Matches in SemOpenAlex for { <https://semopenalex.org/work/W1690739335> ?p ?o ?g. }
- W1690739335 abstract "While depth tends to improve network performances, it also makes gradient-based training more difficult since deeper networks tend to be more non-linear. The recently proposed knowledge distillation approach is aimed at obtaining small and fast-to-execute models, and it has shown that a student network could imitate the soft output of a larger teacher network or ensemble of networks. In this paper, we extend this idea to allow the training of a student that is deeper and thinner than the teacher, using not only the outputs but also the intermediate representations learned by the teacher as hints to improve the training process and final performance of the student. Because the student intermediate hidden layer will generally be smaller than the teacher's intermediate hidden layer, additional parameters are introduced to map the student hidden layer to the prediction of the teacher hidden layer. This allows one to train deeper students that can generalize better or run faster, a trade-off that is controlled by the chosen student capacity. For example, on CIFAR-10, a deep student network with almost 10.4 times less parameters outperforms a larger, state-of-the-art teacher network." @default.
- W1690739335 created "2016-06-24" @default.
- W1690739335 creator A5032466547 @default.
- W1690739335 creator A5044826539 @default.
- W1690739335 creator A5057065873 @default.
- W1690739335 creator A5072194206 @default.
- W1690739335 creator A5080039924 @default.
- W1690739335 creator A5086198262 @default.
- W1690739335 date "2014-12-19" @default.
- W1690739335 modified "2023-10-06" @default.
- W1690739335 title "FitNets: Hints for Thin Deep Nets" @default.
- W1690739335 cites W1606347560 @default.
- W1690739335 cites W1686810756 @default.
- W1690739335 cites W1724438581 @default.
- W1690739335 cites W1814328102 @default.
- W1690739335 cites W1872489089 @default.
- W1690739335 cites W1994197834 @default.
- W1690739335 cites W2012885984 @default.
- W1690739335 cites W2110798204 @default.
- W1690739335 cites W2112796928 @default.
- W1690739335 cites W2119913432 @default.
- W1690739335 cites W2124509324 @default.
- W1690739335 cites W2134797427 @default.
- W1690739335 cites W2136922672 @default.
- W1690739335 cites W2146989110 @default.
- W1690739335 cites W2159291644 @default.
- W1690739335 cites W2163922914 @default.
- W1690739335 cites W2167215970 @default.
- W1690739335 cites W2168894214 @default.
- W1690739335 cites W2171312459 @default.
- W1690739335 cites W2294370754 @default.
- W1690739335 cites W2335728318 @default.
- W1690739335 cites W2950179405 @default.
- W1690739335 cites W2950967261 @default.
- W1690739335 cites W2951603627 @default.
- W1690739335 cites W2952020226 @default.
- W1690739335 cites W3037950864 @default.
- W1690739335 cites W3118608800 @default.
- W1690739335 hasPublicationYear "2014" @default.
- W1690739335 type Work @default.
- W1690739335 sameAs 1690739335 @default.
- W1690739335 citedByCount "371" @default.
- W1690739335 countsByYear W16907393352014 @default.
- W1690739335 countsByYear W16907393352015 @default.
- W1690739335 countsByYear W16907393352016 @default.
- W1690739335 countsByYear W16907393352017 @default.
- W1690739335 countsByYear W16907393352018 @default.
- W1690739335 countsByYear W16907393352019 @default.
- W1690739335 countsByYear W16907393352020 @default.
- W1690739335 countsByYear W16907393352021 @default.
- W1690739335 countsByYear W16907393352022 @default.
- W1690739335 countsByYear W16907393352023 @default.
- W1690739335 crossrefType "posted-content" @default.
- W1690739335 hasAuthorship W1690739335A5032466547 @default.
- W1690739335 hasAuthorship W1690739335A5044826539 @default.
- W1690739335 hasAuthorship W1690739335A5057065873 @default.
- W1690739335 hasAuthorship W1690739335A5072194206 @default.
- W1690739335 hasAuthorship W1690739335A5080039924 @default.
- W1690739335 hasAuthorship W1690739335A5086198262 @default.
- W1690739335 hasConcept C11413529 @default.
- W1690739335 hasConcept C121332964 @default.
- W1690739335 hasConcept C145420912 @default.
- W1690739335 hasConcept C153294291 @default.
- W1690739335 hasConcept C154945302 @default.
- W1690739335 hasConcept C178790620 @default.
- W1690739335 hasConcept C185592680 @default.
- W1690739335 hasConcept C199360897 @default.
- W1690739335 hasConcept C2777211547 @default.
- W1690739335 hasConcept C2779227376 @default.
- W1690739335 hasConcept C33923547 @default.
- W1690739335 hasConcept C41008148 @default.
- W1690739335 hasConcept C48103436 @default.
- W1690739335 hasConcept C98045186 @default.
- W1690739335 hasConceptScore W1690739335C11413529 @default.
- W1690739335 hasConceptScore W1690739335C121332964 @default.
- W1690739335 hasConceptScore W1690739335C145420912 @default.
- W1690739335 hasConceptScore W1690739335C153294291 @default.
- W1690739335 hasConceptScore W1690739335C154945302 @default.
- W1690739335 hasConceptScore W1690739335C178790620 @default.
- W1690739335 hasConceptScore W1690739335C185592680 @default.
- W1690739335 hasConceptScore W1690739335C199360897 @default.
- W1690739335 hasConceptScore W1690739335C2777211547 @default.
- W1690739335 hasConceptScore W1690739335C2779227376 @default.
- W1690739335 hasConceptScore W1690739335C33923547 @default.
- W1690739335 hasConceptScore W1690739335C41008148 @default.
- W1690739335 hasConceptScore W1690739335C48103436 @default.
- W1690739335 hasConceptScore W1690739335C98045186 @default.
- W1690739335 hasLocation W16907393351 @default.
- W1690739335 hasOpenAccess W1690739335 @default.
- W1690739335 hasPrimaryLocation W16907393351 @default.
- W1690739335 hasRelatedWork W1522301498 @default.
- W1690739335 hasRelatedWork W1686810756 @default.
- W1690739335 hasRelatedWork W1821462560 @default.
- W1690739335 hasRelatedWork W2095705004 @default.
- W1690739335 hasRelatedWork W2097117768 @default.
- W1690739335 hasRelatedWork W2108598243 @default.
- W1690739335 hasRelatedWork W2112796928 @default.
- W1690739335 hasRelatedWork W2117539524 @default.
- W1690739335 hasRelatedWork W2134797427 @default.
- W1690739335 hasRelatedWork W2163605009 @default.