Matches in SemOpenAlex for { <https://semopenalex.org/work/W3119009163> ?p ?o ?g. }
Showing items 1 to 84 of
84
with 100 items per page.
- W3119009163 abstract "Deep ReLU networks trained with the square loss have been observed to perform well in classification tasks. We provide here a theoretical justification based on analysis of the associated gradient flow. We show that convergence to a solution with the absolute minimum norm is expected when normalization techniques such as Batch Normalization (BN) or Weight Normalization (WN) are used together with Weight Decay (WD). The main property of the minimizers that bounds their expected error is the norm: we prove that among all the close-to-interpolating solutions, the ones associated with smaller Frobenius norms of the unnormalized weight matrices have better margin and better bounds on the expected classification error. With BN but in the absence of WD, the dynamical system is singular. Implicit dynamical regularization -- that is zero-initial conditions biasing the dynamics towards high margin solutions -- is also possible in the no-BN and no-WD case. The theory yields several predictions, including the role of BN and weight decay, aspects of Papyan, Han and Donoho's Neural Collapse and the constraints induced by BN on the network weights." @default.
- W3119009163 created "2021-01-18" @default.
- W3119009163 creator A5001833084 @default.
- W3119009163 creator A5089682862 @default.
- W3119009163 date "2021-01-04" @default.
- W3119009163 modified "2023-09-27" @default.
- W3119009163 title "Explicit regularization and implicit bias in deep network classifiers trained with the square loss." @default.
- W3119009163 cites W2151718948 @default.
- W3119009163 cites W2904614653 @default.
- W3119009163 cites W2945175884 @default.
- W3119009163 cites W2949117887 @default.
- W3119009163 cites W2952126211 @default.
- W3119009163 cites W2963685250 @default.
- W3119009163 cites W2976713865 @default.
- W3119009163 cites W3065974826 @default.
- W3119009163 cites W3081575329 @default.
- W3119009163 cites W3112700814 @default.
- W3119009163 hasPublicationYear "2021" @default.
- W3119009163 type Work @default.
- W3119009163 sameAs 3119009163 @default.
- W3119009163 citedByCount "5" @default.
- W3119009163 countsByYear W31190091632021 @default.
- W3119009163 crossrefType "posted-content" @default.
- W3119009163 hasAuthorship W3119009163A5001833084 @default.
- W3119009163 hasAuthorship W3119009163A5089682862 @default.
- W3119009163 hasConcept C11413529 @default.
- W3119009163 hasConcept C119857082 @default.
- W3119009163 hasConcept C136886441 @default.
- W3119009163 hasConcept C144024400 @default.
- W3119009163 hasConcept C154945302 @default.
- W3119009163 hasConcept C17744445 @default.
- W3119009163 hasConcept C19165224 @default.
- W3119009163 hasConcept C191795146 @default.
- W3119009163 hasConcept C199539241 @default.
- W3119009163 hasConcept C2776135515 @default.
- W3119009163 hasConcept C28826006 @default.
- W3119009163 hasConcept C2984842247 @default.
- W3119009163 hasConcept C33923547 @default.
- W3119009163 hasConcept C41008148 @default.
- W3119009163 hasConcept C50644808 @default.
- W3119009163 hasConcept C774472 @default.
- W3119009163 hasConceptScore W3119009163C11413529 @default.
- W3119009163 hasConceptScore W3119009163C119857082 @default.
- W3119009163 hasConceptScore W3119009163C136886441 @default.
- W3119009163 hasConceptScore W3119009163C144024400 @default.
- W3119009163 hasConceptScore W3119009163C154945302 @default.
- W3119009163 hasConceptScore W3119009163C17744445 @default.
- W3119009163 hasConceptScore W3119009163C19165224 @default.
- W3119009163 hasConceptScore W3119009163C191795146 @default.
- W3119009163 hasConceptScore W3119009163C199539241 @default.
- W3119009163 hasConceptScore W3119009163C2776135515 @default.
- W3119009163 hasConceptScore W3119009163C28826006 @default.
- W3119009163 hasConceptScore W3119009163C2984842247 @default.
- W3119009163 hasConceptScore W3119009163C33923547 @default.
- W3119009163 hasConceptScore W3119009163C41008148 @default.
- W3119009163 hasConceptScore W3119009163C50644808 @default.
- W3119009163 hasConceptScore W3119009163C774472 @default.
- W3119009163 hasLocation W31190091631 @default.
- W3119009163 hasOpenAccess W3119009163 @default.
- W3119009163 hasPrimaryLocation W31190091631 @default.
- W3119009163 hasRelatedWork W2129658066 @default.
- W3119009163 hasRelatedWork W2766965791 @default.
- W3119009163 hasRelatedWork W2810862998 @default.
- W3119009163 hasRelatedWork W2889756710 @default.
- W3119009163 hasRelatedWork W2894972989 @default.
- W3119009163 hasRelatedWork W2922277331 @default.
- W3119009163 hasRelatedWork W2946840143 @default.
- W3119009163 hasRelatedWork W2950051890 @default.
- W3119009163 hasRelatedWork W2964915616 @default.
- W3119009163 hasRelatedWork W2966363981 @default.
- W3119009163 hasRelatedWork W2971648803 @default.
- W3119009163 hasRelatedWork W3004514911 @default.
- W3119009163 hasRelatedWork W3065974826 @default.
- W3119009163 hasRelatedWork W3088153346 @default.
- W3119009163 hasRelatedWork W3121312397 @default.
- W3119009163 hasRelatedWork W3157503520 @default.
- W3119009163 hasRelatedWork W3166800280 @default.
- W3119009163 hasRelatedWork W3174815695 @default.
- W3119009163 hasRelatedWork W3203654258 @default.
- W3119009163 hasRelatedWork W3213914290 @default.
- W3119009163 isParatext "false" @default.
- W3119009163 isRetracted "false" @default.
- W3119009163 magId "3119009163" @default.
- W3119009163 workType "article" @default.