UKavya Kopparapu, okwishumi elivisayo uhlala eVirginia, wasebenzisa i-smartphone, umshicileli we-3D, kunye nobukrelekrele bokuzenzela ukuze phuhlisa i-eyeagnosis. un Inkqubo ephathekayo yokuqonda isifo kwaye kulula ukufumanisa isifo Ukujonga kwakhona isifo seswekile. Esi sifo sisifo seswekile yonakalisa imithambo yegazi kwiretina yomntu kwaye kunokukhokelela kukungaboni kakuhle.
UKopparapu unomdla kwisayensi bonke ubomi bakhe kwaye emva kokuya kwindibano yocweyo ehlelwe liZiko leSizwe labaseTyhini kunye neTekhnoloji yoLwazi, wongeze iinkqubo kwizinto azithandayo.
Utat'omkhulu kaKopparapu, ohlala eIndiya, yaqala ukubonisa iimpawu kwesi sifo kwi 2013. Ingasoloko ingabonwa, kwaye nangona yafunyanwa ichongiwe kwaye inyangwa, umbono wakhe wonakala. NgokukaKopparapu, ngaphandle kwe Izigidi ezingama-415 zabantu abanesifo seswekile, Isinye kwisithathu siza kuphuhlisa ukubonwa kwakhona kweswekile, kwaye nangona amayeza kunye notyando lunokuyeka okanye lutshintshe ukonakala kwamehlo ukuba lubanjiwe ngokuhamba kwexesha, I-50% ayizukuchongwa, isiqingatha sezigulane ezineefom ezibi kakhulu siya kuba yimfama kwiminyaka emihlanu.
“Ukungabikho koxilongo yeyona ngxaki inkulu. E-India, kukho iinkqubo ezithumela oogqirha ezilalini nasematyotyombeni, kodwa kukho izigulana ezininzi kwaye zininzi kuphela zamehlo.
Wayezibuza ukuba ingaba ikhona na indlela elula nexabiso eliphantsi yokufumanisa isifo, kwaye kwavela uluvo lovavanyo lwe-Eyagnosis, inkqubo enokuthi iguqule inkqubo ende kunye nexabisayo yokufumanisa isifo kwiseshoni elula yokufota. UKopparapu usebenze, echitha ixesha elininzi kuGoogle kwaye ebhalela oogqirha nabaphandi nge-imeyile, ngaphambi kokwenza isicwangciso. Wadibana nomntakwabo kunye nomfundi afunda naye, kwaye Usebenzise uthungelwano lwe-neural network (CNN) ukuseka isifo se-AI emva kwe-Eyeagnosis. Iinethiwekhi zangaphakathi Hlalutya iiseti zedatha enkulu kwaye ujonge iipateni ezifanayo, Kuba uyilo lufana nenkqubo ebonakalayo yengqondo yomntu, ii-CNN zibalasele kudidi.
Isebenzise i-ResNet-50, i-CNN ephuhliswe ngabaphandi beMicrosoft, ukwakha inethiwekhi, kunye nokusebenzisa Izikena ze-retina ezingama-34.000 ifunyenwe kwi database I-EyeGene kwiZiko lezeMpilo leSizwe eMelika (NIH) njengedatha yokufunda, ke yena neqela lakhe banokufundisa inkqubo ye-AI ukuqaphela iimpawu zesifo kwiifoto zamehlo kwaye banike isifo kwangaphambili. Uninzi lweemifanekiso kwindawo yogcino lwedatha zazivezwe kakubi okanye zimfiliba, kodwa ngokukaKopparapu, le nkcukacha incede ukuphucula inkqubo.
«Umgangatho wemifanekiso esetyenzisiweyo ekufundeni uthungelwano lwe-neural lubonisa kakhulu iimeko ezinokufunyanwa ngokusetyenziswa kwe-smartphone«
Iqela lakhe liqeqeshe i-ResNet-50 ukuya ukubona isifo seswekile ngokuchanekileyo njengesifo sokwenyani. Ikwafumanisa kwakhona ii-microaneurysms kunye nemithambo yegazi kumfanekiso ngamnye ngaphandle kwesidingo sokufaka idayi ye-fluorescent kwiliso elichongiweyo.
Ukuwa kokugqibela, iliso leAditya Jyot Isibhedlele eMumbai sivumile ukuvavanya i-Eyeagnosis app, kwaye ngo-Novemba, ithumele i-3D yokuqala eprintiweyo esibhedlele, kwaye inkqubo sele i-hyenza uxilongo oluchanekileyo kwizigulana ezintlanu.
I-Eyeagnosis inendlela ende ekufanele ukuba ihambe kuyo ekuchongeni inani elikhulu lamatyala ukubonisa ukuba yinkqubo ethembekileyo. Iinkqubo zokuqinisekisa ukuba nayiphi na iprojekthi enxulumene neyeza kufuneka ilandele ngokungqongqo kwaye ngokuqinisekileyo kuya kuba nzima kuwe ukufumana inkampani enkulu efuna ukukunceda. Kodwa bonke obu bunzima abuphazamisi kwimpumelelo enkulu yolu lutsha.