Panasonic Imapanga Matekinoloje Awiri Apamwamba AI,
Kuvomerezedwa ku CVPR2021,
Msonkhano Wapadziko Lonse Wotsogola Wadziko Lonse wa AI Technology
[1] Home Action Genome: Kumvetsetsa Zochita Zotsutsana
Ndife okondwa kulengeza kuti tapanga chida chatsopano cha "Home Action Genome" chomwe chimasonkhanitsa zochitika za tsiku ndi tsiku za anthu m'nyumba zawo pogwiritsa ntchito mitundu ingapo ya masensa, kuphatikiza makamera, maikolofoni ndi masensa otentha. Tapanga ndi kutulutsa gulu lalikulu kwambiri padziko lonse lapansi lamitundu yosiyanasiyana yamalo okhala, pomwe ma data ambiri a malo okhala akhala ang'onoang'ono. Pogwiritsa ntchito deta iyi, ofufuza a AI atha kuzigwiritsa ntchito ngati maphunziro ophunzirira makina ndi kafukufuku wa AI kuthandiza anthu okhala m'malo okhala.
Kuphatikiza pa zomwe tafotokozazi, tapanga ukadaulo wophunzirira wothandizana nawo kuti uzindikire zochitika zaulamuliro mumitundu yambiri komanso malingaliro angapo. Pogwiritsa ntchito ukadaulo uwu, titha kuphunzira zinthu zofananira pakati pa malingaliro osiyanasiyana, masensa, machitidwe otsogola, ndi zilembo zatsatanetsatane zamakhalidwe, motero kupititsa patsogolo kuzindikira kwa zochitika zovuta m'malo okhala.
Ukadaulo uwu ndi zotsatira za kafukufuku wopangidwa mogwirizana pakati pa Digital AI Technology Center, Technology Division, ndi Stanford Vision and Learning Lab ku yunivesite ya Stanford.
Chithunzi 1: Cooperative Compositional Action Understanding (CCAU) Kuphunzitsa mogwira mtima njira zonse pamodzi kumatithandiza kuwona bwino ntchito.
Timagwiritsa ntchito maphunziro pogwiritsa ntchito zilembo zamakanema komanso ma atomiki kuti mavidiyo ndi zochita za atomiki zipindule ndi kuyanjana kwapakati pa ziwirizi.
[2] AutoDO: Robust AutoAugment for Biased Data yokhala ndi Label Noise kudzera pa Scalable Probabilistic Implicit Differentiation
Ndifenso okondwa kulengeza kuti tapanga ukadaulo watsopano wophunzirira makina omwe amangopanga kukulitsa kwa data moyenera malinga ndi kugawa kwa data yophunzitsira. Tekinolojeyi ingagwiritsidwe ntchito ku zochitika zenizeni zapadziko lapansi, kumene deta yomwe ilipo ndi yochepa kwambiri. Pali zochitika zambiri m'madera athu akuluakulu amalonda, komwe kumakhala kovuta kugwiritsa ntchito teknoloji ya AI chifukwa cha kuchepa kwa deta yomwe ilipo. Pogwiritsa ntchito ukadaulo uwu, njira yosinthira magawo owonjezera a data imatha kuthetsedwa, ndipo magawowo amatha kusinthidwa zokha. Chifukwa chake, zitha kuyembekezera kuti mitundu yogwiritsira ntchito ukadaulo wa AI ikhoza kufalikira kwambiri. M'tsogolomu, popititsa patsogolo kafukufuku ndi chitukuko cha teknolojiyi, tidzayesetsa kuzindikira luso la AI lomwe lingagwiritsidwe ntchito m'madera enieni monga zipangizo ndi machitidwe omwe amadziwika bwino. Ukadaulo uwu ndi zotsatira za kafukufuku wopangidwa ndi Digital AI Technology Center, Technology Division, AI Laboratory ya Panasonic R&D Company of America.
Chithunzi 2: AutoDO imathetsa vuto la kuwonjezereka kwa deta (Shared-policy DA dilemma) .kugawa kwa deta yowonjezereka ya sitima (yopanda buluu) sikungafanane ndi deta yoyesera (yofiira yolimba) mu malo obisika:
"2" ndiyowonjezereka, pamene "5" ndiyowonjezereka. Zotsatira zake, njira zam'mbuyomu sizingafanane ndi kugawa kwa mayeso ndipo lingaliro la wophunzira wophunzira f(θ) ndilolakwika.
Tsatanetsatane wa matekinolojewa adzaperekedwa ku CVPR2021 (kuti ichitike kuyambira June 19th, 2017).
Pamwambapa uthenga ukuchokera patsamba lovomerezeka la Panasonic!
Nthawi yotumiza: Jun-03-2021
