Asingasviki makore maviri apfuura, DARPA yakatanga chirongwa cheLifelong Learning Machines (L2M) kuti igadzire kuramba ichidzidza marobhoti masisitimu ane zvinhu zvehungwaru hwekugadzira. Chirongwa cheL2M chaifanirwa kutungamira pakubuda kwemapuratifomu ekuzvidzidzira ayo anogona kuzvigadzirisa kune imwe nharaunda isina kutanga chirongwa kana kudzidziswa. Zvichitaurwa zviri nyore, marobhoti aifanira kudzidza kubva mukukanganisa kwavo, uye kwete kudzidza nekupomba seti yetemplate data munzvimbo yerabhoritari.
Chirongwa cheL2M chinosanganisira mapoka makumi matatu ekutsvagisa ane huwandu hwakasiyana hwemari. Nguva pfupi yapfuura, rimwe remapoka kubva kuYunivhesiti yeSouthern California rakaratidza kufambira mberi kunogutsa mukugadzira kuzvidzidzira marobhoti mapuratifomu, sezvakashumwa muna March chinyorwa cheNature Machine Intelligence.
Boka revatsvakurudzi vanobva kuyunivhesiti rinotungamirirwa naFrancisco J. Valero-Cuevas, purofesa we biomedical engineering, biokinesiology uye kurapa kwepanyama. Kubva pane algorithm yakagadzirwa neboka, iyo inobva pane dzimwe nzira dzekushanda kwezvisikwa zvipenyu, kutevedzana kwezviito zvehungwaru zvakagadzirwa kudzidzisa kufamba kwerobhoti pamakumbo mana. Zvinoshumwa kuti makumbo emakumbo mumhando yekutevedzera matendon, tsandanyama nemapfupa zvakakwanisa kudzidza kufamba mukati memaminitsi mashanu mushure mekuita algorithm.
Mushure mekutanga kuvhurwa, maitiro acho aive asina kurongeka uye ane mhirizhonga, asi ipapo iyo AI yakatanga kukurumidza kuchinjika kune izvo chaizvo uye zvakabudirira kutanga kufamba pasina yekutanga chirongwa. Mune ramangwana, iyo yakasikwa nzira yehupenyu hwese kudzidziswa kwemarobhoti pasina yekutanga ML kudzidziswa nemaseti edatha inogona kugadziridzwa kushongedza mota dzevagari vemo neautopilots uye yemauto marobhoti mota. Nekudaro, tekinoroji iyi ine tarisiro yakawanda uye nzvimbo dzekushandisa. Chinhu chikuru ndechekuti algorithm haioni munhu seimwe yezvipingamupinyi mukusimudzira uye haadzidzi chero chakaipa.
Source: 3dnews.ru