FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3036917524> ?p ?o ?g. }

• W3036917524 abstract "In Continual Learning settings, deep neural networks are prone to Catastrophic Forgetting. Orthogonal Gradient Descent (Farajtabar et al., 2019) was proposed to tackle the challenge. However, no theoretical guarantees have been proven yet. We present a theoretical framework to study Continual Learning algorithms in the NTK regime. This framework comprises closed form expression of the model through tasks and proxies for transfer learning, generalisation and tasks similarity. In this framework, we prove that OGD is robust to Catastrophic Forgetting then derive the first generalisation bound for SGD and OGD for Continual Learning. Finally, we study the limits of this framework in practice for OGD and highlight the importance of the NTK variation for Continual Learning." @default.
• W3036917524 created "2020-06-25" @default.
• W3036917524 creator A5028224043 @default.
• W3036917524 creator A5072744508 @default.
• W3036917524 creator A5080464225 @default.
• W3036917524 date "2021-05-04" @default.
• W3036917524 modified "2023-09-27" @default.
• W3036917524 title "Generalisation Guarantees For Continual Learning With Orthogonal Gradient Descent" @default.
• W3036917524 cites W1682403713 @default.
• W3036917524 cites W2007154098 @default.
• W3036917524 cites W2036043322 @default.
• W3036917524 cites W2036963181 @default.
• W3036917524 cites W2060277733 @default.
• W3036917524 cites W2061524864 @default.
• W3036917524 cites W2088135577 @default.
• W3036917524 cites W2112796928 @default.
• W3036917524 cites W2116522068 @default.
• W3036917524 cites W2163605009 @default.
• W3036917524 cites W2296073425 @default.
• W3036917524 cites W2504108613 @default.
• W3036917524 cites W2560647685 @default.
• W3036917524 cites W2579923771 @default.
• W3036917524 cites W2597846060 @default.
• W3036917524 cites W2737492962 @default.
• W3036917524 cites W2782276075 @default.
• W3036917524 cites W2787295326 @default.
• W3036917524 cites W2788388592 @default.
• W3036917524 cites W2790068046 @default.
• W3036917524 cites W2809090039 @default.
• W3036917524 cites W2890870050 @default.
• W3036917524 cites W2894094671 @default.
• W3036917524 cites W2894604724 @default.
• W3036917524 cites W2898892019 @default.
• W3036917524 cites W2902625698 @default.
• W3036917524 cites W2911867426 @default.
• W3036917524 cites W2925181975 @default.
• W3036917524 cites W2932529478 @default.
• W3036917524 cites W2939137134 @default.
• W3036917524 cites W2949117887 @default.
• W3036917524 cites W2962724315 @default.
• W3036917524 cites W2963003887 @default.
• W3036917524 cites W2963540014 @default.
• W3036917524 cites W2963577698 @default.
• W3036917524 cites W2963588172 @default.
• W3036917524 cites W2963649256 @default.
• W3036917524 cites W2964067969 @default.
• W3036917524 cites W2964790801 @default.
• W3036917524 cites W2970176397 @default.
• W3036917524 cites W2970239942 @default.
• W3036917524 cites W2970540478 @default.
• W3036917524 cites W2975529272 @default.
• W3036917524 cites W2978031686 @default.
• W3036917524 cites W2980753060 @default.
• W3036917524 cites W2982079119 @default.
• W3036917524 cites W2995097486 @default.
• W3036917524 cites W2996653965 @default.
• W3036917524 cites W2997359900 @default.
• W3036917524 cites W3007891689 @default.
• W3036917524 cites W3008360155 @default.
• W3036917524 cites W3021137821 @default.
• W3036917524 cites W3035196707 @default.
• W3036917524 cites W3035350369 @default.
• W3036917524 cites W3035525746 @default.
• W3036917524 cites W3035546118 @default.
• W3036917524 cites W3100678069 @default.
• W3036917524 cites W3118608800 @default.
• W3036917524 cites W3120402410 @default.
• W3036917524 hasPublicationYear "2021" @default.
• W3036917524 type Work @default.
• W3036917524 sameAs 3036917524 @default.
• W3036917524 citedByCount "10" @default.
• W3036917524 countsByYear W30369175242020 @default.
• W3036917524 countsByYear W30369175242021 @default.
• W3036917524 crossrefType "journal-article" @default.
• W3036917524 hasAuthorship W3036917524A5028224043 @default.
• W3036917524 hasAuthorship W3036917524A5072744508 @default.
• W3036917524 hasAuthorship W3036917524A5080464225 @default.
• W3036917524 hasConcept C103278499 @default.
• W3036917524 hasConcept C115961682 @default.
• W3036917524 hasConcept C119857082 @default.
• W3036917524 hasConcept C121332964 @default.
• W3036917524 hasConcept C127413603 @default.
• W3036917524 hasConcept C146978453 @default.
• W3036917524 hasConcept C150899416 @default.
• W3036917524 hasConcept C153258448 @default.
• W3036917524 hasConcept C154945302 @default.
• W3036917524 hasConcept C15744967 @default.
• W3036917524 hasConcept C180747234 @default.
• W3036917524 hasConcept C206688291 @default.
• W3036917524 hasConcept C2776637919 @default.
• W3036917524 hasConcept C2778334786 @default.
• W3036917524 hasConcept C41008148 @default.
• W3036917524 hasConcept C44870925 @default.
• W3036917524 hasConcept C50644808 @default.
• W3036917524 hasConcept C7149132 @default.
• W3036917524 hasConceptScore W3036917524C103278499 @default.
• W3036917524 hasConceptScore W3036917524C115961682 @default.
• W3036917524 hasConceptScore W3036917524C119857082 @default.
• W3036917524 hasConceptScore W3036917524C121332964 @default.
• W3036917524 hasConceptScore W3036917524C127413603 @default.

• next