Panasonic Inovandudza Maviri Epamusoro AI Technologies,
Yakagamuchirwa kuCVPR2021,
Musangano wePasirose Unotungamira pasi rose weAI Technology
[1] Home Action Genome: Contrastive Compositional Action Kunzwisisa
Isu tinofara kuzivisa kuti takagadzira dhata nyowani "Home Action Genome" inounganidza zviitiko zvevanhu zvezuva nezuva mudzimba dzavo vachishandisa akati wandei masensor, anosanganisira makamera, maikorofoni uye zvinopisa zvinonzwa. Takavaka uye kuburitsa dhatabheti rakakura kwazvo repasi rose renzvimbo dzekugara, nepo dhatabheti zhinji dzenzvimbo dzekugara dzave diki pachiyero. Nekushandisa iyi dhatabheti, vaongorori veAI vanogona kuishandisa sedhata rekudzidzisa rekudzidza kwemichina uye kutsvaga kweAI kutsigira vanhu munzvimbo yekugara.
Pamusoro pezviri pamusoro, takagadzira tekinoroji yekudyidzana yekudzidza yekuzivikanwa kwechiitwa chiitiko mune akawanda uye akawanda maonero. Nekushandisa tekinoroji iyi, tinogona kudzidza maficha anowirirana pakati pemaonero akasiyana, masensa, hunhu hwepamusoro, uye akadzama maitiro ezvinyorwa, uye nekudaro tinovandudza kucherechedzwa kwekuita kwezviitiko zvakaoma munzvimbo dzekugara.
Iyi tekinoroji mhedzisiro yetsvakiridzo yakaitwa mukubatana pakati peDigital AI Technology Center, Technology Division, neStanford Vision uye Kudzidza Lab paStanford University.
Mufananidzo1: Cooperative Compositional Action Understanding (CCAU)Kushanda pamwe chete kudzidzisa maitiro ese pamwe chete kunotibvumira kuona kuvandudzwa kwekuita.
Isu tinoshandisa kudzidziswa tichishandisa zvese zviri zviviri nhanho-yepamusoro uye maatomu ekuita mavara kutendera zvese mavhidhiyo neatomu zviito kuti zvibatsirwe kubva mukubatanidzwa kwezviviri.
[2] AutoDO: Robust AutoAugment yeBiased Data ine Label Noise kuburikidza neScalable Probabilistic Implicit Differentiation.
Isu tinofarawo kuzivisa kuti takagadzira tekinoroji nyowani yekudzidza muchina iyo inongoita yakakwana yekuwedzera data maererano nekugoverwa kwedhata rekudzidziswa. Iyi tekinoroji inogona kushandiswa kumamiriro epasirese chaiwo, uko data iripo idiki kwazvo. Kune zviitiko zvakawanda munzvimbo dzedu huru dzebhizinesi, uko zvakaoma kushandisa AI tekinoroji nekuda kwekugumira kweiyo data iripo. Nekushandisa iyi tekinoroji, iyo tuning process ye data augmentation paramita inogona kubviswa, uye ma paramita anogona kugadziridzwa otomatiki. Naizvozvo, zvinogona kutarisirwa kuti huwandu hwekushandisa kweAI tekinoroji inogona kupararira zvakanyanya. Mune ramangwana, nekuenderera mberi nekumhanyisa kutsvagisa nekuvandudza tekinoroji iyi, tichashanda kuti tizive tekinoroji yeAI inogona kushandiswa munzvimbo chaidzo dzepasirese senge zvishandiso zvakajairika uye masisitimu. Iyi tekinoroji mhedzisiro yekutsvagisa kwakaitwa neDigital AI Technology Center, Technology Division, AI Laboratory yePanasonic R&D Kambani yeAmerica.
Mufananidzo 2: AutoDO inogadzirisa dambudziko rekuwedzera kwedata (Shared-policy DA dilemma) .kugoverwa kwe data yakawedzerwa yechitima (dashed blue) inogona kunge isingafananidzi data yekuedza (yakasimba tsvuku) munzvimbo yakadzikama:
"2" iri pasi-yakawedzerwa, nepo "5" yakawedzera. Nekuda kweizvozvo, nzira dzekare hadzigone kuenderana nekugovera bvunzo uye sarudzo yeakadzidza classifier f(θ) haina kururama.
Iwo ruzivo rwetekinoroji iyi ichaunzwa kuCVPR2021 (ichaitwa kubva munaJune 19th, 2017).
Pamusoro meseji inobva kuPanasonic yepamutemo webhusaiti!
Nguva yekutumira: Jun-03-2021
