Matches in SemOpenAlex for { <https://semopenalex.org/work/W2748447781> ?p ?o ?g. }
- W2748447781 abstract "We consider the problem of learning function classes computed by neural networks with various activations (e.g. ReLU or Sigmoid), a task believed to be computationally intractable in the worst-case. A major open problem is to understand the minimal assumptions under which these classes admit provably efficient algorithms. In this work we show that a natural distributional assumption corresponding to {em eigenvalue decay} of the Gram matrix yields polynomial-time algorithms in the non-realizable setting for expressive classes of networks (e.g. feed-forward networks of ReLUs). We make no assumptions on the structure of the network or the labels. Given sufficiently-strong polynomial eigenvalue decay, we obtain {em fully}-polynomial time algorithms in {em all} the relevant parameters with respect to square-loss. Milder decay assumptions also lead to improved algorithms. This is the first purely distributional assumption that leads to polynomial-time algorithms for networks of ReLUs, even with one hidden layer. Further, unlike prior distributional assumptions (e.g., the marginal distribution is Gaussian), eigenvalue decay has been observed in practice on common data sets." @default.
- W2748447781 created "2017-08-31" @default.
- W2748447781 creator A5006626709 @default.
- W2748447781 creator A5019144684 @default.
- W2748447781 date "2017-08-11" @default.
- W2748447781 modified "2023-09-27" @default.
- W2748447781 title "Eigenvalue Decay Implies Polynomial-Time Learnability for Neural Networks" @default.
- W2748447781 cites W1499660792 @default.
- W2748447781 cites W1560724230 @default.
- W2748447781 cites W1579249934 @default.
- W2748447781 cites W1839868949 @default.
- W2748447781 cites W1998269045 @default.
- W2748447781 cites W2019363670 @default.
- W2748447781 cites W2084544490 @default.
- W2748447781 cites W2101762657 @default.
- W2748447781 cites W2112545207 @default.
- W2748447781 cites W2119340451 @default.
- W2748447781 cites W2119877753 @default.
- W2748447781 cites W2126464735 @default.
- W2748447781 cites W2134507744 @default.
- W2748447781 cites W2140125566 @default.
- W2748447781 cites W2151683004 @default.
- W2748447781 cites W2160354932 @default.
- W2748447781 cites W2168002876 @default.
- W2748447781 cites W2173323516 @default.
- W2748447781 cites W2221366174 @default.
- W2748447781 cites W2399994860 @default.
- W2748447781 cites W2509435534 @default.
- W2748447781 cites W2573849764 @default.
- W2748447781 cites W2579923771 @default.
- W2748447781 cites W2594845788 @default.
- W2748447781 cites W2735616495 @default.
- W2748447781 cites W2951207584 @default.
- W2748447781 cites W2952318479 @default.
- W2748447781 cites W2952923506 @default.
- W2748447781 cites W2952995998 @default.
- W2748447781 cites W2953243802 @default.
- W2748447781 cites W2962939986 @default.
- W2748447781 cites W2963446085 @default.
- W2748447781 cites W2964220529 @default.
- W2748447781 cites W349434659 @default.
- W2748447781 cites W607505555 @default.
- W2748447781 cites W65382156 @default.
- W2748447781 cites W747345242 @default.
- W2748447781 hasPublicationYear "2017" @default.
- W2748447781 type Work @default.
- W2748447781 sameAs 2748447781 @default.
- W2748447781 citedByCount "5" @default.
- W2748447781 countsByYear W27484477812017 @default.
- W2748447781 countsByYear W27484477812018 @default.
- W2748447781 crossrefType "posted-content" @default.
- W2748447781 hasAuthorship W2748447781A5006626709 @default.
- W2748447781 hasAuthorship W2748447781A5019144684 @default.
- W2748447781 hasConcept C106487976 @default.
- W2748447781 hasConcept C11413529 @default.
- W2748447781 hasConcept C118615104 @default.
- W2748447781 hasConcept C121332964 @default.
- W2748447781 hasConcept C134306372 @default.
- W2748447781 hasConcept C14036430 @default.
- W2748447781 hasConcept C154945302 @default.
- W2748447781 hasConcept C158693339 @default.
- W2748447781 hasConcept C159985019 @default.
- W2748447781 hasConcept C163716315 @default.
- W2748447781 hasConcept C192562407 @default.
- W2748447781 hasConcept C2777723229 @default.
- W2748447781 hasConcept C28826006 @default.
- W2748447781 hasConcept C311688 @default.
- W2748447781 hasConcept C33923547 @default.
- W2748447781 hasConcept C41008148 @default.
- W2748447781 hasConcept C50644808 @default.
- W2748447781 hasConcept C62520636 @default.
- W2748447781 hasConcept C78458016 @default.
- W2748447781 hasConcept C81388566 @default.
- W2748447781 hasConcept C86803240 @default.
- W2748447781 hasConcept C90119067 @default.
- W2748447781 hasConceptScore W2748447781C106487976 @default.
- W2748447781 hasConceptScore W2748447781C11413529 @default.
- W2748447781 hasConceptScore W2748447781C118615104 @default.
- W2748447781 hasConceptScore W2748447781C121332964 @default.
- W2748447781 hasConceptScore W2748447781C134306372 @default.
- W2748447781 hasConceptScore W2748447781C14036430 @default.
- W2748447781 hasConceptScore W2748447781C154945302 @default.
- W2748447781 hasConceptScore W2748447781C158693339 @default.
- W2748447781 hasConceptScore W2748447781C159985019 @default.
- W2748447781 hasConceptScore W2748447781C163716315 @default.
- W2748447781 hasConceptScore W2748447781C192562407 @default.
- W2748447781 hasConceptScore W2748447781C2777723229 @default.
- W2748447781 hasConceptScore W2748447781C28826006 @default.
- W2748447781 hasConceptScore W2748447781C311688 @default.
- W2748447781 hasConceptScore W2748447781C33923547 @default.
- W2748447781 hasConceptScore W2748447781C41008148 @default.
- W2748447781 hasConceptScore W2748447781C50644808 @default.
- W2748447781 hasConceptScore W2748447781C62520636 @default.
- W2748447781 hasConceptScore W2748447781C78458016 @default.
- W2748447781 hasConceptScore W2748447781C81388566 @default.
- W2748447781 hasConceptScore W2748447781C86803240 @default.
- W2748447781 hasConceptScore W2748447781C90119067 @default.
- W2748447781 hasLocation W27484477811 @default.
- W2748447781 hasOpenAccess W2748447781 @default.
- W2748447781 hasPrimaryLocation W27484477811 @default.