Matches in SemOpenAlex for { <https://semopenalex.org/work/W4323641515> ?p ?o ?g. }
Showing items 1 to 77 of
77
with 100 items per page.
- W4323641515 abstract "Reparameterization aims to improve the generalization of deep neural networks by transforming convolutional layers into equivalent multi-branched structures during training. However, there exists a gap in understanding how reparameterization may change and benefit the learning process of neural networks. In this paper, we present a novel spatial gradient scaling method to redistribute learning focus among weights in convolutional networks. We prove that spatial gradient scaling achieves the same learning dynamics as a branched reparameterization yet without introducing structural changes into the network. We further propose an analytical approach that dynamically learns scalings for each convolutional layer based on the spatial characteristics of its input feature map gauged by mutual information. Experiments on CIFAR-10, CIFAR-100, and ImageNet show that without searching for reparameterized structures, our proposed scaling method outperforms the state-of-the-art reparameterization strategies at a lower computational cost." @default.
- W4323641515 created "2023-03-09" @default.
- W4323641515 creator A5006778208 @default.
- W4323641515 creator A5010755344 @default.
- W4323641515 creator A5012650575 @default.
- W4323641515 creator A5028469274 @default.
- W4323641515 creator A5032424832 @default.
- W4323641515 creator A5075850461 @default.
- W4323641515 creator A5081633198 @default.
- W4323641515 date "2023-03-05" @default.
- W4323641515 modified "2023-10-16" @default.
- W4323641515 title "Reparameterization through Spatial Gradient Scaling" @default.
- W4323641515 doi "https://doi.org/10.48550/arxiv.2303.02733" @default.
- W4323641515 hasPublicationYear "2023" @default.
- W4323641515 type Work @default.
- W4323641515 citedByCount "0" @default.
- W4323641515 crossrefType "posted-content" @default.
- W4323641515 hasAuthorship W4323641515A5006778208 @default.
- W4323641515 hasAuthorship W4323641515A5010755344 @default.
- W4323641515 hasAuthorship W4323641515A5012650575 @default.
- W4323641515 hasAuthorship W4323641515A5028469274 @default.
- W4323641515 hasAuthorship W4323641515A5032424832 @default.
- W4323641515 hasAuthorship W4323641515A5075850461 @default.
- W4323641515 hasAuthorship W4323641515A5081633198 @default.
- W4323641515 hasBestOaLocation W43236415151 @default.
- W4323641515 hasConcept C108583219 @default.
- W4323641515 hasConcept C111919701 @default.
- W4323641515 hasConcept C11413529 @default.
- W4323641515 hasConcept C120665830 @default.
- W4323641515 hasConcept C121332964 @default.
- W4323641515 hasConcept C134306372 @default.
- W4323641515 hasConcept C138885662 @default.
- W4323641515 hasConcept C154945302 @default.
- W4323641515 hasConcept C177148314 @default.
- W4323641515 hasConcept C192209626 @default.
- W4323641515 hasConcept C2524010 @default.
- W4323641515 hasConcept C2776401178 @default.
- W4323641515 hasConcept C33923547 @default.
- W4323641515 hasConcept C41008148 @default.
- W4323641515 hasConcept C41895202 @default.
- W4323641515 hasConcept C81363708 @default.
- W4323641515 hasConcept C98045186 @default.
- W4323641515 hasConcept C99844830 @default.
- W4323641515 hasConceptScore W4323641515C108583219 @default.
- W4323641515 hasConceptScore W4323641515C111919701 @default.
- W4323641515 hasConceptScore W4323641515C11413529 @default.
- W4323641515 hasConceptScore W4323641515C120665830 @default.
- W4323641515 hasConceptScore W4323641515C121332964 @default.
- W4323641515 hasConceptScore W4323641515C134306372 @default.
- W4323641515 hasConceptScore W4323641515C138885662 @default.
- W4323641515 hasConceptScore W4323641515C154945302 @default.
- W4323641515 hasConceptScore W4323641515C177148314 @default.
- W4323641515 hasConceptScore W4323641515C192209626 @default.
- W4323641515 hasConceptScore W4323641515C2524010 @default.
- W4323641515 hasConceptScore W4323641515C2776401178 @default.
- W4323641515 hasConceptScore W4323641515C33923547 @default.
- W4323641515 hasConceptScore W4323641515C41008148 @default.
- W4323641515 hasConceptScore W4323641515C41895202 @default.
- W4323641515 hasConceptScore W4323641515C81363708 @default.
- W4323641515 hasConceptScore W4323641515C98045186 @default.
- W4323641515 hasConceptScore W4323641515C99844830 @default.
- W4323641515 hasLocation W43236415151 @default.
- W4323641515 hasOpenAccess W4323641515 @default.
- W4323641515 hasPrimaryLocation W43236415151 @default.
- W4323641515 hasRelatedWork W2731899572 @default.
- W4323641515 hasRelatedWork W2999805992 @default.
- W4323641515 hasRelatedWork W3116150086 @default.
- W4323641515 hasRelatedWork W3133861977 @default.
- W4323641515 hasRelatedWork W3166467183 @default.
- W4323641515 hasRelatedWork W3192840557 @default.
- W4323641515 hasRelatedWork W4200173597 @default.
- W4323641515 hasRelatedWork W4220996320 @default.
- W4323641515 hasRelatedWork W4312417841 @default.
- W4323641515 hasRelatedWork W4321369474 @default.
- W4323641515 isParatext "false" @default.
- W4323641515 isRetracted "false" @default.
- W4323641515 workType "article" @default.