Matches in SemOpenAlex for { <https://semopenalex.org/work/W3092390915> ?p ?o ?g. }
- W3092390915 abstract "A recent breakthrough in deep learning theory shows that the training of over-parameterized deep neural networks can be characterized by a kernel function called textit{neural tangent kernel} (NTK). However, it is known that this type of results does not perfectly match the practice, as NTK-based analysis requires the network weights to stay very close to their initialization throughout training, and cannot handle regularizers or gradient noises. In this paper, we provide a generalized neural tangent kernel analysis and show that noisy gradient descent with weight decay can still exhibit a kernel-like behavior. This implies that the training loss converges linearly up to a certain accuracy. We also establish a novel generalization error bound for two-layer neural networks trained by noisy gradient descent with weight decay." @default.
- W3092390915 created "2020-10-15" @default.
- W3092390915 creator A5037709424 @default.
- W3092390915 creator A5051448391 @default.
- W3092390915 creator A5082974417 @default.
- W3092390915 creator A5086329948 @default.
- W3092390915 date "2020-02-10" @default.
- W3092390915 modified "2023-09-27" @default.
- W3092390915 title "A Generalized Neural Tangent Kernel Analysis for Two-layer Neural Networks" @default.
- W3092390915 cites W1944672 @default.
- W3092390915 cites W2103972570 @default.
- W3092390915 cites W2130401121 @default.
- W3092390915 cites W2136144249 @default.
- W3092390915 cites W2160815625 @default.
- W3092390915 cites W2163605009 @default.
- W3092390915 cites W2257979135 @default.
- W3092390915 cites W2579923771 @default.
- W3092390915 cites W2593709294 @default.
- W3092390915 cites W2613481513 @default.
- W3092390915 cites W2618069152 @default.
- W3092390915 cites W2618398196 @default.
- W3092390915 cites W2625063094 @default.
- W3092390915 cites W2741952635 @default.
- W3092390915 cites W2805926765 @default.
- W3092390915 cites W2808157231 @default.
- W3092390915 cites W2808958252 @default.
- W3092390915 cites W2809090039 @default.
- W3092390915 cites W2886067286 @default.
- W3092390915 cites W2899748887 @default.
- W3092390915 cites W2900959181 @default.
- W3092390915 cites W2911742574 @default.
- W3092390915 cites W2913892099 @default.
- W3092390915 cites W2914811257 @default.
- W3092390915 cites W2945175884 @default.
- W3092390915 cites W2952126211 @default.
- W3092390915 cites W2952204734 @default.
- W3092390915 cites W2952318479 @default.
- W3092390915 cites W2962698540 @default.
- W3092390915 cites W2962857907 @default.
- W3092390915 cites W2963095610 @default.
- W3092390915 cites W2963236897 @default.
- W3092390915 cites W2963239103 @default.
- W3092390915 cites W2963481221 @default.
- W3092390915 cites W2963534251 @default.
- W3092390915 cites W2963664410 @default.
- W3092390915 cites W2963798163 @default.
- W3092390915 cites W2964031251 @default.
- W3092390915 cites W2964161337 @default.
- W3092390915 cites W2964915616 @default.
- W3092390915 cites W2965772785 @default.
- W3092390915 cites W2970176397 @default.
- W3092390915 cites W2970330753 @default.
- W3092390915 cites W2970332347 @default.
- W3092390915 cites W2970388773 @default.
- W3092390915 cites W2970443625 @default.
- W3092390915 cites W2970862128 @default.
- W3092390915 cites W2970890756 @default.
- W3092390915 cites W2971043187 @default.
- W3092390915 cites W2977969175 @default.
- W3092390915 cites W2981407587 @default.
- W3092390915 cites W2982079119 @default.
- W3092390915 cites W2983999360 @default.
- W3092390915 cites W2994110075 @default.
- W3092390915 cites W3000127803 @default.
- W3092390915 cites W3000593479 @default.
- W3092390915 cites W3005475467 @default.
- W3092390915 cites W3007957341 @default.
- W3092390915 cites W3119586787 @default.
- W3092390915 cites W423603040 @default.
- W3092390915 cites W607505555 @default.
- W3092390915 cites W2991401328 @default.
- W3092390915 hasPublicationYear "2020" @default.
- W3092390915 type Work @default.
- W3092390915 sameAs 3092390915 @default.
- W3092390915 citedByCount "5" @default.
- W3092390915 countsByYear W30923909152019 @default.
- W3092390915 countsByYear W30923909152020 @default.
- W3092390915 countsByYear W30923909152021 @default.
- W3092390915 crossrefType "posted-content" @default.
- W3092390915 hasAuthorship W3092390915A5037709424 @default.
- W3092390915 hasAuthorship W3092390915A5051448391 @default.
- W3092390915 hasAuthorship W3092390915A5082974417 @default.
- W3092390915 hasAuthorship W3092390915A5086329948 @default.
- W3092390915 hasConcept C11413529 @default.
- W3092390915 hasConcept C114466953 @default.
- W3092390915 hasConcept C118615104 @default.
- W3092390915 hasConcept C122280245 @default.
- W3092390915 hasConcept C12267149 @default.
- W3092390915 hasConcept C134306372 @default.
- W3092390915 hasConcept C138187205 @default.
- W3092390915 hasConcept C153258448 @default.
- W3092390915 hasConcept C154945302 @default.
- W3092390915 hasConcept C165464430 @default.
- W3092390915 hasConcept C177148314 @default.
- W3092390915 hasConcept C199360897 @default.
- W3092390915 hasConcept C2524010 @default.
- W3092390915 hasConcept C28826006 @default.
- W3092390915 hasConcept C33923547 @default.
- W3092390915 hasConcept C41008148 @default.
- W3092390915 hasConcept C50644808 @default.