Matches in SemOpenAlex for { <https://semopenalex.org/work/W2912971006> ?p ?o ?g. }
Showing items 1 to 70 of
70
with 100 items per page.
- W2912971006 abstract "Softmax is a standard final layer used in Neural Nets (NNs) to summarize information encoded in the trained NN and return a prediction. However, Softmax leverages only a subset of the class-specific structure encoded in the trained model and ignores potentially valuable information: During training, models encode an array $D$ of class response distributions, where $D_{ij}$ is the distribution of the $j^{th}$ pre-Softmax readout neuron's responses to the $i^{th}$ class. Given a test sample, Softmax implicitly uses only the row of this array $D$ that corresponds to the readout neurons' responses to the sample's true class. Leveraging more of this array $D$ can improve classifier accuracy, because the likelihoods of two competing classes can be encoded in other rows of $D$. To explore this potential resource, we develop a hybrid classifier (Softmax-Pooling Hybrid, $SPH$) that uses Softmax on high-scoring samples, but on low-scoring samples uses a log-likelihood method that pools the information from the full array $D$. We apply $SPH$ to models trained on a vectorized MNIST dataset to varying levels of accuracy. $SPH$ replaces only the final Softmax layer in the trained NN, at test time only. All training is the same as for Softmax. Because the pooling classifier performs better than Softmax on low-scoring samples, $SPH$ reduces test set error by 6% to 23%, using the exact same trained model, whatever the baseline Softmax accuracy. This reduction in error reflects hidden capacity of the trained NN that is left unused by Softmax." @default.
- W2912971006 created "2019-02-21" @default.
- W2912971006 creator A5043283568 @default.
- W2912971006 creator A5083450863 @default.
- W2912971006 creator A5087127737 @default.
- W2912971006 date "2019-01-26" @default.
- W2912971006 modified "2023-09-27" @default.
- W2912971006 title "Money on the Table: Statistical information ignored by Softmax can improve classifier accuracy." @default.
- W2912971006 cites W1503398984 @default.
- W2912971006 cites W2032026767 @default.
- W2912971006 cites W2163614729 @default.
- W2912971006 cites W2964113296 @default.
- W2912971006 cites W2964212410 @default.
- W2912971006 cites W3118608800 @default.
- W2912971006 hasPublicationYear "2019" @default.
- W2912971006 type Work @default.
- W2912971006 sameAs 2912971006 @default.
- W2912971006 citedByCount "1" @default.
- W2912971006 countsByYear W29129710062020 @default.
- W2912971006 crossrefType "posted-content" @default.
- W2912971006 hasAuthorship W2912971006A5043283568 @default.
- W2912971006 hasAuthorship W2912971006A5083450863 @default.
- W2912971006 hasAuthorship W2912971006A5087127737 @default.
- W2912971006 hasConcept C119857082 @default.
- W2912971006 hasConcept C153180895 @default.
- W2912971006 hasConcept C154945302 @default.
- W2912971006 hasConcept C169903167 @default.
- W2912971006 hasConcept C188441871 @default.
- W2912971006 hasConcept C190502265 @default.
- W2912971006 hasConcept C41008148 @default.
- W2912971006 hasConcept C50644808 @default.
- W2912971006 hasConcept C70437156 @default.
- W2912971006 hasConcept C95623464 @default.
- W2912971006 hasConceptScore W2912971006C119857082 @default.
- W2912971006 hasConceptScore W2912971006C153180895 @default.
- W2912971006 hasConceptScore W2912971006C154945302 @default.
- W2912971006 hasConceptScore W2912971006C169903167 @default.
- W2912971006 hasConceptScore W2912971006C188441871 @default.
- W2912971006 hasConceptScore W2912971006C190502265 @default.
- W2912971006 hasConceptScore W2912971006C41008148 @default.
- W2912971006 hasConceptScore W2912971006C50644808 @default.
- W2912971006 hasConceptScore W2912971006C70437156 @default.
- W2912971006 hasConceptScore W2912971006C95623464 @default.
- W2912971006 hasLocation W29129710061 @default.
- W2912971006 hasOpenAccess W2912971006 @default.
- W2912971006 hasPrimaryLocation W29129710061 @default.
- W2912971006 hasRelatedWork W141827852 @default.
- W2912971006 hasRelatedWork W1517669128 @default.
- W2912971006 hasRelatedWork W1554466213 @default.
- W2912971006 hasRelatedWork W1596950930 @default.
- W2912971006 hasRelatedWork W1911848126 @default.
- W2912971006 hasRelatedWork W2014000357 @default.
- W2912971006 hasRelatedWork W2383218674 @default.
- W2912971006 hasRelatedWork W2423124209 @default.
- W2912971006 hasRelatedWork W2469841107 @default.
- W2912971006 hasRelatedWork W2527527138 @default.
- W2912971006 hasRelatedWork W2784175621 @default.
- W2912971006 hasRelatedWork W2794642338 @default.
- W2912971006 hasRelatedWork W2802447513 @default.
- W2912971006 hasRelatedWork W2903350505 @default.
- W2912971006 hasRelatedWork W2911881000 @default.
- W2912971006 hasRelatedWork W2942657220 @default.
- W2912971006 hasRelatedWork W2963542578 @default.
- W2912971006 hasRelatedWork W3092527263 @default.
- W2912971006 hasRelatedWork W3173685130 @default.
- W2912971006 hasRelatedWork W2865836616 @default.
- W2912971006 isParatext "false" @default.
- W2912971006 isRetracted "false" @default.
- W2912971006 magId "2912971006" @default.
- W2912971006 workType "article" @default.