Matches in SemOpenAlex for { <https://semopenalex.org/work/W3122807519> ?p ?o ?g. }
Showing items 1 to 85 of
85
with 100 items per page.
- W3122807519 abstract "The intuitive connection to robustness and convincing empirical evidence have made the flatness of the loss surface an attractive measure of generalizability for neural networks. Yet it suffers from various problems such as computational difficulties, reparametrization issues, and a growing concern that it may only be an epiphenomenon of optimization methods. We provide empirical evidence that under the cross-entropy loss once a neural network reaches a non-trivial training error, the flatness correlates (via Pearson Correlation Coefficient) well to the classification margins, which allows us to better reason about the concerns surrounding flatness. Our results lead to the practical recommendation that when assessing generalizability one should consider a margin-based measure instead, as it is computationally more efficient, provides further insight, and is highly correlated to flatness. We also use our insight to replace the misleading folklore that small-batch methods generalize better because they are able to escape sharp minima. Instead, we argue that large-batch methods did not have enough time to maximize margins and hence generalize worse." @default.
- W3122807519 created "2021-02-01" @default.
- W3122807519 creator A5042380467 @default.
- W3122807519 creator A5080848978 @default.
- W3122807519 creator A5083174192 @default.
- W3122807519 creator A5089059572 @default.
- W3122807519 date "2021-05-04" @default.
- W3122807519 modified "2023-09-27" @default.
- W3122807519 title "On Flat Minima, Large Margins and Generalizability" @default.
- W3122807519 hasPublicationYear "2021" @default.
- W3122807519 type Work @default.
- W3122807519 sameAs 3122807519 @default.
- W3122807519 citedByCount "0" @default.
- W3122807519 crossrefType "journal-article" @default.
- W3122807519 hasAuthorship W3122807519A5042380467 @default.
- W3122807519 hasAuthorship W3122807519A5080848978 @default.
- W3122807519 hasAuthorship W3122807519A5083174192 @default.
- W3122807519 hasAuthorship W3122807519A5089059572 @default.
- W3122807519 hasConcept C104317684 @default.
- W3122807519 hasConcept C105795698 @default.
- W3122807519 hasConcept C106301342 @default.
- W3122807519 hasConcept C11413529 @default.
- W3122807519 hasConcept C119857082 @default.
- W3122807519 hasConcept C121332964 @default.
- W3122807519 hasConcept C134306372 @default.
- W3122807519 hasConcept C154945302 @default.
- W3122807519 hasConcept C185592680 @default.
- W3122807519 hasConcept C186633575 @default.
- W3122807519 hasConcept C26405456 @default.
- W3122807519 hasConcept C27158222 @default.
- W3122807519 hasConcept C2778530986 @default.
- W3122807519 hasConcept C33923547 @default.
- W3122807519 hasConcept C41008148 @default.
- W3122807519 hasConcept C50644808 @default.
- W3122807519 hasConcept C55493867 @default.
- W3122807519 hasConcept C62520636 @default.
- W3122807519 hasConcept C63479239 @default.
- W3122807519 hasConcept C774472 @default.
- W3122807519 hasConceptScore W3122807519C104317684 @default.
- W3122807519 hasConceptScore W3122807519C105795698 @default.
- W3122807519 hasConceptScore W3122807519C106301342 @default.
- W3122807519 hasConceptScore W3122807519C11413529 @default.
- W3122807519 hasConceptScore W3122807519C119857082 @default.
- W3122807519 hasConceptScore W3122807519C121332964 @default.
- W3122807519 hasConceptScore W3122807519C134306372 @default.
- W3122807519 hasConceptScore W3122807519C154945302 @default.
- W3122807519 hasConceptScore W3122807519C185592680 @default.
- W3122807519 hasConceptScore W3122807519C186633575 @default.
- W3122807519 hasConceptScore W3122807519C26405456 @default.
- W3122807519 hasConceptScore W3122807519C27158222 @default.
- W3122807519 hasConceptScore W3122807519C2778530986 @default.
- W3122807519 hasConceptScore W3122807519C33923547 @default.
- W3122807519 hasConceptScore W3122807519C41008148 @default.
- W3122807519 hasConceptScore W3122807519C50644808 @default.
- W3122807519 hasConceptScore W3122807519C55493867 @default.
- W3122807519 hasConceptScore W3122807519C62520636 @default.
- W3122807519 hasConceptScore W3122807519C63479239 @default.
- W3122807519 hasConceptScore W3122807519C774472 @default.
- W3122807519 hasLocation W31228075191 @default.
- W3122807519 hasOpenAccess W3122807519 @default.
- W3122807519 hasPrimaryLocation W31228075191 @default.
- W3122807519 hasRelatedWork W2099968818 @default.
- W3122807519 hasRelatedWork W2407014344 @default.
- W3122807519 hasRelatedWork W2465753173 @default.
- W3122807519 hasRelatedWork W2549189808 @default.
- W3122807519 hasRelatedWork W2566079294 @default.
- W3122807519 hasRelatedWork W2596692027 @default.
- W3122807519 hasRelatedWork W2605372163 @default.
- W3122807519 hasRelatedWork W2731468224 @default.
- W3122807519 hasRelatedWork W2787999646 @default.
- W3122807519 hasRelatedWork W2966091527 @default.
- W3122807519 hasRelatedWork W2979200397 @default.
- W3122807519 hasRelatedWork W3004717130 @default.
- W3122807519 hasRelatedWork W3035352032 @default.
- W3122807519 hasRelatedWork W3093309139 @default.
- W3122807519 hasRelatedWork W3101700548 @default.
- W3122807519 hasRelatedWork W3110088573 @default.
- W3122807519 hasRelatedWork W3174732583 @default.
- W3122807519 hasRelatedWork W3206275068 @default.
- W3122807519 hasRelatedWork W3211468576 @default.
- W3122807519 hasRelatedWork W3212331305 @default.
- W3122807519 isParatext "false" @default.
- W3122807519 isRetracted "false" @default.
- W3122807519 magId "3122807519" @default.
- W3122807519 workType "article" @default.