Matches in SemOpenAlex for { <https://semopenalex.org/work/W3168760371> ?p ?o ?g. }
Showing items 1 to 90 of
90
with 100 items per page.
- W3168760371 endingPage "8947" @default.
- W3168760371 startingPage "8936" @default.
- W3168760371 abstract "A recent series of theoretical works showed that the dynamics of neural networks with a certain initialisation are well-captured by kernel methods. Concurrent empirical work demonstrated that kernel methods can come close to the performance of neural networks on some image classification tasks. These results raise the question of whether neural networks only learn successfully if kernels also learn successfully, despite neural networks being more expressive. Here, we show theoretically that two-layer neural networks (2LNN) with only a few hidden neurons can beat the performance of kernel learning on a simple Gaussian mixture classification task. We study the high-dimensional limit where the number of samples is linearly proportional to the input dimension, and show that while small 2LNN achieve near-optimal performance on this task, lazy training approaches such as random features and kernel methods do not. Our analysis is based on the derivation of a closed set of equations that track the learning dynamics of the 2LNN and thus allow to extract the asymptotic performance of the network as a function of signal-to-noise ratio and other hyperparameters. We finally illustrate how over-parametrising the neural network leads to faster convergence, but does not improve its final performance." @default.
- W3168760371 created "2021-06-22" @default.
- W3168760371 creator A5046159570 @default.
- W3168760371 creator A5057039281 @default.
- W3168760371 creator A5068236230 @default.
- W3168760371 creator A5089268172 @default.
- W3168760371 date "2021-07-18" @default.
- W3168760371 modified "2023-10-17" @default.
- W3168760371 title "Classifying high-dimensional Gaussian mixtures: Where kernel methods fail and neural networks succeed" @default.
- W3168760371 hasPublicationYear "2021" @default.
- W3168760371 type Work @default.
- W3168760371 sameAs 3168760371 @default.
- W3168760371 citedByCount "0" @default.
- W3168760371 crossrefType "proceedings-article" @default.
- W3168760371 hasAuthorship W3168760371A5046159570 @default.
- W3168760371 hasAuthorship W3168760371A5057039281 @default.
- W3168760371 hasAuthorship W3168760371A5068236230 @default.
- W3168760371 hasAuthorship W3168760371A5089268172 @default.
- W3168760371 hasConcept C11413529 @default.
- W3168760371 hasConcept C114614502 @default.
- W3168760371 hasConcept C119857082 @default.
- W3168760371 hasConcept C121332964 @default.
- W3168760371 hasConcept C122280245 @default.
- W3168760371 hasConcept C12267149 @default.
- W3168760371 hasConcept C134306372 @default.
- W3168760371 hasConcept C151201525 @default.
- W3168760371 hasConcept C153180895 @default.
- W3168760371 hasConcept C154945302 @default.
- W3168760371 hasConcept C162324750 @default.
- W3168760371 hasConcept C163716315 @default.
- W3168760371 hasConcept C2777303404 @default.
- W3168760371 hasConcept C33923547 @default.
- W3168760371 hasConcept C41008148 @default.
- W3168760371 hasConcept C50522688 @default.
- W3168760371 hasConcept C50644808 @default.
- W3168760371 hasConcept C62520636 @default.
- W3168760371 hasConcept C7218915 @default.
- W3168760371 hasConcept C74193536 @default.
- W3168760371 hasConcept C8642999 @default.
- W3168760371 hasConceptScore W3168760371C11413529 @default.
- W3168760371 hasConceptScore W3168760371C114614502 @default.
- W3168760371 hasConceptScore W3168760371C119857082 @default.
- W3168760371 hasConceptScore W3168760371C121332964 @default.
- W3168760371 hasConceptScore W3168760371C122280245 @default.
- W3168760371 hasConceptScore W3168760371C12267149 @default.
- W3168760371 hasConceptScore W3168760371C134306372 @default.
- W3168760371 hasConceptScore W3168760371C151201525 @default.
- W3168760371 hasConceptScore W3168760371C153180895 @default.
- W3168760371 hasConceptScore W3168760371C154945302 @default.
- W3168760371 hasConceptScore W3168760371C162324750 @default.
- W3168760371 hasConceptScore W3168760371C163716315 @default.
- W3168760371 hasConceptScore W3168760371C2777303404 @default.
- W3168760371 hasConceptScore W3168760371C33923547 @default.
- W3168760371 hasConceptScore W3168760371C41008148 @default.
- W3168760371 hasConceptScore W3168760371C50522688 @default.
- W3168760371 hasConceptScore W3168760371C50644808 @default.
- W3168760371 hasConceptScore W3168760371C62520636 @default.
- W3168760371 hasConceptScore W3168760371C7218915 @default.
- W3168760371 hasConceptScore W3168760371C74193536 @default.
- W3168760371 hasConceptScore W3168760371C8642999 @default.
- W3168760371 hasLocation W31687603711 @default.
- W3168760371 hasOpenAccess W3168760371 @default.
- W3168760371 hasPrimaryLocation W31687603711 @default.
- W3168760371 hasRelatedWork W1973584826 @default.
- W3168760371 hasRelatedWork W2113517874 @default.
- W3168760371 hasRelatedWork W2225087728 @default.
- W3168760371 hasRelatedWork W2372579376 @default.
- W3168760371 hasRelatedWork W2737836580 @default.
- W3168760371 hasRelatedWork W2739997491 @default.
- W3168760371 hasRelatedWork W2762534372 @default.
- W3168760371 hasRelatedWork W2785333883 @default.
- W3168760371 hasRelatedWork W2890311316 @default.
- W3168760371 hasRelatedWork W2899008792 @default.
- W3168760371 hasRelatedWork W2914317573 @default.
- W3168760371 hasRelatedWork W2914753953 @default.
- W3168760371 hasRelatedWork W2925974453 @default.
- W3168760371 hasRelatedWork W2945319244 @default.
- W3168760371 hasRelatedWork W2982282910 @default.
- W3168760371 hasRelatedWork W2987612595 @default.
- W3168760371 hasRelatedWork W3005381706 @default.
- W3168760371 hasRelatedWork W3103732786 @default.
- W3168760371 hasRelatedWork W3129819231 @default.
- W3168760371 hasRelatedWork W3172467154 @default.
- W3168760371 hasVolume "139" @default.
- W3168760371 isParatext "false" @default.
- W3168760371 isRetracted "false" @default.
- W3168760371 magId "3168760371" @default.
- W3168760371 workType "article" @default.