngokuNxibelela ngokuNzululwazi (DQN). Ii-DQN ziyafana ne-Q-Learning kuba ziqikelela amaxabiso entshukumo zisebenzisa inethiwekhi ye-neural enzulu kunetafile.
Oku kubenza bajongane noseto olukhulu, oluntsokothileyo kunye nezenzo ezininzi ezizezinye. Ii-DQN zisetyenziselwe ukuqeqesha ii-agent ukuba zidlale imidlalo efana ne-Go kunye ne-Dota 2, kunye nokudala iirobhothi ezinokufunda ukuhamba.
5. IiNethiwekhi zeNeural eziRecurrent (RNNs)
IiRNN luhlobo lwenethiwekhi ye-neural ekwazi ukuqhubekekisa idatha elandelelanayo ngelixa igcina imeko yangaphakathi. Yicinge into efanayo nomntu ofunda incwadi, apho igama ngalinye letyiswa ngokunxulumene nalawo avela ngaphambi kwayo.
IiRNN ke ngoko zifanelekile kwimisebenzi efana nokuqondwa kwentetho, ukuguqulela ulwimi, kunye nokuxela kwangaphambili igama elilandelayo kwibinzana.
IiRNNs zisebenza ngokusebenzisa iilophu zengxelo ukudibanisa imveliso yenyathelo ngalinye lokubuyela emva kwigalelo lenyathelo elilandelayo. Oku kwenza inethiwekhi isebenzise ulwazi lwamanyathelo angaphambili ukwazisa uqikelelo lwamanyathelo exesha elizayo. Ngelishwa, oku kukwathetha ukuba ii-RNNs zisesichengeni somba wokunyamalala wegradient, apho igradient ezisetyenziselwa uqeqesho ziba zincinci kakhulu kwaye inethiwekhi iyasokola ukufunda ubudlelwane bexesha elide.
Ngaphandle kwesi sithintelo esibonakalayo, iiRNN zifumene ukusetyenziswa kuluhlu olubanzi lwezicelo. Ezi zicelo ziquka ukusetyenzwa kolwimi lwendalo, ukuqondwa kwentetho, kunye nokuveliswa komculo.
Isiguquleli sikaGoogle, umzekelo, usebenzisa inkqubo esekwe kwiRNN ukuguqulela kwiilwimi zonke, ngelixa uSiri, umncedisi wenyani, usebenzisa inkqubo esekwe kwiRNN ukufumanisa ilizwi. Ii-RNN zikwasetyenziselwa ukuqikelela amaxabiso esitokhwe kunye nokudala isicatshulwa esinenyani kunye nemizobo.
6. Iinethiwekhi zeCapsule
I-Capsule Networks luhlobo olutsha loyilo lwenethiwekhi ye-neural enokuchonga iipateni kunye nokulungelelaniswa kwedatha ngokusebenzayo ngakumbi. Baququzelela i-neurons ibe "i-capsules" ezifakela iinkalo ezithile zegalelo.
Ngale ndlela banokwenza uqikelelo oluchane ngakumbi. I-Capsule Networks ikhupha iipropathi ezinobunzima obuqhubekayo ukusuka kwidatha yegalelo ngokusebenzisa uluhlu oluninzi lwee-capsules.
Ubuchule beCapsule Networks bubenza ukuba bafunde ukumelwa ngokwemigangatho yegalelo elinikiweyo. Banokudibanisa ngokufanelekileyo unxibelelwano lwesithuba phakathi kwezinto ezingaphakathi komfanekiso ngokunxibelelana phakathi kwee-capsules.
Ukuchongwa kwento, ulwahlulo lwemifanekiso, kunye nokulungiswa kolwimi lwendalo zonke zizicelo zeCapsule Networks.
IiNethiwekhi zeCapsule zinethuba lokuqashwa ukuqhuba ngokuzimela ubugcisa. Bancedisa inkqubo ekuboneni nasekuhlukaniseni phakathi kwezinto ezinjengeemoto, abantu, kunye neempawu zendlela. Ezi nkqubo zinokunqanda ukungqubana ngokwenza uqikelelo oluchanekileyo malunga nokuziphatha kwezinto kwindawo yazo.
7. Ii-Autoencoders ezahlukeneyo (VAEs)
Ii-VAE luhlobo lwesixhobo sokufunda nzulu esisetyenziselwa ukufunda kungajongwanga. Ngokufaka ikhowudi kwisithuba esisezantsi-ntathu kwaye emva koko bayiguqulele kwifomathi yokuqala, banokufunda ukubona iipateni kwidatha.
Bafana negqwirha elikwaziyo ukuguqula umvundla ube ngumnqwazi lize liphinde libe yimvundla! Ii-VAEs ziluncedo ekuveliseni izinto ezibonwayo zokwenyani okanye umculo. Kwaye, zingasetyenziselwa ukuvelisa idatha entsha enokuthelekiswa nedatha yokuqala.
Ii-VAE zifana ne-codebreaker eyimfihlo. Bayakwazi ukufumanisa undoqo ubume bedatha ngokuyicalula ibe ngamasuntswana alula, kufana nendlela iphazili ecalulwa ngayo. Basenokusebenzisa olo lwazi ukwakha idatha entsha ekhangeleka ngathi yeyokuqala emva kokuba balungise iindawo.
Oku kunokuba luncedo ekucinezeleni iifayile ezinkulu okanye ukuvelisa iigraphics ezintsha okanye umculo kwisitayile esithile. Ii-VAE zinokuvelisa umxholo omtsha, onje ngamabali eendaba okanye amazwi omculo.
8. IiNethiwekhi zoNxibelelwano lweNkathazo (GANs)
I-GANs (i-Generative Adversarial Networks) luhlobo lwenkqubo yokufunda enzulu eyenza idatha entsha efana neyokuqala. Basebenza ngokuqeqesha amanethiwekhi amabini: i-generator kunye nenethiwekhi yocalucalulo.
Ijeneretha ivelisa idatha entsha enokuthelekiswa neyokuqala.
Kwaye, umcaluli uzama ukwahlula phakathi kwedatha yokuqala kunye neyokudala. Amanethiwekhi amabini aqeqeshwe kwi-tandem, kunye nejeneretha ezama ukukhohlisa umcalucalulo kunye nocalucalulo oluzama ukuchonga ngokufanelekileyo idatha yokuqala.
Zithathele ingqalelo ii-GAN njengonxibelelwano phakathi komqambi nomcuphi. Ijenereyitha isebenza ngokufanayo nekhambi, ivelisa umsebenzi omtsha wobugcisa ofana nowokuqala.
Umcalu-calulo usebenza njengomcuphi, ezama ukwahlula phakathi komsebenzi wobugcisa wokwenene kunye nomgunyathi. Uthungelwano olubini luqeqeshwe ngokuhambelanayo, kunye nejenereyitha ephuculayo ekwenzeni i-fakes ebonakalayo kunye nomcaluli uphucula ekuqapheleni kwabo.
Ii-GAN zinezinto ezininzi ezisetyenziswayo, ukusuka ekuveliseni imifanekiso yokwenene yabantu okanye yezilwanyana ukuya ekudaleni umculo omtsha okanye ukubhala. Zisenokusetyenziselwa ukwandiswa kwedatha, okubandakanya ukudibanisa idatha eveliswayo kunye nedatha yokwenyani ukwakha i-dataset enkulu kwiimodeli zokufunda zoomatshini.
9. IiNethiwekhi ezinzulu ze-Q (DQNs)
I-Deep Q-Networks (i-DQNs) luhlobo lwe-algorithm yokufunda yokuqinisa ukwenza izigqibo. Basebenza ngokufunda umsebenzi we-Q oqikelela umvuzo olindelekileyo wokwenza isenzo esithile kwimeko ethile.
I-Q-function ifundiswa ngovavanyo kunye nephutha, kunye ne-algorithm ezama izenzo ezahlukeneyo kunye nokufunda kwiziphumo.
Yicinge njenge umdlalo yevidiyo umlinganiswa ozama ngezenzo ezahlukeneyo aze afumanise ukuba zeziphi ezikhokelela empumelelweni! Ii-DQNs ziqeqesha i-Q-function isebenzisa inethiwekhi ye-neural enzulu, izenza izixhobo ezisebenzayo kwimisebenzi enzima yokwenza izigqibo.
Bade boyisa iintshatsheli zabantu kwimidlalo efana neGo kunye nechess, kunye nerobhothi kunye neemoto eziziqhubayo. Ke, lilonke, ii-DQNs zisebenza ngokufunda kumava ukuphucula izakhono zabo zokuthatha izigqibo ekuhambeni kwexesha.
10. Uthungelwano lwe-Radial Basis Function Networks (RBFNs)
I-Radial Basis Function Networks (RBFNs) luhlobo lwenethiwekhi ye-neural esetyenziselwa ukuqikelela imisebenzi kunye nokwenza imisebenzi yokuhlelwa. Basebenza ngokuguqula idatha yegalelo kwindawo ephezulu-dimensional usebenzisa ingqokelela yemisebenzi yesiseko se-radial.
Imveliso yomsebenzi womnatha yindibaniselwano yomgca yesiseko semisebenzi, kwaye umsebenzi ngamnye wesiseko seradial umele indawo esembindini kwindawo yongeniso.
Ii-RBFN zisebenza ngokukodwa kwiimeko ezinonxibelelwano oluntsonkothileyo lwegalelo-mveliso, kwaye zinokufundiswa kusetyenziswa uluhlu olubanzi lweendlela, kubandakanywa ukufunda okubekwe esweni kunye nokungagadwanga. Zisetyenziselwe nantoni na ukusuka kuqikelelo lwemali ukuya kumfanekiso kunye nokuqatshelwa kwentetho ukuya kuxilongo lwezonyango.
Qwalasela ii-RBFN njengenkqubo yeGPS esebenzisa uthotho lwamanqaku e-ankile ukufumana indlela yawo kumhlaba ocela umngeni. Imveliso yenethiwekhi yindibaniselwano yamanqaku e-anchor, emele imisebenzi yesiseko se-radial.
Sinokukhangela kulwazi oluntsonkothileyo kwaye sivelise uqikelelo oluchanekileyo malunga nokuba imeko iyakwenzeka njani na ngokuqesha ii-RBFN.
11. IiPerceptron zaMacala amaninzi (MLPs)
Uhlobo oluqhelekileyo lothungelwano lwe-neural olubizwa ngokuba yi-multilayer perceptron (MLP) lusetyenziselwa imisebenzi yokufunda ebekwe esweni njengokuhlela nokuhlehla. Zisebenza ngokupakisha iileya ezininzi zeenodi eziqhagamshelweyo, okanye i-neuron, kunye nomaleko ngamnye ngokungaguquguqukiyo kwedatha engenayo.
Kwi-MLP, i-neuron nganye ifumana igalelo kwi-neuron ekumaleko angezantsi kwaye ithumela isignali kwi-neurons ekumaleko ongentla. Imveliso nganye ye-neuron igqitywe ngokusebenzisa umsebenzi wokuvula, onika unxibelelwano lwenethiwekhi.
Bayakwazi ukufunda ukubonakaliswa okuphucukileyo kwedatha yegalelo kuba banokuba neeleya ezininzi ezifihliweyo.
Ii-MLPs zisetyenziswe kwimisebenzi eyahlukeneyo, njengokuhlalutya kweemvakalelo, ukufumanisa ubuqhetseba, kunye nokuqatshelwa kwezwi kunye nomfanekiso. Ii-MLPs zinokuthelekiswa neqela labaphandi abasebenzisanayo ukuqhekeza ityala elinzima.
Ngokudibeneyo, banokudibanisa iinyani kwaye basombulule ulwaphulo-mthetho ngaphandle kwento yokuba ngamnye unendawo ethile ekhethekileyo.
12. Convolutional Neural Networks (CNNs)
Imifanekiso kunye neevidiyo zicutshungulwa kusetyenziswa i-convolutional neural networks (CNNs), uhlobo lothungelwano lwe-neural. Basebenza ngokuqesha iseti yezihluzi ezifundekayo, okanye iinkozo, ukukhupha iimpawu ezibalulekileyo kwidatha yegalelo.
Izihluzi zityibilika phezu komfanekiso wegalelo, zenza iiconvolutions ukwakha imephu yefitsha ebamba imiba ebalulekileyo yomfanekiso.
Njengoko ii-CNNs zikwazi ukufunda ukumelwa ngokwemigangatho yeempawu zemifanekiso, ziluncedo ngakumbi kwiimeko ezibandakanya imithamo emikhulu yedatha ebonakalayo. Iinkqubo ezininzi ziye zazisebenzisa, ezinje ngobhaqo lwento, ukuhlelwa kwemifanekiso, kunye nokubhaqwa kobuso.
Qwalasela ii-CNNs njengomzobi osebenzisa iibrashi ezininzi ukwenza umsebenzi wobugcisa. Ibhrashi nganye yinkozo, kwaye umzobi unokwakha umfanekiso ontsokothileyo, wokwenyani ngokuxuba iinkozo ezininzi. Sinokukhupha iimpawu ezibalulekileyo kwiifoto kwaye sizisebenzise ukuqikelela ngokuchanekileyo imixholo yomfanekiso ngokusebenzisa ii-CNNs.
13. Uthungelwano lweNkolo enzulu (DBNs)
Ii-DBN luhlobo lothungelwano lwe-neural olusetyenziselwa imisebenzi yokufunda engajongwanga njengokunciphisa ubungakanani kunye nokufunda kweempawu. Zisebenza ngokupakisha iileya ezininzi zoomatshini be-Boltzmann abaMiselweyo (ii-RBMs), eziziinethiwekhi ze-neural ezinomaleko-mbini ezikwaziyo ukufunda ukuphinda kuqulunqwe idatha yegalelo.
I-DBN inenzuzo kakhulu kwimiba yedatha ephezulu ngenxa yokuba inokufunda ukubonakaliswa okuncinci kunye nokufanelekileyo kwegalelo. Zisetyenziselwe nantoni na ukusuka ekuqondeni ilizwi ukuya kuluhlu lwemifanekiso ukuya ekufumaneni iziyobisi.
Ngokomzekelo, abaphandi baqeshe i-DBN ukuqikelela ubudlelwane obubophezelayo babaviwa bamayeza kwi-estrogen receptor. I-DBN yaqeqeshwa kwingqokelela yeempawu zeekhemikhali kunye nee-affinities ezibophelelayo, kwaye yakwazi ukuqikelela ngokuchanekileyo ubudlelwane obubophelelayo babaviwa bamachiza anoveli.
Oku kugxininisa ukusetyenziswa kwe-DBN kuphuhliso lweziyobisi kunye nezinye izicelo zedatha ephezulu.
14. Iikhowudi ezizenzekelayo
I-autoencoders ziinethiwekhi ze-neural ezisetyenziselwa imisebenzi yokufunda engajongwanga. Zijonge ukwakhiwa ngokutsha kwedatha yegalelo, nto leyo ethetha ukuba baya kufunda ukubethelela ulwazi kwi-compact representation baze emva koko bayihlaziye bayibuyisele kwigalelo lokuqala.
Iikhowudi ezizenzekelayo zisebenza kakhulu kuxinzelelo lwedatha, ukususwa kwengxolo, kunye nokubhaqwa okungaqhelekanga. Zisenokusetyenziswa nasekufundeni inkalo, apho i-autoencoder's compact representation ifakwa kumsebenzi wokufunda obekwe esweni.
Qwalasela ii-autoencoders ukuba ibe ngabafundi abathatha amanqaku eklasini. Umfundi umamela intetho aze abhale awona manqaku afanelekileyo ngendlela emfutshane nesebenzayo.
Kamva, umfundi usenokusifundisisa aze asikhumbule eso sifundo esebenzisa amanqaku akhe. I-autoencoder, kwelinye icala, ifaka iikhowudi kwidatha yegalelo ibe ngumboniso ohlangeneyo onokuthi emva koko usetyenziselwe iinjongo ezahlukeneyo ezifana nokubhaqwa okungaqhelekanga okanye ucinezelo lwedatha.
15. Oomatshini beBoltzmann abathintelweyo(RBMs)
Ii-RBMs (Oomatshini beBoltzmann abaThintelweyo) luhlobo lothungelwano lwe-neural oluvelisayo olusetyenziselwa imisebenzi yokufunda engajongwanga. Zenziwe ngoluhlu olubonakalayo kunye noluhlu olufihliweyo, kunye ne-neurons kwinqanaba ngalinye, elidibeneyo kodwa lingekho ngaphakathi kwinqanaba elifanayo.
Ii-RBMs ziqeqeshwa ngokusebenzisa indlela eyaziwa ngokuba yintlukwano echaseneyo, ebandakanya ukutshintsha iintsimbi phakathi kweeleya ezibonakalayo nezifihlakeleyo ukuze kwandiswe ukuba nokwenzeka kwedatha yoqeqesho. Ii-RBMs zinokudala idatha entsha emva kokuqeqeshwa ngesampulu kulwabiwo olufundiweyo.
Ukuqondwa komfanekiso kunye nentetho, ukuhluza ngentsebenziswano, kunye nokubhaqwa okungaqhelekanga zizo zonke iinkqubo ezisebenzise ii-RBMs. Zikwasetyenziswe kwiisistim zokucebisa ukwenza iingcebiso ezilungiselelwe ngokufunda iipatheni zokuziphatha kwabasebenzisi.
Ii-RBMs nazo zisetyenzisiwe kwinkalo yokufunda ukwenza umboniso obambeneyo nofanelekileyo wedatha enomgangatho ophezulu.
Ukusonga kunye noPhuhliso oluthembisayo kwi-Horizon
Iindlela zokufunda ezinzulu, ezinje ngeConvolutional Neural Networks (CNNs) kunye neRecurrent Neural Networks (RNNs), ziphakathi kweyona ndlela iphambili yobukrelekrele bokwenziwa. Ii-CNNs ziguqule umfanekiso kunye nokuqatshelwa kwe-audio, ngelixa i-RNN iqhubele phambili kakhulu ekuqhubeni ulwimi lwendalo kunye nohlalutyo lwedatha olulandelelanayo.
Inyathelo elilandelayo ekuveleni kwezi ndlela kunokwenzeka ukuba lijolise ekuphuculeni ukusebenza kakuhle kunye nokulinganisa, ukubavumela ukuba bahlalutye iiseti zedatha ezinkulu nezintsonkothileyo, kunye nokuphucula ukutolika kunye nokukwazi ukufunda kwiidatha ezinombhalo omncinci.
Ukufunda nzulu kunethuba lokuvumela impumelelo kwiinkalo ezifana nezempilo, imali, kunye neenkqubo ezizimeleyo njengoko iqhubela phambili.
Shiya iMpendulo