Ngaba wakha wazibuza ukuba imoto eziqhubayo iyazi njani ukuba imele ime kwisibane esibomvu okanye ukuba ifowuni yakho inokubona njani ubuso bakho?
Kulapho iConvolutional Neural Network okanye iCNN ngokufutshane ingena khona.
I-CNN ithelekiseka nengqondo yomntu ekwazi ukuhlalutya imifanekiso ukuze ibone ukuba kwenzeka ntoni kuyo. Ezi ndlela zothungelwano ziyakwazi ukubona nezinto abantu ebebenokungazijongi!
Kule post, siza kuhlola i-CNN kwi ukufunda okunzulu umxholo. Makhe sibone ukuba lo mmandla uchulumancisayo unokusinika ntoni na!
Yintoni iMfundo eNzululwazi?
Ukufunda nzulu luhlobo oluthile kukubhadla okungeyonyani. Ivumela iikhompyutha ukuba zifunde.
Iinkqubo zokufunda nzulu idatha kusetyenziswa iimodeli ezintsonkothileyo zemathematika. Ke ngoko, ikhompyuter iyakwazi ukubona iipateni kwaye ihlukanise idatha.
Emva koqeqesho olunemizekelo emininzi, inokwenza izigqibo.
Kutheni sinomdla kwii-CNN kwiSifundo esiNzulu?
Convolutional Neural Networks (CNNs) yinxalenye ebalulekileyo yokufunda nzulu.
Bavumela iikhompyuter ukuba ziqonde imifanekiso kunye nezinye idatha ebonakalayo. Sinokuqeqesha iikhomputha ukuba zibone iipatheni kwaye zichonge izinto ezisekelwe kwinto "ezibonayo" ngokuqesha ii-CNN ekufundeni nzulu.
Ii-CNNs zisebenza njengamehlo okufunda nzulu, zinceda iikhompyutha ekuqondeni okusingqongileyo!
Impembelelo evela kwi-Brain's Architecture
Ii-CNN zithatha ukuphefumlelwa kwazo kwindlela ingqondo etolika ngayo ulwazi. I-neurons yokwenziwa, okanye ii-node, kwii-CNNs, zamkela amagalelo, ziwaqhubele phambili, kwaye zihambise iziphumo njengemveliso, kanye ngendlela ii-neuron zobuchopho zenza ngayo yonke imizimba.
Uluhlu Lokufaka
Umaleko wegalelo lomgangatho inethiwekhi yomnatha ifumana amagalelo ngendlela yoluhlu, njengeepikseli zemifanekiso. Kwii-CNNs, umfanekiso unikezelwa njengegalelo kumaleko okufakwayo.
Iileya ezifihlakeleyo
Kukho iileya ezininzi ezifihliweyo kwii-CNN, ezisebenzisa imathematika ukukhupha iimpawu kumfanekiso. Kukho iindidi ezininzi zeeleya, ezibandakanya ukudibanisa ngokupheleleyo, iiyunithi zomgca ezilungisiweyo, ukudityaniswa, kunye neeleya zeconvolution.
Convolution Layer
Umaleko wokuqala wokukhupha iimpawu kumfanekiso wegalelo ngumaleko weconvolution. Umfanekiso wegalelo uphantsi kohluzo, kwaye isiphumo yimephu yefitsha eqaqambisa izinto eziphambili zomfanekiso.
Ukudibanisa Kamva
Umaleko wokudibanisa usetyenziselwa ukucutha ubungakanani bemephu yesici. Iqinisa ukuchasana kwemodeli ekutshintsheni indawo yomfanekiso wegalelo.
Ulwaleko lweYunithi eLungisiweyo (ReLU)
Umaleko we-ReLU uqeshwe ukunika imodeli engahambelaniyo. Isiphumo somaleko owandulelayo sivulwe ngulo maleko.
Umaleko oQhagamshelwe ngokupheleleyo
Umaleko oqhagamshelwe ngokupheleleyo uhlela into kwaye uyinike i-ID ekhethekileyo kuluhlu lwemveliso ngumaleko oqhagamshelwe ngokupheleleyo.
Ii-CNNs ziiNethiwekhi zeFeedforward
Idatha ihamba kuphela ukusuka kwigalelo ukuya kwiziphumo ngendlela enye. Uyilo lwabo luphefumlelwe yi-cortex ebonakalayo yengqondo, eyenziwe ngamaleko atshintshanayo weeseli ezisisiseko kunye nezintsonkothileyo.
Ziqeqeshwa njani ii-CNNs?
Cinga ukuba uzama ukufundisa ikhompyutha ukuchonga ikati.
Ubonisa kuyo imifanekiso emininzi yeekati ngelixa usithi, "Nantsi ikati." Emva kokujonga imifanekiso eyaneleyo yeekati, ikhomputha iqala ukubona iimpawu ezinjengeendlebe ezitsolo kunye namabhovu.
Indlela iCNN esebenza ngayo iyafana. Kukho iifoto ezininzi ekhompyutheni, kwaye amagama ezinto ezikumfanekiso ngamnye anikiwe.
Nangona kunjalo, i-CNN yohlula imifanekiso ibe ngamaqhekeza amancinci, njengemimandla. Kwaye, ifunda ukuchonga iimpawu kuloo mimandla kunokujonga nje imifanekiso iyonke.
Ke, umaleko wokuqala we-CNN unokubona kuphela iimpawu ezisisiseko njengemiphetho okanye iikona. Emva koko, umaleko olandelayo wakhela kuloo nto ukuqonda iimpawu ezineenkcukacha ezingakumbi njengeefom okanye ukuthungwa.
Amaleko ahlala ehlengahlengisa kwaye ephucula ezo mpawu njengoko ikhompyuter ijonga imifanekiso emininzi. Iyaqhubeka ide ibe nobuchule ekuchongeni nantoni na ebiqeqeshelwe kuyo, nokuba ziikati, ubuso, okanye nantoni na eyenye.
Isixhobo esinamandla sokufunda ngokunzulu: Indlela i-CNNs eguqule ngayo ukuQwaqwa koMfanekiso
Ngokuchonga kunye nokwenza ingqiqo yeepateni kwimifanekiso, ii-CNNs, ziguqule ukuqatshelwa komfanekiso. Kuba zibonelela ngeziphumo ngezinga eliphezulu lokuchaneka, ii-CNNs zezona zakhiwo zisebenzayo zokuhlelwa kwemifanekiso, ukufunyanwa, kunye nokufumanisa izicelo.
Ngokuqhelekileyo bavelisa iziphumo ezintle. Kwaye, bakhomba ngokuchanekileyo kwaye bachonge izinto kwiifoto kusetyenziso lwehlabathi lokwenyani.
Ukufumana iipateni kulo Naliphi na iNdawo yoMfanekiso
Nokuba ipateni ivela phi emfanekisweni, ii-CNNs ziyilelwe ukuyiqonda. Bangakwazi ukukhupha ngokuzenzekelayo iimpawu ezibonakalayo kuyo nayiphi na indawo emfanekisweni.
Oku kunokwenzeka ngenxa yesakhono sabo esibizwa ngokuba "yi-spatial invariance." Ngokwenza lula inkqubo, ii-CNNs zinokufunda ngokuthe ngqo kwiifoto ngaphandle kwesidingo sokutsalwa kweempawu zabantu.
Isantya sokuLungiselela ngakumbi kunye neMemori encinci eSetyenzisiweyo
Iinkqubo ze-CNN zemifanekiso ngokukhawuleza nangempumelelo kuneenkqubo zemveli. Esi sisiphumo semaleko okudibanisa, athoba inani leeparamitha ezifunekayo ukusetyenzwa komfanekiso.
Ngale ndlela, banciphisa ukusetyenziswa kwememori kunye neendleko zokulungisa. Iindawo ezininzi zisebenzisa ii-CNN, ezifana; ukuqaphela ubuso, ukuhlelwa kwevidiyo, kunye nohlalutyo lwemifanekiso. Bade baqhele hlela iminyele.
Ubomi bokwenyani imizekelo
Imifanekiso kaGoogle yindlela enye yokusetyenziswa kwee-CNN kwihlabathi lokwenyani eliziqeshayo ukuchonga abantu kunye nezinto ezisemifanekisweni. Ngaphezu koko, I-Azure kwaye Amazon bonelela ngemifanekiso ye-APIs ethegi kwaye ichonge izinto kusetyenziswa ii-CNNs.
Ujongano lwe-intanethi loqeqesho kuthungelwano lwe-neural usebenzisa iiseti zedatha, kubandakanywa nemisebenzi yokuqaphela imifanekiso, ibonelelwa liqonga lokufunda nzulu. Iidijithi zeNVIDIA.
Ezi zicelo zibonisa indlela ii-CNNs ezinokuthi zisetyenziswe ngayo kwimisebenzi eyahlukeneyo, ukusuka kwiimeko ezincinci zokusetyenziswa kwezorhwebo ukuya ekulungiseleleni iifoto zomntu. Mininzi eminye imizekelo ekunokucingwa ngayo.
Ingaba iConvolutional Neural Networks iyakuvela njani?
Ukhathalelo lwempilo lishishini elinomdla apho ii-CNNs kulindeleke ukuba zibe nempembelelo enkulu. Ngokomzekelo, zinokusetyenziselwa ukuvavanya imifanekiso yezonyango njenge-X-reyi kunye ne-MRI scans. Banokuncedisa oogqirha ngokukhawuleza nangokuchanekileyo ukufumanisa izifo.
Iimoto eziziqhubayo sesinye isicelo esinomdla apho ii-CNNs zinokusetyenziselwa ukuchonga into. Inokuphucula indlela izithuthi eziqonda ngayo kwaye zisabela kakuhle kwindawo ezingqongileyo.
Inani elikhulayo labantu likwanomdla ekudaleni izakhiwo ze-CNN ezikhawulezayo nezisebenzayo ngakumbi, kubandakanywa nee-CNN ezihambayo. Kulindeleke ukuba zisetyenziswe kwigajethi zamandla aphantsi njengee-smartphones kunye nezinto ezinxitywayo.
Shiya iMpendulo