FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3034632403> ?p ?o ?g. }

• W3034632403 abstract "Modern neural network performance typically improves as model size increases. A recent line of research on the Neural Tangent Kernel (NTK) of over-parameterized networks indicates that the improvement with size increase is a product of a better conditioned loss landscape. In this work, we investigate a form of over-parameterization achieved through ensembling, where we define collegial ensembles (CE) as the aggregation of multiple independent models with identical architectures, trained as a single model. We show that the optimization dynamics of CE simplify dramatically when the number of models in the ensemble is large, resembling the dynamics of wide models, yet scale much more favorably. We use recent theoretical results on the finite width corrections of the NTK to perform efficient architecture search in a space of finite width CE that aims to either minimize capacity, or maximize trainability under a set of constraints. The resulting ensembles can be efficiently implemented in practical architectures using group convolutions and block diagonal layers. Finally, we show how our framework can be used to analytically derive optimal group convolution modules originally found using expensive grid searches, without having to train a single model." @default.
• W3034632403 created "2020-06-19" @default.
• W3034632403 creator A5014759491 @default.
• W3034632403 creator A5031329242 @default.
• W3034632403 creator A5033404184 @default.
• W3034632403 creator A5049929304 @default.
• W3034632403 creator A5082311640 @default.
• W3034632403 creator A5091159803 @default.
• W3034632403 date "2020-06-13" @default.
• W3034632403 modified "2023-09-27" @default.
• W3034632403 title "Collegial Ensembles." @default.
• W3034632403 cites W2163605009 @default.
• W3034632403 cites W2549139847 @default.
• W3034632403 cites W2793904650 @default.
• W3034632403 cites W2809090039 @default.
• W3034632403 cites W2889798024 @default.
• W3034632403 cites W2950220847 @default.
• W3034632403 cites W2963125010 @default.
• W3034632403 cites W2963739978 @default.
• W3034632403 cites W2963966020 @default.
• W3034632403 cites W2963993763 @default.
• W3034632403 cites W2965862350 @default.
• W3034632403 cites W2981615156 @default.
• W3034632403 cites W2987473824 @default.
• W3034632403 cites W3118608800 @default.
• W3034632403 hasPublicationYear "2020" @default.
• W3034632403 type Work @default.
• W3034632403 sameAs 3034632403 @default.
• W3034632403 citedByCount "2" @default.
• W3034632403 countsByYear W30346324032021 @default.
• W3034632403 crossrefType "posted-content" @default.
• W3034632403 hasAuthorship W3034632403A5014759491 @default.
• W3034632403 hasAuthorship W3034632403A5031329242 @default.
• W3034632403 hasAuthorship W3034632403A5033404184 @default.
• W3034632403 hasAuthorship W3034632403A5049929304 @default.
• W3034632403 hasAuthorship W3034632403A5082311640 @default.
• W3034632403 hasAuthorship W3034632403A5091159803 @default.
• W3034632403 hasConcept C11413529 @default.
• W3034632403 hasConcept C118615104 @default.
• W3034632403 hasConcept C126255220 @default.
• W3034632403 hasConcept C130367717 @default.
• W3034632403 hasConcept C138187205 @default.
• W3034632403 hasConcept C154945302 @default.
• W3034632403 hasConcept C165464430 @default.
• W3034632403 hasConcept C173608175 @default.
• W3034632403 hasConcept C177264268 @default.
• W3034632403 hasConcept C187691185 @default.
• W3034632403 hasConcept C199360897 @default.
• W3034632403 hasConcept C2524010 @default.
• W3034632403 hasConcept C2777210771 @default.
• W3034632403 hasConcept C33923547 @default.
• W3034632403 hasConcept C41008148 @default.
• W3034632403 hasConcept C45347329 @default.
• W3034632403 hasConcept C50644808 @default.
• W3034632403 hasConcept C68339613 @default.
• W3034632403 hasConcept C74193536 @default.
• W3034632403 hasConceptScore W3034632403C11413529 @default.
• W3034632403 hasConceptScore W3034632403C118615104 @default.
• W3034632403 hasConceptScore W3034632403C126255220 @default.
• W3034632403 hasConceptScore W3034632403C130367717 @default.
• W3034632403 hasConceptScore W3034632403C138187205 @default.
• W3034632403 hasConceptScore W3034632403C154945302 @default.
• W3034632403 hasConceptScore W3034632403C165464430 @default.
• W3034632403 hasConceptScore W3034632403C173608175 @default.
• W3034632403 hasConceptScore W3034632403C177264268 @default.
• W3034632403 hasConceptScore W3034632403C187691185 @default.
• W3034632403 hasConceptScore W3034632403C199360897 @default.
• W3034632403 hasConceptScore W3034632403C2524010 @default.
• W3034632403 hasConceptScore W3034632403C2777210771 @default.
• W3034632403 hasConceptScore W3034632403C33923547 @default.
• W3034632403 hasConceptScore W3034632403C41008148 @default.
• W3034632403 hasConceptScore W3034632403C45347329 @default.
• W3034632403 hasConceptScore W3034632403C50644808 @default.
• W3034632403 hasConceptScore W3034632403C68339613 @default.
• W3034632403 hasConceptScore W3034632403C74193536 @default.
• W3034632403 hasLocation W30346324031 @default.
• W3034632403 hasOpenAccess W3034632403 @default.
• W3034632403 hasPrimaryLocation W30346324031 @default.
• W3034632403 hasRelatedWork W2151380595 @default.
• W3034632403 hasRelatedWork W2540616960 @default.
• W3034632403 hasRelatedWork W2748902989 @default.
• W3034632403 hasRelatedWork W2785606994 @default.
• W3034632403 hasRelatedWork W2883816343 @default.
• W3034632403 hasRelatedWork W2949067459 @default.
• W3034632403 hasRelatedWork W2951568266 @default.
• W3034632403 hasRelatedWork W2963844898 @default.
• W3034632403 hasRelatedWork W3011193154 @default.
• W3034632403 hasRelatedWork W3037850847 @default.
• W3034632403 hasRelatedWork W3039116410 @default.
• W3034632403 hasRelatedWork W3040010236 @default.
• W3034632403 hasRelatedWork W3046890301 @default.
• W3034632403 hasRelatedWork W3090123627 @default.
• W3034632403 hasRelatedWork W3105037848 @default.
• W3034632403 hasRelatedWork W3177963998 @default.
• W3034632403 hasRelatedWork W3185945913 @default.
• W3034632403 hasRelatedWork W3206378469 @default.
• W3034632403 hasRelatedWork W3214411122 @default.
• W3034632403 hasRelatedWork W2898071315 @default.
• W3034632403 isParatext "false" @default.
• W3034632403 isRetracted "false" @default.

• next