Ngaphandle kokungabaza, umhlaba wesayensi yamakhompiyutha nokufunda ngomshini ukhula ngokweqile. Kulesi senzakalo kufanele sikhulume ngomsebenzi wakamuva owenziwe yithimba labaphenyi abangabodumo IKamitani Lab we IKyoto University (Japan), abakwazile ukwenza isoftware, ngokwezivivinyo zokuqala, ekwaziyo ukufunda nokukhiqiza esikrinini lokho umuntu akucabangayo.
Ngaphandle kokungabaza, sikhuluma ngokuthuthuka okubaluleke kakhulu kunalokho okungabonakala ekuqaleni umthelela iphrojekthi efana nalena engaba nawo kunoma iyiphi indawo yezimpilo zethu mikhulu impela Futhi konke lokhu, kunjalo, ngaphandle kokungena ezindabeni zezempi, umkhakha ozokuthakazelisa ngokukhethekile ukwazi mathupha ukuthi ubuchwepheshe obufana nalobu buqhubekela kanjani phambili.
Bakha inethiwekhi ye-neural ekwazi ukubona lokho ocabanga ngakho
Njengoba sekwenzekile ephepheni elishicilelwe ngomagazini ohlonishwayo nowaziwayo njenge Isayensi, ngokusobala inselelo eyinkimbinkimbi kunazo zonke ebhekane nethimba labaphenyi baseJapan kube ukufeza dala i-algorithm ekwazi ukuhumusha futhi, ngaphezu kwakho konke, yokukhiqiza ngokunembile izithombe ezirekhodwe ubuchopho bomuntu ngemuva kokubabuka esikrinini. Ngokulandelayo, i-algorithm nayo iyakwazi ukutolika nokukhiqiza kabusha lokho umuntu akukhumbulayo kusuka ezithombeni azibonile phambilini.
Ngokuphikisana nalokho okungabonakala, sibhekene nengqophamlando okungekho noyedwa okwazile ukuyenza kabusha ngenye indlela. Ukubeka le ntuthuko kancane futhi sikuqonde kangcono okufeziwe, mane ngikutshele ukuthi kuze kube manje, Imizamo yokwakha i-algorithm ekwazi ukwenza lo msebenzi yayi, ngokubabiza ngandlela thile, kulinganiselwe impela. Umehluko phakathi komsebenzi owenziwe nephuzu lapho amaqembu amaningana ayefinyelele khona ukuthi le nethiwekhi ye-neural ayigcinanga ngokwandisa ukuqonda nokukhiqizwa kwekhompyutha kwezithombe zengqondo, kepha futhi ivumele amafomu akhona kuphela ukuthi ahunyushwe futhi aphindwe. umcabango womuntu.
Amavolontiya amathathu enele ukuqeqesha le software ethokozisayo
Njengoba sekuchaziwe, ukusiza ukuqeqeshwa nokuthuthukiswa kwenethiwekhi ye-neural, abacwaningi abakha iqembu banquma ukusebenzisa amavolontiya amathathu anombono ojwayelekile kulabo abebezokwethula izithombe ezahlukahlukene zezigaba ezifana nemvelo, izinhlamvu kanye namajamo wejometri.
Umqondo ngalokhu kuhlolwa ukuthi lapho kubhekwa izithombe, kuzokwenziwa umsebenzi ku-cortex yobuchopho yalezi zisebenzi zokuzithandela ezizodluliselwa futhi zihlaziywe kwinethiwekhi ye-neural. Ukuze ikhule futhi ifunde ekuphenduleni okuhlukile, amavolontiya kwakudingeka enze njalo buka izithombe ezingaphezu kwezingu-1.000 XNUMX izikhathi eziningi. Phakathi kwezithombe, ukuze sikwazi ukuthola umqondo ongcono, singathola inhlanzi, ukubukeka okulula kwemibala noma indiza.
Ngaphandle kwenqubekela phambili enkulu elethwe ukwenziwa kwale software, abathuthukisi bayo basenomsebenzi omningi ngaphambi kwabo
Ukurekhoda umsebenzi wobuchopho wevolontiya ngalinye, abacwaningi basebenzise amasu afana nalawa i-resonance esebenzayo kazibuthe, elinganisa ukugeleza kwegazi okukhona ezifundeni ezithile zobuchopho futhi ngaleyo ndlela iveze umsebenzi we-neural. Ngesithombe ngasinye senziwa umsebenzi wobuchopho bomuntu ngamunye. Ngenxa yalo msebenzi omkhulu kangaka, ekugcineni kungenzeka ukuthi ikhompyutha ibe namandla anele okwakha kabusha isithombe esivela ebuchosheni obenziwa yilowo nalowo muntu nganoma isiphi isikhathi.
Njengemininingwane, mane ubeke amazwana lokho ukwakhiwa kabusha kwesithombe akuyona into esheshayo kodwa inethiwekhi ye-neural ichaza okufanayo kumarounds angaba ngu-200 selokhu umbono wokuthi umuntu uthola ukusabela umuntu anakho esithombeni esethulwe noma esikhunjulwayo kufanele uqhathaniswe ngokoqobo nalokho aye wakugcina. Ekugcineni, njengemininingwane, hhayi kuphela inethiwekhi ye-neural ekwazi ukuphindaphinda isithombe sobuchopho, kepha ithola okuyiqiniso ngokwengeziwe ngenxa ye-algorithm ekhethekile esetshenziswe ngokungeziwe.