Okuqukethwe[Fihla][Bonisa]
Ikusasa selifikile. Futhi, kule mishini yesikhathi esizayo iqonda umhlaba oyizungezile ngendlela efanayo abantu abawenza ngayo. Amakhompyutha angakwazi ukushayela izimoto, ahlole izifo, futhi abikezele ikusasa ngokunembile.
Lokhu kungase kubonakale njengenganekwane yesayensi, kodwa amamodeli okufunda ajulile akwenza kube ngokoqobo.
Lawa ma-algorithms ayinkimbinkimbi embula izimfihlo ze ukuhlakanipha okungekhona okwangempela, okuvumela amakhompyutha ukuthi azifundele futhi athuthuke. Kulokhu okuthunyelwe, sizongena emkhakheni wamamodeli okufunda ajulile.
Futhi, sizophenya amandla amakhulu abanawo okuguqula izimpilo zethu. Lungiselela ukufunda ngobuchwepheshe besimanje obushintsha ikusasa lesintu.
Imaphi Ngempela Amamodeli Okufunda Okujulile?
Ingabe uke wadlala umdlalo lapho kufanele uhlonze khona umehluko phakathi kwezithombe ezimbili?
Kuyajabulisa nokho, kungase futhi kube nzima, akunjalo? Zibone ngeso lengqondo ukwazi ukufundisa ikhompuyutha ukudlala lowo mdlalo futhi uwine ngaso sonke isikhathi. Amamodeli okufunda okujulile enza lokho kanye!
Amamodeli okufunda okujulile afana nemishini ehlakaniphe kakhulu engahlola inani elikhulu lezithombe futhi inqume ukuthi ifana ngani. Lokhu bakufeza ngokuqaqa izithombe futhi bafunde ngayinye ngayinye.
Bese besebenzisa abakufundile ukuze babone amaphethini futhi benze izibikezelo mayelana nezithombe ezintsha abangakaze bazibone ngaphambilini.
Amamodeli okufunda okujulile amanethiwekhi e-neural okwenziwa angafunda futhi akhiphe amaphethini ayinkimbinkimbi nezici kumadathasethi amakhulu. Lawa mamodeli akhiwe izendlalelo ezimbalwa zamanodi axhunyiwe, noma ama-neurons, ahlaziya futhi aguqule idatha engenayo ukuze enze okukhiphayo.
Amamodeli okufunda okujulile ayifanele kakhulu imisebenzi edinga ukunemba okukhulu nokunemba, njengokuhlonza isithombe, ukuqaphela inkulumo, ukucutshungulwa kolimi lwemvelo, namarobhothi.
Zisetshenziswe kuyo yonke into kusukela ezimotweni ezizishayelayo kuye ekuxilongweni kwezempilo, izinhlelo zokuncoma, kanye i-analytics yokubikezela.
Nansi inguqulo eyenziwe lula yokuboniswa ukuze ubonise ukuhamba kwedatha kumodeli yokufunda ejulile.
Idatha yokufaka igelezela kusendlalelo okokufaka semodeli, ebese idlulisela idatha kunombolo yezendlalelo ezifihliwe ngaphambi kokunikeza isibikezelo sokuphumayo.
Isendlalelo ngasinye esifihliwe senza uchungechunge lwemisebenzi yezibalo kudatha yokufaka ngaphambi kokuyidlulisela kusendlalelo esilandelayo, esinikeza isibikezelo sokugcina.
Manje, ake sibone ukuthi yimaphi amamodeli okufunda ajulile nokuthi singawasebenzisa kanjani ekuphileni kwethu.
1. I-Convolutional Neural Networks (CNNs)
Ama-CNN ayimodeli yokufunda ejulile eguqule indawo yombono wekhompyutha. Ama-CNN asetshenziselwa ukuhlukanisa izithombe, ukubona izinto, kanye nesegimenti yezithombe. Isakhiwo nomsebenzi we-visual cortex yomuntu wazisa ukwakheka kwama-CNN.
Zisebenza Kanjani?
I-CNN yenziwe ngenani lezendlalelo eziguquguqukayo, izendlalelo zokuhlanganisa, nezendlalelo ezixhunywe ngokugcwele. Okokufaka kuyisithombe, futhi okukhiphayo kuwukuqagela ilebula yeklasi yesithombe.
Izendlalelo ze-CNN ze-convolution zakha imephu yesici ngokwenza umkhiqizo wamachashazi phakathi kwesithombe sokufaka kanye nesethi yezihlungi. Izendlalelo zokuhlanganisa zehlisa usayizi wemephu yesici ngokuyenza isampula.
Okokugcina, isici semephu sisetshenziswa izendlalelo ezixhumeke ngokugcwele ukubikezela ilebula yesigaba sesithombe.
Kungani Abalulekile Ama-CNN?
Ama-CNN abalulekile ngoba angafunda ukuthola amaphethini nezici ezithombeni abantu abakuthola kunzima ukuzibona. Ama-CNN angafundiswa ukubona izici ezinjengamachopho, amakhona, nokuthungwa kusetshenziswa amasethi edatha amakhulu. Ngemva kokufunda lezi zakhiwo, i-CNN ingazisebenzisa ukuhlonza izinto ezithombeni ezintsha. Ama-CNN abonise ukusebenza okuphambili ezinhlelweni ezihlukahlukene zokuhlonza izithombe.
Sisebenzisa kuphi ama-CNN
Ukunakekelwa kwezempilo, imboni yezimoto, kanye nokudayisa kuyimikhakha embalwa nje eqasha ama-CNN. Embonini yezokunakekelwa kwempilo, zingaba nenzuzo ekuxilongeni ukugula, ukuthuthukiswa kwemithi, nokuhlaziywa kwesithombe sezokwelapha.
Emkhakheni wezimoto, basiza ngokutholwa komzila, ukutholwa kwento, kanye nokushayela ngokuzenzakalelayo. Zibuye zisetshenziswe kakhulu ekuthengiseni ukusesha okubonakalayo, isincomo somkhiqizo osuselwe ezithombeni, nokulawulwa kwempahla.
Ngokwesibonelo; I-Google isebenzisa ama-CNN kuzinhlelo zokusebenza ezihlukene, okuhlanganisa I-Google Lens, ithuluzi lokuhlonza isithombe elithandwa kakhulu. Uhlelo lusebenzisa ama-CNN ukuhlola izithombe futhi lunikeze abasebenzisi ulwazi.
I-Google Lens, ngokwesibonelo, ingabona izinto esithombeni futhi inikeze imininingwane ngazo, njengohlobo lwembali.
Ingase futhi ihumushe umbhalo okhishwe esithombeni uwuyise ezilimini eziningi. I-Google Lens iyakwazi ukunikeza abathengi ulwazi oluwusizo ngenxa yosizo lwama-CNN ekuhlonzeni ngokunembile izinto nokukhipha izici ezithombeni.
2. Amanethiwekhi Enkumbulo Yesikhathi Esifushane (LSTM).
Amanethiwekhi Enkumbulo Yesikhathi Esifushane (i-LSTM) adalelwe ukubhekana nokushiyeka kwamanethiwekhi avamile e-neural ajwayelekile (ama-RNN). Amanethiwekhi e-LSTM alungele imisebenzi efuna ukucutshungulwa kokulandelana kwedatha ngaso sonke isikhathi.
Asebenza ngokusebenzisa iseli yenkumbulo ethile kanye nezindlela ezintathu zokungena.
Balawula ukungena nokuphuma kolwazi kuseli. Isango lokufaka, isango lokukhohlwa kanye nesango lokuphumayo amasango amathathu.
Isango lokufaka lilawula ukugeleza kwedatha kuseli yenkumbulo, isango lokukhohlwa lilawula ukususwa kwedatha kuseli, futhi isango lokuphumayo lilawula ukugeleza kwedatha ephuma kuseli.
Kuyini Ukubaluleka kwazo?
Amanethiwekhi e-LSTM awusizo ngoba angamela ngempumelelo futhi abikezele ukulandelana kwedatha nobudlelwano besikhathi eside. Bangakwazi ukurekhoda futhi bagcine ulwazi mayelana nokokufaka kwangaphambilini, okubavumela ukuthi benze izibikezelo ezinembe kakhulu mayelana nokokufaka kwesikhathi esizayo.
Ukuqashelwa kwenkulumo, ukunakwa kokubhala ngesandla, ukucutshungulwa kolimi lwemvelo, namagama-ncazo wesithombe ezinye zezinhlelo zokusebenza ezisebenzise amanethiwekhi e-LSTM.
Siwasebenzisaphi Amanethiwekhi e-LSTM?
Izinhlelo zokusebenza eziningi zesofthiwe nobuchwepheshe zisebenzisa amanethiwekhi e-LSTM, okuhlanganisa nezinhlelo zokuqaphela inkulumo, amathuluzi okucubungula izilimi zemvelo njenge ukuhlaziywa kwemizwa, amasistimu wokuhumusha ngomshini, namasistimu okukhiqiza umbhalo nezithombe.
Zibuye zasetshenziswa ekwakhiweni kwezimoto ezizishayelayo namarobhothi, kanye nasemkhakheni wezezimali ukuze kutholwe ukukhwabanisa nokulindela. stock market ukunyakaza.
3. I-Generative Adversarial Networks (GANs)
Ama-GAN a ukufunda okujulile inqubo esetshenziselwa ukukhiqiza amasampula edatha amasha afana nedathasethi enikeziwe. Ama-GAN akhiwe ngamabili amanethiwekhi we-neural: eyodwa efunda ukukhiqiza amasampula amasha futhi efunda ukuhlukanisa phakathi kwamasampuli angempela nakhiqizwayo.
Ngendlela efanayo, lawa manethiwekhi amabili aqeqeshwa ndawonye kuze kube yilapho ijeneretha ingakwazi ukukhiqiza amasampula angenakuhlukaniswa kwamanye angempela.
Kungani Sisebenzisa ama-GAN
Ama-GAN abalulekile ngenxa yekhono lawo lokukhiqiza izinga eliphezulu idatha yokwenziwa engase isetshenziselwe izinhlelo zokusebenza ezihlukahlukene, okuhlanganisa ukukhiqizwa kwezithombe namavidiyo, ukukhiqizwa kombhalo, ngisho nokukhiqizwa komculo.
Ama-GAN nawo asetshenziselwe ukukhulisa idatha, okuwukukhiqizwa kwe idatha yokwenziwa ukwengeza idatha yomhlaba wangempela kanye nokwenza ngcono ukusebenza kwamamodeli okufunda ngomshini.
Ngaphezu kwalokho, ngokwakha idatha yokwenziwa engasetshenziswa ukuqeqesha amamodeli nokulingisa izivivinyo, ama-GAN anamandla okuguqula imikhakha efana nemithi nokuthuthukiswa kwezidakamizwa.
Izicelo zama-GAN
Ama-GAN angakwazi ukwengeza amasethi edatha, adale izithombe ezintsha noma amamuvi, futhi akhiqize idatha yokwenziwa yokulingisa kwesayensi. Ngaphezu kwalokho, ama-GAN anamandla okuqashwa ezinhlotsheni ezihlukahlukene kusukela kwezokuzijabulisa kuye kwezokwelapha.
iminyaka namavidiyo. I-StyleGAN2 ye-NVIDIA, isibonelo, isetshenziswe ukwenza izithombe zekhwalithi ephezulu zosaziwayo nomsebenzi wobuciko.
4. I-Deep Belief Networks (DBNs)
Ama-Deep Belief Networks (DBNs) akhona ukuhlakanipha okungekhona okwangempela amasistimu angafunda ukubona amaphethini kudatha. Bafeza lokhu ngokuhlukanisa idatha ibe yizingxenyana ezincane nezincane, bathole ukuyiqonda kahle kakhulu kuleveli ngayinye.
Ama-DBN angafunda kudatha ngaphandle kokwaziswa ukuthi iyini (lokhu kubizwa ngokuthi “ukufunda okungagadiwe”). Lokhu kuwenza abe yigugu kakhulu ekutholeni amaphethini kudatha umuntu angayithola kunzima noma kungenzeki ukuyibona.
Yini Eyenza I-DBN Ibaluleke?
Ama-DBN abalulekile ngenxa yekhono lawo lokufunda ukumelwa kwedatha yesigaba. Lezi izethulo zingasetshenziswa ezinhlelweni ezihlukahlukene ezifana nokuhlukanisa, ukutholwa okudidayo, nokunciphisa ubukhulu.
Amandla ama-DBN okwenza ukuqeqeshwa kwangaphambili okungagadiwe, okungakhuphula ukusebenza kwamamodeli okufunda okujulile anedatha encane enelebula, kuyinzuzo enkulu.
Yiziphi Izicelo zama-DBN?
Enye yezinhlelo zokusebenza ezibaluleke kakhulu ukutholwa kwento, lapho kusetshenziswa khona ama-DBN ukubona izinhlobo ezithile zezinto ezifana nezindiza, izinyoni, nabantu. Ziphinde zisetshenziselwe ukukhiqiza izithombe nokuhlukanisa, ukutholwa kokunyakaza kumafilimu, nokuqonda kolimi lwemvelo ukuze kucutshungulwe izwi.
Ngaphezu kwalokho, ama-DBN avame ukusetshenziswa kumasethi wedatha ukuhlola ukuma kwabantu. Ama-DBN ayithuluzi elihle lezimboni ezahlukahlukene, kufaka phakathi ezokunakekelwa kwezempilo namabhange, kanye nobuchwepheshe.
5. Amanethiwekhi Wokufunda Okujulile (DRLs)
Deep Ukuqinisa Ukufunda Amanethiwekhi (DRLs) ahlanganisa amanethiwekhi e-neural ajulile namasu okuqinisa ukufunda ukuze avumele ama-ejenti ukuthi afunde endaweni eyinkimbinkimbi ngokuzama nangephutha.
Ama-DRL asetshenziselwa ukufundisa abenzeli indlela yokuthuthukisa isiginali yomvuzo ngokusebenzisana nendawo ebazungezile nokufunda emaphutheni abo.
Yini Ebenza Baphawuleke?
Zisetshenziswe ngempumelelo ezinhlelweni ezihlukene, ezifaka amageyimu, amarobhothi, nokushayela ngokuzenzakalelayo. Ama-DRL abalulekile ngoba angafunda ngokuqondile ekufakweni kwezinzwa okungaphekiwe, okuvumela ama-ejenti ukuthi enze izinqumo ngokusekelwe ekusebenzelaneni kwawo nendawo ezungezile.
Izinhlelo zokusebenza ezibalulekile
Ama-DRL asetshenziswa ezimweni zomhlaba wangempela ngoba angakwazi ukusingatha izinkinga ezinzima.
Ama-DRL afakwe kuma-software amaningi avelele kanye nezinkundla zobuchwepheshe, okuhlanganisa ne-OpenAI's Gym, I-ML-Agents ye-Unity, kanye ne-DeepMind Lab yakwa-Google. AlphaGo, eyakhiwe ngabakwa-Google Deepmind, isibonelo, isebenzisa i-DRL ukuze idlale igeyimu yebhodi ethi Hamba ezingeni lompetha bomhlaba.
Okunye ukusetshenziswa kwe-DRL kukumarobhothi, lapho isetshenziselwa khona ukulawula ukunyakaza kwezingalo zerobhothi ukwenza imisebenzi efana nokubamba izinto noma amabhlogo wokupakisha. Ama-DRL anokusetshenziswa okuningi futhi ayithuluzi eliwusizo ama-ejenti okuqeqesha ukuze afunde futhi wenze izinqumo ezimweni eziyinkimbinkimbi.
6. Ama-autoencoder
Ama-Autoencoder awuhlobo oluthokozisayo lwe inethiwekhi ye-neural lokho kubambe intshisekelo yazo zombili izazi nososayensi bedatha. Zenzelwe ngokuyisisekelo ukufunda indlela yokucindezela nokubuyisela idatha.
Idatha yokufaka inikezwa ngokulandelana kwezendlalelo ezehlisa kancane kancane ubukhulu bedatha ize iminyaniswe ibe isendlalelo sebhodlela esinamanodi ambalwa kunezendlalelo zokokufaka nezokuphumayo.
Lesi sethulo esicindezelwe sibe sesisetshenziswa ukuze kudalwe kabusha idatha yokokufaka yoqobo kusetshenziswa ukulandelana kwezendlalelo ezikhuphula kancane kancane ubukhulu bedatha bubuyele esimweni sayo sokuqala.
Kungani Kubalulekile?
Ama-Autoencoder ayingxenye ebalulekile ukufunda okujulile ngoba benza ukukhishwa kwesici nokwehliswa kwedatha kwenzeke.
Bayakwazi ukuhlonza izici ezibalulekile zedatha engenayo futhi bazihumushele kwifomu elicindezelwe elingase lisetshenziswe kweminye imisebenzi efana nokuhlukanisa, ukuhlanganisa, noma ukudalwa kwedatha entsha.
Siwasebenzisaphi Ama-Autoencoder?
Ukutholwa okungaqondakali, ukucutshungulwa kolimi lwemvelo, kanye umbono wekhompyutha ezinye zeziyalo ezimbalwa lapho kusetshenziswa khona ama-autoencoder. Ama-autoencoder, isibonelo, angasetshenziselwa ukuminyanisa isithombe, ukususa umsindo, nokuhlanganiswa kwezithombe ekubonweni kwekhompyutha.
Singasebenzisa ama-Autoencoder emisebenzini efana nokudala umbhalo, ukuhlelwa kombhalo, nokufingqa kombhalo ekucubungulweni kolimi lwemvelo. Ingahlonza umsebenzi ongaqondakali kudatha echezuka kokujwayelekile ekukhombeni okudidayo.
7. Amanethiwekhi we-Capsule
I-Capsule Networks iyisakhiwo sokufunda esijulile esisha esakhiwa esikhundleni se-Convolutional Neural Networks (CNNs).
Ama-Capsule Networks asekelwe emcabangweni wokuhlanganisa amayunithi obuchopho abizwa ngokuthi ama-capsules anomthwalo wemfanelo wokubona ubukhona bento ethile esithombeni futhi ibhale ngekhodi izici zayo, njengokuma kwayo nendawo, kumavekhtha azo aphumayo. Ngakho-ke i-Capsule Networks ingaphatha ukusebenzisana kwendawo nokuguquguquka kokubuka kangcono kunama-CNN.
Kungani sikhetha ama-Capsule Networks kune-CNN's?
I-Capsule Networks iyasiza ngoba inqoba ubunzima be-CNN ekuthwebuleni ubudlelwano besigaba phakathi kwezinto ezisesithombeni. Ama-CNN angakwazi ukubona izinto ezinosayizi abahlukahlukene kodwa akuthola kunzima ukuqonda ukuthi lezi zinto zixhumana kanjani nezinye.
Ngakolunye uhlangothi, i-Capsule Networks, ingafunda ukubona izinto nezingcezu zazo, kanye nendlela ezibekwe ngayo endaweni esithombeni, okuzenza zibe yimbangi esebenzayo yezinhlelo zokusebenza zombono wekhompyutha.
Izindawo Zokusebenza
I-Capsule Networks isivele ibonise imiphumela ethembisayo ezinhlobonhlobo zezinhlelo zokusebenza, okuhlanganisa ukuhlukaniswa kwesithombe, ukuhlonza into, nokuhlukaniswa kwesithombe.
Zisetshenziselwe ukuhlukanisa izinto ezithombeni zezokwelapha, ukubona abantu kumafilimu, ngisho nokudala amamodeli e-3D ezithombeni ze-2D.
Ukuze kwandiswe ukusebenza kwawo, ama-Capsule Networks ahlanganiswe nezinye izakhiwo zokufunda ezijulile ezifana ne-Generative Adversarial Networks (GANs) kanye ne-Variational Autoencoder (VAEs). I-Capsule Networks ibikezelwa ukuthi izodlala indima ebaluleke kakhulu ekuthuthukiseni ubuchwepheshe bokubona ngekhompyutha njengoba isayensi yokufunda okujulile ithuthuka.
Ngokwesibonelo; Nibabel iyithuluzi elaziwa kakhulu lePython lokufunda nokubhala izinhlobo zamafayela we-neuroimaging. Ngokuhlukaniswa kwesithombe, kusebenzisa ama-Capsule Networks.
8. Amamodeli asekelwe ekunakeni
Amamodeli okufunda ajulile aziwa ngokuthi amamodeli asuselwe ekunakekeni, abuye aziwe ngokuthi yizindlela zokunaka, alwela ukukhulisa ukunemba amamodeli wokufunda wemishini. Lawa mamodeli asebenza ngokugxila ezicini ezithile zedatha engenayo, okuholela ekucutshungulweni okuphumelelayo nangempumelelo.
Emisebenzini yokucubungula ulimi lwemvelo njengokuhumusha komshini nokuhlaziya imizwa, izindlela zokunaka zibonise ukuthi ziphumelele impela.
Kuyini Ukubaluleka Kwazo?
Amamodeli asuselwe ekunakekelweni awusizo ngoba anika amandla ukucutshungulwa okusebenzayo nokusebenza kahle kwedatha eyinkimbinkimbi.
Amanethiwekhi e-neural endabuko hlola yonke idatha yokufaka njengebaluleke ngokulinganayo, okuholela ekucutshungulweni okuhamba kancane nokuncipha kokunemba. Izinqubo zokunaka zigxila ezicini ezibalulekile zedatha yokufaka, okuvumela ukuqagela okusheshayo nokunembe kakhudlwana.
Izindawo Zokusebenzisa
Emkhakheni wobuhlakani bokwenziwa, izindlela zokunaka zinezinhlobonhlobo zezinhlelo zokusebenza, okuhlanganisa ukucutshungulwa kolimi lwemvelo, ukubonwa kwezithombe nomsindo, kanye nezimoto ezingashayeli.
Izindlela zokunaka, isibonelo, zingasetshenziswa ukuthuthukisa ukuhumusha komshini ekucutshungulweni kolimi lwemvelo ngokuvumela isistimu ukuthi igxile kumagama athile noma imishwana ebalulekile kumongo.
Izindlela zokunaka ezimotweni ezizimele zingasetshenziswa ukusiza isistimu ukuthi igxile ezintweni ezithile noma izinselele endaweni ezungezile.
9. Transformer Networks
Amanethiwekhi e-Transformer angamamodeli okufunda ajulile ahlola futhi akhiqize ukulandelana kwedatha. Asebenza ngokucubungula ukulandelana kokufaka into eyodwa ngesikhathi futhi akhiqize ukulandelana kokukhiphayo kobude obufanayo noma obuhlukile.
Amanethiwekhi e-Transformer, ngokungafani namamodeli ajwayelekile okulandelana-kuya-kulandelana, awacubunguli ukulandelana esebenzisa amanethiwekhi e-neural aqhubekayo (ama-RNN). Kunalokho, basebenzisa izinqubo zokuzinaka ukuze bafunde izixhumanisi phakathi kwezingcezu zokulandelana.
Kuyini Ukubaluleka Kwamanethiwekhi eTransformer?
Amanethiwekhi e-Transformer akhule ngokuduma eminyakeni yamuva nje ngenxa yokusebenza kwawo okungcono emisebenzini yokucubungula ulimi lwemvelo.
Ziyifanele kakhulu imisebenzi yokudala umbhalo njengokuhumusha ulimi, ukufinyezwa kombhalo, nokukhiqizwa kwengxoxo.
Amanethiwekhi eTransformer asebenza kahle kakhulu ngokubala kunamamodeli asuselwa ku-RNN, okuwenza abe yinketho ekhethwayo yezinhlelo zokusebenza ezinkulu.
Ungawatholaphi Amanethiwekhi eTransformer?
Amanethiwekhi e-Transformer asetshenziswa kabanzi ezinhlobonhlobo zezinhlelo zokusebenza, ikakhulukazi ukucubungula ulimi lwemvelo.
Uchungechunge lwe-GPT (Generative Pre-trained Transformer) luyimodeli evelele esekelwe ku-transformer esetshenziselwe imisebenzi efana nokuhumusha ulimi, ukufinyezwa kombhalo, nokwenza i-chatbot.
I-BERT (Ukumelwa kwe-Bidirectional Encoder from Transformers) ingenye imodeli evamile esekelwe ku-transformer eye yasetshenziselwa izinhlelo zokusebenza zokuqonda ulimi lwemvelo ezifana nokuphendula imibuzo nokuhlaziya imizwelo.
Bobabili I-GPT kanye ne-BERT yadalwa nge I-PyTorch, uhlaka lokufunda okujulile olunomthombo ovulekile obeludumile ekuthuthukiseni amamodeli asekelwe ku-transformer.
10. Imishini ye-Boltzmann Ekhawulelwe (ama-RBM)
Imishini ye-Boltzmann Ekhawulelwe (ama-RBM) iwuhlobo lwenethiwekhi ye-neural engagadiwe efunda ngendlela yokukhiqiza. Ngenxa yekhono labo lokufunda nokukhipha izici ezibalulekile kudatha enobukhulu obuphezulu, zisetshenziswe kabanzi emikhakheni yokufunda komshini nokufunda okujulile.
Ama-RBM akhiwe ngezingqimba ezimbili, ezibonakalayo nezifihliwe, futhi isendlalelo ngasinye esineqembu lama-neurons axhunywe imiphetho enesisindo. Ama-RBM aklanyelwe ukufunda ukusatshalaliswa kwamathuba achaza idatha yokufaka.
Iyini Imishini Ye-Boltzmann Ekhawulelwe?
Ama-RBM asebenzisa isu lokufunda elikhiqizayo. Kuma-RBM, isendlalelo esibonakalayo sibonisa idatha yokufaka, kuyilapho isendlalelo esingcwatshiwe sibhala izici zedatha yokufaka. Izisindo zezendlalelo ezibonakalayo nezifihliwe zibonisa amandla esixhumanisi sazo.
Ama-RBM alungisa izisindo nokuchema phakathi kwezendlalelo ngesikhathi sokuqeqeshwa esebenzisa indlela eyaziwa ngokuthi ukuhluka okuhlukile. Ukwehlukana okuphambene kuyisu lokufunda elingagadiwe elikhulisa amathuba okuqagela imodeli.
Yini ukubaluleka kweMishini ye-Boltzmann Ekhawulelwe?
Ama-RBM abalulekile ku ukufunda imishini nokufunda okujulile ngoba bangafunda futhi bakhiphe izici ezifanele emananini amakhulu edatha.
Zisebenza kahle kakhulu ekubonweni kwesithombe nenkulumo, futhi zisetshenziswe kuzinhlelo zokusebenza ezihlukene ezifana namasistimu wokuncoma, ukutholwa okudidayo, nokunciphisa ubukhulu. Ama-RBM angathola amaphethini kumadathasethi amaningi, okuholela ekuqaguleni okuphezulu nemininingwane.
Ingasetshenziswa kuphi Imishini Ye-Boltzmann Ekhawulelwe?
Izicelo zama-RBM zifaka phakathi ukuncishiswa kobukhulu, ukutholwa okudidayo, namasistimu wokuncoma. Ama-RBM awusizo ikakhulukazi ekuhlaziyeni imizwelo kanye ukumodela isihloko esimeni sokucutshungulwa kolimi lwemvelo.
Amanethiwekhi ezinkolelo ezijulile, uhlobo lwenethiwekhi ye-neural esetshenziselwa ukuqashelwa kwezwi nesithombe, aphinde asebenzise ama-RBM. I-Deep Belief Network Toolbox, I-TensorFlow, Futhi Theano ezinye zezibonelo ezithile zesofthiwe noma ubuchwepheshe obusebenzisa ama-RBM.
Qedani
Amamodeli e-Deep Learning aya ngokuya ebaluleka kakhulu ezimbonini ezahlukahlukene, okubandakanya ukuqaphela inkulumo, ukucubungula ulimi lwemvelo, nokubona ngekhompyutha.
I-Convolutional Neural Networks (CNNs) kanye Ne-Recurrent Neural Networks (RNNs) abonise isithembiso esikhulu futhi asetshenziswa kakhulu ezinhlelweni eziningi, nokho, wonke amamodeli e-Deep Learning anezinzuzo zawo kanye nokubi.
Nokho, abacwaningi basabheka Imishini Ye-Boltzmann Ekhawulelwe (ama-RBM) nezinye izinhlobo zamamodeli e-Deep Learning ngoba nawo anezinzuzo ezikhethekile.
Kulindeleke ukuthi kudalwe amamodeli amasha nanobuchule njengoba indawo yokufunda ejulile iqhubeka nokuthuthuka ukuze kubhekwane nezinkinga ezinzima.
shiya impendulo