Matches in SemOpenAlex for { <https://semopenalex.org/work/W2949233439> ?p ?o ?g. }
- W2949233439 abstract "Empirical evidence suggests that neural networks with ReLU activations generalize better with over-parameterization. However, there is currently no theoretical analysis that explains this observation. In this work, we provide theoretical and empirical evidence that, in certain cases, overparameterized convolutional networks generalize better than small networks because of an interplay between weight clustering and feature exploration at initialization. We demonstrate this theoretically for a 3-layer convolutional neural network with max-pooling, in a novel setting which extends the XOR problem. We show that this interplay implies that with overparamterization, gradient descent converges to global minima with better generalization performance compared to global minima of small networks. Empirically, we demonstrate these phenomena for a 3-layer convolutional neural network in the MNIST task." @default.
- W2949233439 created "2019-06-27" @default.
- W2949233439 creator A5029118855 @default.
- W2949233439 creator A5047817959 @default.
- W2949233439 date "2018-10-06" @default.
- W2949233439 modified "2023-09-27" @default.
- W2949233439 title "Why do Larger Models Generalize Better? A Theoretical Perspective via the XOR Problem" @default.
- W2949233439 cites W1522301498 @default.
- W2949233439 cites W2591714514 @default.
- W2949233439 cites W2593958421 @default.
- W2949233439 cites W2758053331 @default.
- W2949233439 cites W2777843033 @default.
- W2949233439 cites W2786622092 @default.
- W2949233439 cites W2807299122 @default.
- W2949233439 cites W2894604724 @default.
- W2949233439 cites W2900103278 @default.
- W2949233439 cites W2949804919 @default.
- W2949233439 cites W2962698540 @default.
- W2949233439 cites W2962930448 @default.
- W2949233439 cites W2963100491 @default.
- W2949233439 cites W2963417959 @default.
- W2949233439 cites W2963519230 @default.
- W2949233439 cites W2963695615 @default.
- W2949233439 cites W2963744427 @default.
- W2949233439 cites W607505555 @default.
- W2949233439 hasPublicationYear "2018" @default.
- W2949233439 type Work @default.
- W2949233439 sameAs 2949233439 @default.
- W2949233439 citedByCount "0" @default.
- W2949233439 crossrefType "posted-content" @default.
- W2949233439 hasAuthorship W2949233439A5029118855 @default.
- W2949233439 hasAuthorship W2949233439A5047817959 @default.
- W2949233439 hasConcept C114466953 @default.
- W2949233439 hasConcept C12713177 @default.
- W2949233439 hasConcept C134306372 @default.
- W2949233439 hasConcept C138885662 @default.
- W2949233439 hasConcept C153258448 @default.
- W2949233439 hasConcept C154945302 @default.
- W2949233439 hasConcept C177148314 @default.
- W2949233439 hasConcept C178790620 @default.
- W2949233439 hasConcept C185592680 @default.
- W2949233439 hasConcept C186633575 @default.
- W2949233439 hasConcept C190502265 @default.
- W2949233439 hasConcept C199360897 @default.
- W2949233439 hasConcept C2776401178 @default.
- W2949233439 hasConcept C2779227376 @default.
- W2949233439 hasConcept C33923547 @default.
- W2949233439 hasConcept C41008148 @default.
- W2949233439 hasConcept C41895202 @default.
- W2949233439 hasConcept C50644808 @default.
- W2949233439 hasConcept C70437156 @default.
- W2949233439 hasConcept C73555534 @default.
- W2949233439 hasConcept C81363708 @default.
- W2949233439 hasConceptScore W2949233439C114466953 @default.
- W2949233439 hasConceptScore W2949233439C12713177 @default.
- W2949233439 hasConceptScore W2949233439C134306372 @default.
- W2949233439 hasConceptScore W2949233439C138885662 @default.
- W2949233439 hasConceptScore W2949233439C153258448 @default.
- W2949233439 hasConceptScore W2949233439C154945302 @default.
- W2949233439 hasConceptScore W2949233439C177148314 @default.
- W2949233439 hasConceptScore W2949233439C178790620 @default.
- W2949233439 hasConceptScore W2949233439C185592680 @default.
- W2949233439 hasConceptScore W2949233439C186633575 @default.
- W2949233439 hasConceptScore W2949233439C190502265 @default.
- W2949233439 hasConceptScore W2949233439C199360897 @default.
- W2949233439 hasConceptScore W2949233439C2776401178 @default.
- W2949233439 hasConceptScore W2949233439C2779227376 @default.
- W2949233439 hasConceptScore W2949233439C33923547 @default.
- W2949233439 hasConceptScore W2949233439C41008148 @default.
- W2949233439 hasConceptScore W2949233439C41895202 @default.
- W2949233439 hasConceptScore W2949233439C50644808 @default.
- W2949233439 hasConceptScore W2949233439C70437156 @default.
- W2949233439 hasConceptScore W2949233439C73555534 @default.
- W2949233439 hasConceptScore W2949233439C81363708 @default.
- W2949233439 hasLocation W29492334391 @default.
- W2949233439 hasOpenAccess W2949233439 @default.
- W2949233439 hasPrimaryLocation W29492334391 @default.
- W2949233439 hasRelatedWork W2618907597 @default.
- W2949233439 hasRelatedWork W2765861987 @default.
- W2949233439 hasRelatedWork W2787442904 @default.
- W2949233439 hasRelatedWork W2897080984 @default.
- W2949233439 hasRelatedWork W2899748887 @default.
- W2949233439 hasRelatedWork W2927724204 @default.
- W2949233439 hasRelatedWork W2944820194 @default.
- W2949233439 hasRelatedWork W2945236295 @default.
- W2949233439 hasRelatedWork W2955304652 @default.
- W2949233439 hasRelatedWork W2961321674 @default.
- W2949233439 hasRelatedWork W2982079119 @default.
- W2949233439 hasRelatedWork W2991290085 @default.
- W2949233439 hasRelatedWork W3092830796 @default.
- W2949233439 hasRelatedWork W3119215351 @default.
- W2949233439 hasRelatedWork W3119755129 @default.
- W2949233439 hasRelatedWork W3164547988 @default.
- W2949233439 hasRelatedWork W3166177065 @default.
- W2949233439 hasRelatedWork W3199400829 @default.
- W2949233439 hasRelatedWork W3034194698 @default.
- W2949233439 hasRelatedWork W3090270091 @default.
- W2949233439 isParatext "false" @default.
- W2949233439 isRetracted "false" @default.
- W2949233439 magId "2949233439" @default.