Reseau capsule
Reseau capsule yi architecture neuronal lañu yuy boole neuron yi ci ay 'capsule' yuy génne ay vecteur yuy kode ndax amna màndarga ak pose bi (position, jubluwaay, echel).
Résumé
They aim to fix a core blindness in standard convolutional networks: losing track of spatial relationships between parts.
Plongeur bu xóot
Geoffrey Hinton, Sara Sabour ak Nicholas Frosst ñoo ko taxawal ci 2017, reso kapsul yi dañuy wecci génnug neuron scalaire ak vecteur. Guddaayi fi vecteur bi dafay wane probabilite bi am ci entite (lu melni bët wala bakkan), fekk jubluwaayam dafay kode pose parametre yi. Niveau capsules yu gëna suufe yi ñooy wax luy waaja am ci capsules yu gëna kawe yi jaaraleko ci matrices yu ñuy soppi, ak anam wu ñuy woowe dynamic routing-by-agreement mooy xam ban xam-xam lañu wara wóolu. Sudee pàrti kapsul yu bari bokk nañu ci benn lëmm, yoon wi dafay gëna dëgëral lëkkaloo bi. CapsNet bu njëkk bi amna ay resilta yu am solo ci MNIST te nekkoon na lu am doole ci lim yuy jaxasoo ak coppite affine, di saafara 'jafe-jafe Picasso' fu CNN yi nangu melokaani kanam yu jaxasoo ni kanam bu baax.
Gis-gis xarala
Mekanism bi gëna am solo mooy 'squash' nonlinearite biy wàññi vecteur yu gàtt yi dem ci nul ak vecteur yu gudd yi ci guddaay benn, kon magnitude vecteur bi dafay nekk probabilite. Routing Dynamic dafay def ay iteration yu néew ci jéego déggoob softmax-pondéré: capsule bu nekk ci suuf dafay yónnee liñu ko wax, ba noppi coefficients couplage yi dañuy yokk ngir capsule yu gëna kawe yi seen génnekaay méngoo (jaaraleko ci produit dot) ak li ñu wax. Loolu mooy wecci max-pooling, di denc leeral yi gëna jubal ci barab bi, duñu ko sànni.
njeextalu pexe
Njëgg ak budget
Dogal yi architecture di jël dañuy indi njariñ ak njëgu liggéey bi ay at ci ginaaw.
dogal yu gëna leer
Njàngalem xarala yi dafay jàppale ekip yi ñu tànn li gën, te baña yam ci li gëna bees daal.
Xool kalite
Tanneef yu gëna baax ci wàllu ingeñër dina wàññi jafe-jafe yi ci wàllu wóor ci liggéey bi.
Ëlëgu Reseau Capsule
Reseau capsule yi dañuy gëna nekk yoon wi ñuy gëstu, moo gëna bari ndax routing dynamique dafa seer lool ci wàllu ordinatër te du mëna dem ba ci nataal yu yaatu yu melni ImageNet. Liggéey bi ci topp dafa xoolaat yoon EM (Matrix Capsules) ak yoon wi lalu ci bàyyi xel ci sa bopp ngir gëna mëna liggéey. Ginaaw bëgg-bëggu equivariance, efficacité misaal, ak hierarchies yuñ mëna tekki dañuy gëna màgg, xalaati capsule yi dañuy wéy di am njeexital ci gëstu yi, boole ci xalaatu GLOM bu Hinton ci ginaaw, doonte Transformers ñoo ëpp doole ci gis-gis bu mag bi.
Doxal ci àdduna dëgg
Xaajale nimero yiñ bind ak loxo ci MNIST bi ñuy tabaxaat liñu dugal ci vecteur capsule yi, di wane ni parametru pose yi dañu am solo.
Tàqale ñaari lim yuy jaxasoo (liggéey MultiMNIST) ci xaaj pixel yi bokk ci ban entite.
Gëstu ci nataali medsin di jëfandikoo ay kapsul ngir gis ay nodule ci poumon yi wala ay tumër ci yuur gi, muy barab bu am solo ci diggante wàll wi ak barab bi yépp.
Xamme mbir yi ci anamu gis-gis bu bees ak misaali tàggat yu néew, jàppale architecture bi ci biir gis-gis buy méngoo.
Risk yi ak balustrade yi
Optimize benn benchmark mën na nëbb ñakk kattan yu gëna yaatu ci sistem bi.
Njëg li ñuy fay ci infrastructure yi ak ci toppatoo dañuy faral di suufeel.
Bu sistem yi di gëna xawa jafee xam, jafe-jafe yi am ci wàllu kaaraange ak seetlu mën nañu gëna bari.
Roadmap ngir samp gi
Mandargal latency, kalite, ak njëg yi laata ngay jëfandikoo.
Benchmark ci biir sargal ak done yu dëggu.
Jumtukaay bi di saytu njuumte yi, derive bi ak njeextalu jëfandikukat bi.
Waajal rollback ak yooni tontu ci jafe-jafe yi laata ngay eskale.
Weyal di banneexu
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Capsule Networks quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Gis bi ci topp
Reseau autoroute yi ak lëkkaloo yi
Laaj yi ñuy faral di laaj
What is Capsule Networks?
Reseau capsule yi architecture neuronal lañu yuy boole neuron yi ci ay 'capsule' yuy génne ay vecteur yuy kode ndax amna màndarga ak pose bi (position, jubluwaay, echel). Liñu bëgga mooy saafara jafe-jafe yi ci reso convolutionnel yiñ miin: ñàkka mëna topp lëkkaloo gi am ci diggante pàcc yi.
Ci biir reso capsule, lan mooy LENGTH vecteur biy génne ci capsule bi?
Guddaayi vecteur bi (magnitude) dafay kode probabilite bi entite bi mëna am, fekk jubluwaayam dafay kode pose parametre yu melni position ak wëréelu.
Ban jafe-jafe ak CNN yiñ miin lañu defar ngir saafara reso capsule yi?
CNNs ak max-pooling sànni relasioŋ spatial yu leer, kon kanam bu am màndarga yu jaarul yoon mën na am poñ yu bari. Lii mooy 'jafe-jafe Picasso' bi kapsul yi di jéema saafara.
Luy turu algorithm biy wax ban kapsul bu gëna suufe mooy boole ak ban kapsul bu gëna kawe?
Dynamic routing-by-agreement dafay gëna dëgëral lëkkaloo gi am ci digganté capsules yi nga xamni li ñuy seentu méngoo ak génnug capsule bu gëna rëy.
Kan moo dugal reso capsule bi am routage dynamique ci 2017?
Sara Sabour, Nicholas Frosst ak Geoffrey Hinton ñoo bind këyitu 2017 bi tuddu 'Yoon buy Soppeeku ci Diggante Kapsul yi'.
Lan mooy njariñu fonction 'squash' ci reso capsule?
Nonlinearite squash dafay wàññi vecteur yu gàtt yi ci nul ba noppi tëj vecteur yu gudd yi ci wetu benn guddaay, suko defee magnitude bi mën nañu ko jàng ni probabilite fekk jubluwaay (pose) bi dafay wéy.