Matches in SemOpenAlex for { <https://semopenalex.org/work/W3186021038> ?p ?o ?g. }
- W3186021038 abstract "We revisit on-average algorithmic stability of Gradient Descent (GD) for training overparameterised shallow neural networks and prove new generalisation and excess risk bounds without the Neural Tangent Kernel (NTK) or Polyak-Łojasiewicz (PL) assumptions. In particular, we show oracle type bounds which reveal that the generalisation and excess risk of GD is controlled by an interpolating network with the shortest GD path from initialisation (in a sense, an interpolating network with the smallest relative norm). While this was known for kernelised interpolants, our proof applies directly to networks trained by GD without intermediate kernelisation. At the same time, by relaxing oracle inequalities developed here we recover existing NTK-based risk bounds in a straightforward way, which demonstrates that our analysis is tighter. Finally, unlike most of the NTK-based analyses we focus on regression with label noise and show that GD with early stopping is consistent." @default.
- W3186021038 created "2021-08-02" @default.
- W3186021038 creator A5041854594 @default.
- W3186021038 creator A5068525966 @default.
- W3186021038 date "2021-07-27" @default.
- W3186021038 modified "2023-10-06" @default.
- W3186021038 title "Stability & Generalisation of Gradient Descent for Shallow Neural Networks without the Neural Tangent Kernel" @default.
- W3186021038 cites W1542886316 @default.
- W3186021038 cites W1560724230 @default.
- W3186021038 cites W2034978228 @default.
- W3186021038 cites W2139338362 @default.
- W3186021038 cites W2579923771 @default.
- W3186021038 cites W2604451472 @default.
- W3186021038 cites W2741470927 @default.
- W3186021038 cites W2752366553 @default.
- W3186021038 cites W2795605442 @default.
- W3186021038 cites W2809090039 @default.
- W3186021038 cites W2912260645 @default.
- W3186021038 cites W2963094221 @default.
- W3186021038 cites W2963239103 @default.
- W3186021038 cites W2963334018 @default.
- W3186021038 cites W2963794891 @default.
- W3186021038 cites W2964098911 @default.
- W3186021038 cites W2964161337 @default.
- W3186021038 cites W2965157832 @default.
- W3186021038 cites W2995015865 @default.
- W3186021038 cites W3021189130 @default.
- W3186021038 cites W3034851139 @default.
- W3186021038 cites W3037144731 @default.
- W3186021038 cites W3039141380 @default.
- W3186021038 cites W3046508829 @default.
- W3186021038 cites W3092612324 @default.
- W3186021038 cites W3102429474 @default.
- W3186021038 cites W3112275392 @default.
- W3186021038 cites W3119586787 @default.
- W3186021038 cites W3121965758 @default.
- W3186021038 cites W3126211418 @default.
- W3186021038 cites W3141595720 @default.
- W3186021038 cites W3157503520 @default.
- W3186021038 cites W3158239976 @default.
- W3186021038 cites W3165421101 @default.
- W3186021038 cites W3173288964 @default.
- W3186021038 cites W3191067499 @default.
- W3186021038 cites W3214104957 @default.
- W3186021038 cites W607505555 @default.
- W3186021038 hasPublicationYear "2021" @default.
- W3186021038 type Work @default.
- W3186021038 sameAs 3186021038 @default.
- W3186021038 citedByCount "0" @default.
- W3186021038 crossrefType "posted-content" @default.
- W3186021038 hasAuthorship W3186021038A5041854594 @default.
- W3186021038 hasAuthorship W3186021038A5068525966 @default.
- W3186021038 hasConcept C112972136 @default.
- W3186021038 hasConcept C11413529 @default.
- W3186021038 hasConcept C114614502 @default.
- W3186021038 hasConcept C115903868 @default.
- W3186021038 hasConcept C119857082 @default.
- W3186021038 hasConcept C138187205 @default.
- W3186021038 hasConcept C153258448 @default.
- W3186021038 hasConcept C154945302 @default.
- W3186021038 hasConcept C17744445 @default.
- W3186021038 hasConcept C191795146 @default.
- W3186021038 hasConcept C199539241 @default.
- W3186021038 hasConcept C2524010 @default.
- W3186021038 hasConcept C28826006 @default.
- W3186021038 hasConcept C33923547 @default.
- W3186021038 hasConcept C41008148 @default.
- W3186021038 hasConcept C50644808 @default.
- W3186021038 hasConcept C55166926 @default.
- W3186021038 hasConcept C74193536 @default.
- W3186021038 hasConceptScore W3186021038C112972136 @default.
- W3186021038 hasConceptScore W3186021038C11413529 @default.
- W3186021038 hasConceptScore W3186021038C114614502 @default.
- W3186021038 hasConceptScore W3186021038C115903868 @default.
- W3186021038 hasConceptScore W3186021038C119857082 @default.
- W3186021038 hasConceptScore W3186021038C138187205 @default.
- W3186021038 hasConceptScore W3186021038C153258448 @default.
- W3186021038 hasConceptScore W3186021038C154945302 @default.
- W3186021038 hasConceptScore W3186021038C17744445 @default.
- W3186021038 hasConceptScore W3186021038C191795146 @default.
- W3186021038 hasConceptScore W3186021038C199539241 @default.
- W3186021038 hasConceptScore W3186021038C2524010 @default.
- W3186021038 hasConceptScore W3186021038C28826006 @default.
- W3186021038 hasConceptScore W3186021038C33923547 @default.
- W3186021038 hasConceptScore W3186021038C41008148 @default.
- W3186021038 hasConceptScore W3186021038C50644808 @default.
- W3186021038 hasConceptScore W3186021038C55166926 @default.
- W3186021038 hasConceptScore W3186021038C74193536 @default.
- W3186021038 hasLocation W31860210381 @default.
- W3186021038 hasOpenAccess W3186021038 @default.
- W3186021038 hasPrimaryLocation W31860210381 @default.
- W3186021038 hasRelatedWork W1560379207 @default.
- W3186021038 hasRelatedWork W2065976408 @default.
- W3186021038 hasRelatedWork W2129247014 @default.
- W3186021038 hasRelatedWork W2164688468 @default.
- W3186021038 hasRelatedWork W2753713566 @default.
- W3186021038 hasRelatedWork W2755553805 @default.
- W3186021038 hasRelatedWork W2963956929 @default.
- W3186021038 hasRelatedWork W2996822312 @default.
- W3186021038 hasRelatedWork W3039771339 @default.