Matches in SemOpenAlex for { <https://semopenalex.org/work/W2910142656> ?p ?o ?g. }
- W2910142656 abstract "For a long time, designing neural architectures that exhibit high performance was considered a dark art that required expert hand-tuning. One of the few well-known guidelines for architecture design is the avoidance of exploding gradients, though even this guideline has remained relatively vague and circumstantial. We introduce the nonlinearity coefficient (NLC), a measurement of the complexity of the function computed by a neural network that is based on the magnitude of the gradient. Via an extensive empirical study, we show that the NLC is a powerful predictor of test error and that attaining a right-sized NLC is essential for optimal performance. The NLC exhibits a range of intriguing and important properties. It is closely tied to the amount of information gained from computing a single network gradient. It is tied to the error incurred when replacing the nonlinearity operations in the network with linear operations. It is not susceptible to the confounders of multiplicative scaling, additive bias and layer width. It is stable from layer to layer. Hence, we argue that the NLC is the first robust predictor of overfitting in deep networks." @default.
- W2910142656 created "2019-01-25" @default.
- W2910142656 creator A5037655437 @default.
- W2910142656 creator A5062362044 @default.
- W2910142656 date "2018-09-27" @default.
- W2910142656 modified "2023-09-27" @default.
- W2910142656 title "The Nonlinearity Coefficient - Predicting Generalization in Deep Neural Networks" @default.
- W2910142656 cites W1533861849 @default.
- W2910142656 cites W1677182931 @default.
- W2910142656 cites W1815076433 @default.
- W2910142656 cites W1836465849 @default.
- W2910142656 cites W2107878631 @default.
- W2910142656 cites W2126229007 @default.
- W2910142656 cites W2141473882 @default.
- W2910142656 cites W2161388792 @default.
- W2910142656 cites W2208555118 @default.
- W2910142656 cites W2253400648 @default.
- W2910142656 cites W2402758762 @default.
- W2910142656 cites W2513671774 @default.
- W2910142656 cites W2556364298 @default.
- W2910142656 cites W2591954064 @default.
- W2910142656 cites W2624413595 @default.
- W2910142656 cites W2752851182 @default.
- W2910142656 cites W2794922239 @default.
- W2910142656 cites W2808460953 @default.
- W2910142656 cites W2949427019 @default.
- W2910142656 cites W2962958829 @default.
- W2910142656 cites W2962971773 @default.
- W2910142656 cites W2963042606 @default.
- W2910142656 cites W2963063862 @default.
- W2910142656 cites W2963148870 @default.
- W2910142656 cites W2963172539 @default.
- W2910142656 cites W2963405349 @default.
- W2910142656 cites W2963504252 @default.
- W2910142656 cites W2963570896 @default.
- W2910142656 cites W2963679562 @default.
- W2910142656 cites W2963982496 @default.
- W2910142656 cites W2964088238 @default.
- W2910142656 cites W2964294232 @default.
- W2910142656 cites W2964313743 @default.
- W2910142656 cites W3037932933 @default.
- W2910142656 cites W194249466 @default.
- W2910142656 hasPublicationYear "2018" @default.
- W2910142656 type Work @default.
- W2910142656 sameAs 2910142656 @default.
- W2910142656 citedByCount "4" @default.
- W2910142656 countsByYear W29101426562019 @default.
- W2910142656 countsByYear W29101426562020 @default.
- W2910142656 crossrefType "posted-content" @default.
- W2910142656 hasAuthorship W2910142656A5037655437 @default.
- W2910142656 hasAuthorship W2910142656A5062362044 @default.
- W2910142656 hasConcept C11413529 @default.
- W2910142656 hasConcept C121332964 @default.
- W2910142656 hasConcept C127413603 @default.
- W2910142656 hasConcept C134306372 @default.
- W2910142656 hasConcept C14036430 @default.
- W2910142656 hasConcept C146978453 @default.
- W2910142656 hasConcept C154945302 @default.
- W2910142656 hasConcept C158622935 @default.
- W2910142656 hasConcept C159985019 @default.
- W2910142656 hasConcept C177148314 @default.
- W2910142656 hasConcept C192562407 @default.
- W2910142656 hasConcept C193415008 @default.
- W2910142656 hasConcept C204323151 @default.
- W2910142656 hasConcept C22019652 @default.
- W2910142656 hasConcept C2524010 @default.
- W2910142656 hasConcept C2779227376 @default.
- W2910142656 hasConcept C33923547 @default.
- W2910142656 hasConcept C38652104 @default.
- W2910142656 hasConcept C41008148 @default.
- W2910142656 hasConcept C42747912 @default.
- W2910142656 hasConcept C50644808 @default.
- W2910142656 hasConcept C62520636 @default.
- W2910142656 hasConcept C78458016 @default.
- W2910142656 hasConcept C86803240 @default.
- W2910142656 hasConcept C99844830 @default.
- W2910142656 hasConceptScore W2910142656C11413529 @default.
- W2910142656 hasConceptScore W2910142656C121332964 @default.
- W2910142656 hasConceptScore W2910142656C127413603 @default.
- W2910142656 hasConceptScore W2910142656C134306372 @default.
- W2910142656 hasConceptScore W2910142656C14036430 @default.
- W2910142656 hasConceptScore W2910142656C146978453 @default.
- W2910142656 hasConceptScore W2910142656C154945302 @default.
- W2910142656 hasConceptScore W2910142656C158622935 @default.
- W2910142656 hasConceptScore W2910142656C159985019 @default.
- W2910142656 hasConceptScore W2910142656C177148314 @default.
- W2910142656 hasConceptScore W2910142656C192562407 @default.
- W2910142656 hasConceptScore W2910142656C193415008 @default.
- W2910142656 hasConceptScore W2910142656C204323151 @default.
- W2910142656 hasConceptScore W2910142656C22019652 @default.
- W2910142656 hasConceptScore W2910142656C2524010 @default.
- W2910142656 hasConceptScore W2910142656C2779227376 @default.
- W2910142656 hasConceptScore W2910142656C33923547 @default.
- W2910142656 hasConceptScore W2910142656C38652104 @default.
- W2910142656 hasConceptScore W2910142656C41008148 @default.
- W2910142656 hasConceptScore W2910142656C42747912 @default.
- W2910142656 hasConceptScore W2910142656C50644808 @default.
- W2910142656 hasConceptScore W2910142656C62520636 @default.
- W2910142656 hasConceptScore W2910142656C78458016 @default.
- W2910142656 hasConceptScore W2910142656C86803240 @default.