Chikwata chevaongorori kubva kuYunivhesiti yePadua (Italy) neYunivhesiti yeDelft (Netherlands) vakaburitsa nzira yekushandisa muchina kudzidza kugadzirazve yakapinda PIN kodhi kubva kuvhidhiyo kurekodha yenzvimbo yakavharwa nemaoko yeATM. . Paunenge uchipinda 4-digit PIN kodhi, mukana wekufanotaura kodhi chaiyo inofungidzirwa pa41%, uchifunga nezve mukana wekuedza katatu usati wavhara. Kune mashanu-manhamba ePIN kodhi, mukana wekufungidzira waiva 5%. Kuedza kwakasiyana kwakaitwa umo vazvipiri makumi manomwe neshanu vakaedza kufanotaura PIN kodhi kubva kumavhidhiyo akarekodhwa akafanana. Muchiitiko ichi, mukana wekufanotaura kwakabudirira waiva 30% mushure mekuedza katatu.
Paunenge uchivhara dhijitari pani yeATM nechanza chako, chikamu cheruoko icho chinoiswa chinoramba chisina kuvharwa, izvo zvakakwana kufanotaura kudzvanya nekushandura chinzvimbo cheruoko uye kushandura iyo isina kuvharwa zvigunwe. Paunenge uchiongorora mapindiro edijiti rega rega, iyo sisitimu inobvisa makiyi asingagone kudzvanywa uchifunga nezve chinzvimbo cheruoko rwekuvhara, uye zvakare inoverengera zvingangoitika sarudzo dzekutsikirira zvichienderana nenzvimbo yeruoko rwekudzvanya maererano nenzvimbo yemakiyi. . Kuti uwedzere mukana wekuonekwa kwekupinza, ruzha rwemakiyi anogona kuwedzera kurekodhwa, izvo zvakasiyana zvishoma kune imwe neimwe kiyi.
Chiyedzo chakashandisa muchina wekufunda sisitimu yakavakirwa pakushandiswa kweiyo convolutional neural network (CNN) uye inodzokororwa neural network yakavakirwa paLSTM (Nguva Yenguva Yakareba Memory) yekuvaka. Iyo CNN network yaive nebasa rekuburitsa dhata renzvimbo kune yega yega furemu, uye network yeLSTM yakashandisa iyi data kuburitsa nguva-inosiyana mapatani. Iyo modhi yakadzidziswa pamavhidhiyo e58 vanhu vakasiyana vanopinda maPIN kodhi vachishandisa vatori vechikamu-vakasarudzwa nzira dzekuvhara yekupinda (mutori wega wega akapinda zana akasiyana macode, i.e., 100 yekuisa mienzaniso yakashandiswa pakudzidziswa). Munguva yekudzidziswa, zvakaratidzwa kuti vashandisi vazhinji vanoshandisa imwe yenzira nhatu huru dzekuvhara mapindiro.
Kudzidzisa muchina wekudzidza modhi, sevha yakavakirwa paXeon E5-2670 processor ine 128 GB ye RAM uye matatu Tesla K20m makadhi ane 5GB yendangariro imwe neimwe yakashandiswa. Chikamu chesoftware chakanyorwa muPython uchishandisa raibhurari yeKeras uye Tensorflow chikuva. Sezvo mapaneru ekuisa eATM akasiyana uye mhedzisiro yekufanotaura zvinoenderana nehunhu hwakaita sekiyi saizi uye topology, kudzidziswa kwakasiyana kunodiwa kune yega yega yepaneru.
Sematanho ekudzivirira kubva kune yakarongwa nzira yekurwisa, zvinokurudzirwa, kana zvichibvira, kushandisa maPIN kodhi e5 manhamba panzvimbo ye4, uye zvakare edza kuvhara yakawanda yenzvimbo yekupinza sezvinobvira neruoko rwako (iyo nzira inoramba ichishanda kana inenge 75% yenzvimbo yekupinza yakafukidzwa neruoko rwako). Vagadziri veATM vanokurudzirwa kushandisa zvakakosha zvidziviriro zvinovanza kupinza, uye kwete zvemakanika, asi kubata mapenera ekuisa, chinzvimbo chenhamba dzinochinja zvisina tsarukano.
Source: opennet.ru