IPanasonic ithuthukisa ubuchwepheshe obukhulu be-AI ad,
Yamukelwe ku-CVPR2021,
Ingqungquthela yezobuchwepheshe ehamba phambili yezwe le-AI
[1] Isenzo sasekhaya Genome: Ukuqonda kwesenzo sokuhlanganisa
Siyajabula ukumemezela ukuthi sithuthukise i-dataset entsha "Isenzo sesenzo sasekhaya" esiqoqa imisebenzi yansuku zonke emizini yaso esebenzisa izinhlobo eziningana zezinzwa, kufaka phakathi amakhamera, imakrofoni nezinzwa ezishisayo. Sakhe sakha futhi sakhipha i-dataset enkulu kunazo zonke ye-multimodal yezikhala zokuphila, kanti iningi lemininingwane yezikhala zokuphila lincane. Ngokusebenzisa le datha, abacwaningi be-AI bangayisebenzisa njengemininingwane yokuqeqesha yokufunda komshini kanye nocwaningo lwe-AI ukusekela abantu endaweni yokuhlala.
Ngaphezu kwalokhu okungenhla, sithuthukise ubuchwepheshe bokufunda ngokubambisana kokuqashelwa komsebenzi we-Hierarchical kuma-multimodal kanye nemibono eminingi. Ngokusebenzisa le teknoloji, singafunda izici ezingaguquki phakathi kwemibono ehlukene, izinzwa, izindlela zokuziphatha ezinemibala, namalebula wokuziphatha anemininingwane, futhi ngaleyo ndlela athuthukise ukusebenza kokuqashelwa kwemisebenzi eyinkimbinkimbi ezindaweni zokuphila.
Lobu buchwepheshe bungumphumela wocwaningo olwenziwe ngokubambisana phakathi kwedijithali AI Technology Center, ukuhlukaniswa kobuchwepheshe, kanye nombono weStanford kanye neLebhu yokufunda eStanford University.
Umdwebo1: Ukubambisana kwezenzo zokuhlanganisa ukuqonda (CCAU) Ukuqeqesha ukusebenzisana wonke ama-modalies ndawonye kusivumela ukuthi sibone ukusebenza okuthuthukile.
Sisebenzisa ukuqeqeshwa kusetshenziswa amalebula womabili we-video kanye ne-atomic isenzo ukuvumela amavidiyo nezenzo ze-athomu ukuze zizuze ekuxhumaneni okwenziwe phakathi kwalokhu okubili.
[2] I-AutoDo: I-Autougnment enobuqili yedatha enobandlululo ngomsindo welebula nge-slabikeal probubililistic
Siyajabula futhi ukumemezela ukuthi sithuthukise ubuchwepheshe obusha bokufunda umshini owenza ngokuzenzakalelayo ukwengezwa kwedatha efanelekile ngokuya ngokusatshalaliswa kwemininingwane yokuqeqeshwa. Lobu buchwepheshe bungasetshenziswa ezimweni zomhlaba wangempela, lapho idatha etholakalayo incane kakhulu. Kunezimo eziningi ezindaweni zethu eziphambili zebhizinisi, lapho kunzima ukusebenzisa ubuchwepheshe be-AI ngenxa yemikhawulo yedatha etholakalayo. Ngokusebenzisa le teknoloji, inqubo yokuhleleka kwamapharamitha we-agcation data angaqedwa, futhi amapharamitha angalungiswa ngokuzenzakalelayo. Ngakho-ke, kungalindeleka ukuthi uhla lwezinhlelo zokusebenza zobuchwepheshe be-AI lungasakazeka kabanzi. Esikhathini esizayo, ngokusheshisa ucwaningo kanye nokuthuthukiswa kwalobu buchwepheshe, sizosebenzela ukuthola ubuchwepheshe be-AI obungasetshenziswa ezindaweni zangempela zomhlaba ezifana namadivayisi ajwayelekile nezinhlelo. Lobu buchwepheshe bungumphumela wocwaningo olwenziwe yisikhungo sobuchwepheshe be-Digital AI, ukuhlukaniswa kobuchwepheshe, ilabhoreji ye-AI R & D inkampani yaseMelika.
Umdwebo 2: I-AutoDo ixazulula inkinga yokwenyuselwa idatha (Inqubomgomo eyabelwe inqubomgomo ye-DA)
"2" Ungaphansi kweminyaka yobudala, ngenkathi u- "5" afakwe ngokweqile. Ngenxa yalokhu, izindlela zangaphambilini azikwazi ukufanisa ukusatshalaliswa kokuhlolwa kanye nesinqumo se-Classifier F (θ) ayilungile.
Imininingwane yalobu buchwepheshe izokwethulwa ku-CVPR2021 (okuzoba khona kusukela ngoJuni 19, 2017).
Umyalezo ongenhla uqhamuka kuwebhusayithi yePanasonic esemthethweni!
Isikhathi Sokuthumela: Jun-03-2021
