Matches in SemOpenAlex for { <https://semopenalex.org/work/W2914902801> ?p ?o ?g. }
- W2914902801 abstract "Neural networks with a large number of parameters admit a mean-field description, which has recently served as a theoretical explanation for the favorable training properties of overparameterized models. In this regime, gradient descent obeys a deterministic partial differential equation (PDE) that converges to a globally optimal solution for networks with a single hidden layer under appropriate assumptions. In this work, we propose a non-local mass transport dynamics that leads to a modified PDE with the same minimizer. We implement this non-local dynamics as a stochastic neuronal birth-death process and we prove that it accelerates the rate of convergence in the mean-field limit. We subsequently realize this PDE with two classes of numerical schemes that converge to the mean-field equation, each of which can easily be implemented for neural networks with finite numbers of parameters. We illustrate our algorithms with two models to provide intuition for the mechanism through which convergence is accelerated." @default.
- W2914902801 created "2019-02-21" @default.
- W2914902801 creator A5044193587 @default.
- W2914902801 creator A5048166612 @default.
- W2914902801 creator A5071403558 @default.
- W2914902801 creator A5086430646 @default.
- W2914902801 date "2019-02-05" @default.
- W2914902801 modified "2023-09-27" @default.
- W2914902801 title "Global convergence of neuron birth-death dynamics." @default.
- W2914902801 cites W102487131 @default.
- W2914902801 cites W1480347379 @default.
- W2914902801 cites W1540706608 @default.
- W2914902801 cites W1738603105 @default.
- W2914902801 cites W2071048859 @default.
- W2914902801 cites W2095705004 @default.
- W2914902801 cites W2103496339 @default.
- W2914902801 cites W2113442785 @default.
- W2914902801 cites W2160960847 @default.
- W2914902801 cites W2162036626 @default.
- W2914902801 cites W2163605009 @default.
- W2914902801 cites W2166116275 @default.
- W2914902801 cites W2239176660 @default.
- W2914902801 cites W2402846924 @default.
- W2914902801 cites W2519388618 @default.
- W2914902801 cites W2596367596 @default.
- W2914902801 cites W2778749116 @default.
- W2914902801 cites W2789285779 @default.
- W2914902801 cites W2798986185 @default.
- W2914902801 cites W2897080984 @default.
- W2914902801 cites W2946963372 @default.
- W2914902801 cites W2962742373 @default.
- W2914902801 cites W2966530573 @default.
- W2914902801 cites W3103801447 @default.
- W2914902801 cites W2770298516 @default.
- W2914902801 hasPublicationYear "2019" @default.
- W2914902801 type Work @default.
- W2914902801 sameAs 2914902801 @default.
- W2914902801 citedByCount "12" @default.
- W2914902801 countsByYear W29149028012019 @default.
- W2914902801 countsByYear W29149028012020 @default.
- W2914902801 countsByYear W29149028012021 @default.
- W2914902801 countsByYear W29149028012022 @default.
- W2914902801 crossrefType "posted-content" @default.
- W2914902801 hasAuthorship W2914902801A5044193587 @default.
- W2914902801 hasAuthorship W2914902801A5048166612 @default.
- W2914902801 hasAuthorship W2914902801A5071403558 @default.
- W2914902801 hasAuthorship W2914902801A5086430646 @default.
- W2914902801 hasConcept C111472728 @default.
- W2914902801 hasConcept C121332964 @default.
- W2914902801 hasConcept C126255220 @default.
- W2914902801 hasConcept C132010649 @default.
- W2914902801 hasConcept C134306372 @default.
- W2914902801 hasConcept C138885662 @default.
- W2914902801 hasConcept C151201525 @default.
- W2914902801 hasConcept C153258448 @default.
- W2914902801 hasConcept C154945302 @default.
- W2914902801 hasConcept C162324750 @default.
- W2914902801 hasConcept C202213908 @default.
- W2914902801 hasConcept C26517878 @default.
- W2914902801 hasConcept C2777303404 @default.
- W2914902801 hasConcept C28826006 @default.
- W2914902801 hasConcept C33923547 @default.
- W2914902801 hasConcept C38652104 @default.
- W2914902801 hasConcept C41008148 @default.
- W2914902801 hasConcept C50522688 @default.
- W2914902801 hasConcept C50644808 @default.
- W2914902801 hasConcept C51955184 @default.
- W2914902801 hasConcept C57869625 @default.
- W2914902801 hasConcept C62520636 @default.
- W2914902801 hasConcept C93779851 @default.
- W2914902801 hasConceptScore W2914902801C111472728 @default.
- W2914902801 hasConceptScore W2914902801C121332964 @default.
- W2914902801 hasConceptScore W2914902801C126255220 @default.
- W2914902801 hasConceptScore W2914902801C132010649 @default.
- W2914902801 hasConceptScore W2914902801C134306372 @default.
- W2914902801 hasConceptScore W2914902801C138885662 @default.
- W2914902801 hasConceptScore W2914902801C151201525 @default.
- W2914902801 hasConceptScore W2914902801C153258448 @default.
- W2914902801 hasConceptScore W2914902801C154945302 @default.
- W2914902801 hasConceptScore W2914902801C162324750 @default.
- W2914902801 hasConceptScore W2914902801C202213908 @default.
- W2914902801 hasConceptScore W2914902801C26517878 @default.
- W2914902801 hasConceptScore W2914902801C2777303404 @default.
- W2914902801 hasConceptScore W2914902801C28826006 @default.
- W2914902801 hasConceptScore W2914902801C33923547 @default.
- W2914902801 hasConceptScore W2914902801C38652104 @default.
- W2914902801 hasConceptScore W2914902801C41008148 @default.
- W2914902801 hasConceptScore W2914902801C50522688 @default.
- W2914902801 hasConceptScore W2914902801C50644808 @default.
- W2914902801 hasConceptScore W2914902801C51955184 @default.
- W2914902801 hasConceptScore W2914902801C57869625 @default.
- W2914902801 hasConceptScore W2914902801C62520636 @default.
- W2914902801 hasConceptScore W2914902801C93779851 @default.
- W2914902801 hasLocation W29149028011 @default.
- W2914902801 hasOpenAccess W2914902801 @default.
- W2914902801 hasPrimaryLocation W29149028011 @default.
- W2914902801 hasRelatedWork W1522579744 @default.
- W2914902801 hasRelatedWork W2016079886 @default.
- W2914902801 hasRelatedWork W2798986185 @default.
- W2914902801 hasRelatedWork W2921395510 @default.