Ngaphandle kwamathandabuzo, ihlabathi lekhompyutha kunye nokufunda koomatshini likhula ngokuxhuma kunye nemida. Ngesi sihlandlo kufuneka sithethe ngomsebenzi wamva nje ophuhliswe liqela labaphandi abangabadumileyo Kamitani Lab de la IYunivesithi yaseKyoto (eJapan), abaye bakwazi ukuphuhlisa isoftware ethi, ngokwemifuniselo yokuqala, ikwazi ukufunda nokuvelisa kwiscreen oko umntu akucingayo.
Ngaphandle kwamathandabuzo sithetha malunga nokuhamba phambili okufanelekileyo kunokuba kunokubonakala ngathi xa uqala, kuba Impembelelo enokuthi iprojekthi enjengale ibe nayo kuyo nayiphi na indawo yobomi bethu inkulu kakhulu. kwaye konke oku, ngokuqinisekileyo, ngaphandle kokungena kwimiba yezomkhosi, icandelo eliya kuba nomdla ngakumbi ekwazini indlela iteknoloji efana nale nkqubela phambili ngayo.
Baphuhlisa inethiwekhi ye-neural ekwaziyo ukuqaphela into oyicingayo
Njengoko kutyhilwe kwiphepha elipapashwe ngemagazini edumileyo neyaziwayo njenge inzululwazi, ekubonakala ukuba olona celomngeni luntsonkothileyo eliye lajamelana nalo iqela labaphandi baseJapan kukufumana yenza i-algorithm ekwaziyo ukutolika kwaye, ngaphezu kwakho konke, ukuvelisa ngokuchanekileyo imifanekiso erekhodiweyo yingqondo yomntu. emva kokubabona kwiscreen. Kwangaxeshanye, i-algorithm iyakwazi ukutolika nokuvelisa into umntu ayikhumbulayo kwimifanekiso ayibonayo ngaphambili.
Ngokuchaseneyo noko kubonakala ngathi, sijongene nesiganeko esibalulekileyo ekungekho mntu ukwaziyo ukusidala kwakhona nangayiphi na enye indlela. Ukubeka olu phuhliso luncinci kwaye uqonde ngcono oko kuphunyeziweyo, xelela nje ukuba ukuza kuthi ga ngoku, Iinzame zokuphuhlisa i-algorithm ekwazi ukwenza lo msebenzi yayi, ukubabiza ngandlela thile, lilinganiselwe. Umahluko phakathi komsebenzi owenziweyo kunye nenqanaba apho amaqela aliqela aye afikelela khona kwinto yokuba le nethiwekhi ye-neural ayizange yandise nje ukuqonda kunye nokuveliswa kwemifanekiso yengqondo ngekhompyutha kodwa ikwavumele iifom ebezikho ngaphambili ukuba zitolikwe kwaye zitolikwe. kwingcinga yomntu.
Amavolontiya amathathu anele ukuqeqesha le software inomdla
Njengoko kuchaziwe, ukunceda uqeqesho kunye nophuhliso lwenethiwekhi ye-neural, abaphandi abenza iqela banquma ukusebenzisa amavolontiya amathathu anombono oqhelekileyo ababeza kubonisa kubo iifoto ezahlukeneyo zeendidi ezifana nendalo, oonobumba kunye neemilo zejometri.
Ingcinga ngolu lingelo yayikukuba xa kujongwa imifanekiso, umsebenzi uya kuveliswa kwi-cortex ye-cerebral yelinye lamavolontiya aya kuhanjiswa kwaye ahlalutywe kwinethiwekhi ye-neural. Ukuze oku kuphuhle kwaye kufundwe kwiindlela ezahlukeneyo zokusabela, amavolontiya kwafuneka enjenjalo jonga ngaphezu kwe-1.000 yemifanekiso ngamaxesha amaninzi. Phakathi kwemifanekiso, ukuze sifumane ingcamango engcono, sinokufumana intlanzi, imilo enemibala elula okanye indiza.
Ngaphandle kwenkqubela phambili enkulu eyenziwe yindalo yale software, abaphuhlisi bayo basenomsebenzi omninzi phambi kwabo.
Ukurekhoda umsebenzi wengqondo wevolontiya ngalinye, abaphandi basebenzisa iindlela ezifana isandi esisebenzayo semagneti, elinganisa ukuhamba kwegazi ekhoyo kwimimandla ethile yengqondo kwaye ngaloo ndlela ityhila umsebenzi we-neuronal. Ngomfanekiso ngamnye, umsebenzi wobuchopho bomntu wahlalutywa.. Enkosi kulo msebenzi mkhulu kangaka, ekugqibeleni kwaba nokwenzeka ukuba ikhomputha ibenomthamo owaneleyo wokwakha kwakhona umfanekiso osuka kumsebenzi wobuchopho lowo nawuphi na umntu ebewubonisa nangaliphi na ixesha.
Njengenkcukacha, phawula nje loo nto Ukwakhiwa kwakhona komfanekiso akwenzeki kwangoko kodwa iprofayile yenethiwekhi ye-neural iwuchaza ngaphezulu kwemijikelo engama-200. ekubeni umele uthelekise ngokoqobo imbono oyifumanayo kwindlela umntu asabela ngayo kumfanekiso owenziweyo okanye okhunjulweyo naleyo uyigcinileyo. Ekugqibeleni, njengenkcukacha, ayisiyiyo kuphela inethiwekhi ye-neural elawula ukuphindaphinda umfanekiso wengqondo, kodwa ikwafumana ubunyani obungakumbi bombulelo kwi-algorithm ekhethekileyo ethe yaphunyezwa.