Ngaphantsi kweminyaka emibini edlulileyo, i-DARPA yasungula inkqubo ye-Lifelong Learning Machines (L2M) ukuze idale ngokuqhubekayo ukufunda iinkqubo zerobhothi ezinezinto zobukrelekrele bokwenziwa. Inkqubo ye-L2M yayifanele ukuba ikhokele ekuveleni kwamaqonga okuzifundela anokuziqhelanisa nendawo entsha ngaphandle kweprogram okanye uqeqesho lwangaphambili. Ukubeka ngokulula, iirobhothi kwafuneka zifunde kwiimpazamo zabo, kwaye zingafundi ngokumpompa iisethi zedatha yetemplate kwindawo yebhubhoratri.
Inkqubo ye-L2M ibandakanya amaqela ophando angama-30 anezixa-mali ezahlukeneyo zenkxaso-mali. Kutshanje, elinye lamaqela avela kwiYunivesithi yaseSouthern California libonise inkqubela phambili eqinisekileyo ekudaleni amaqonga e-robotic okufundayo, njengoko kuchaziwe kwinqaku likaMatshi loBukrelekrele boMshini weNdalo.
Iqela labaphandi beyunivesithi likhokelwa nguFrancisco J. Valero-Cuevas, uprofesa wobunjineli bezinto eziphilayo, i-biokinesiology kunye nonyango lomzimba. Ngokusekelwe kwi-algorithm ephuhliswe liqela, esekelwe kwiindlela ezithile zokusebenza kwezinto eziphilayo, ulandelelwano lwezenzo zobukrelekrele bokwenziwa zadalwa ukufundisa iintshukumo zerobhothi kumalungu amane. Kuxelwa ukuba amalungu okwenziwa ngendlela yokulinganisa i-tendon, imisipha kunye namathambo akwazi ukufunda ukuhamba phakathi kwemizuzu emihlanu emva kokuqhuba i-algorithm.
Emva kokuqaliswa kokuqala, inkqubo yayingacwangciswanga kwaye inesiphithiphithi, kodwa ke i-AI yaqala ukuziqhelanisa ngokukhawuleza nezinto eziyinyani kwaye yaqala ngempumelelo ukuhamba ngaphandle kwenkqubo yangaphambili. Kwixesha elizayo, indlela eyiliweyo yoqeqesho lobomi bonke beerobhothi ngaphandle koqeqesho lokuqala lweML kunye neeseti zedatha inokuhlengahlengiswa ukuze kuxhotyiswe iimoto zabahlali ngoopilothi kunye neenqwelo zomkhosi zerobhothi. Nangona kunjalo, le teknoloji inamathuba amaninzi kunye neendawo zokusetyenziswa. Into ephambili kukuba i-algorithm ayiboni umntu njengomqobo ekuphuhliseni kwaye ayifundi nantoni na embi.
umthombo: 3dnews.ru