FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3014077729> ?p ?o ?g. }

• W3014077729 abstract "Understanding the inner workings of deep neural networks (DNNs) is essentialto provide trustworthy artificial intelligence techniques for practicalapplications. Existing studies typically involve linking semantic concepts tounits or layers of DNNs, but fail to explain the inference process. In thispaper, we introduce neural architecture disentanglement (NAD) to fill the gap.Specifically, NAD learns to disentangle a pre-trained DNN intosub-architectures according to independent tasks, forming information flowsthat describe the inference processes. We investigate whether, where, and howthe disentanglement occurs through experiments conducted with handcrafted andautomatically-searched network architectures, on both object-based andscene-based datasets. Based on the experimental results, we present three newfindings that provide fresh insights into the inner logic of DNNs. First, DNNscan be divided into sub-architectures for independent tasks. Second, deeperlayers do not always correspond to higher semantics. Third, the connection typein a DNN affects how the information flows across layers, leading to differentdisentanglement behaviors. With NAD, we further explain why DNNs sometimes givewrong predictions. Experimental results show that misclassified images have ahigh probability of being assigned to task sub-architectures similar to thecorrect ones. Code will be available at: https://github.com/hujiecpp/NAD." @default.
• W3014077729 created "2020-04-03" @default.
• W3014077729 creator A5014628588 @default.
• W3014077729 creator A5015317495 @default.
• W3014077729 creator A5016080094 @default.
• W3014077729 creator A5017100408 @default.
• W3014077729 creator A5039152554 @default.
• W3014077729 creator A5067413001 @default.
• W3014077729 creator A5071593501 @default.
• W3014077729 creator A5074759940 @default.
• W3014077729 creator A5082634513 @default.
• W3014077729 date "2020-03-30" @default.
• W3014077729 modified "2023-09-26" @default.
• W3014077729 title "Architecture Disentanglement for Deep Neural Networks" @default.
• W3014077729 cites W1686810756 @default.
• W3014077729 cites W1825675169 @default.
• W3014077729 cites W1849277567 @default.
• W3014077729 cites W2038721957 @default.
• W3014077729 cites W2108598243 @default.
• W3014077729 cites W2179423374 @default.
• W3014077729 cites W2194775991 @default.
• W3014077729 cites W2242818861 @default.
• W3014077729 cites W2253993278 @default.
• W3014077729 cites W2295107390 @default.
• W3014077729 cites W2418098761 @default.
• W3014077729 cites W2732026016 @default.
• W3014077729 cites W2753738274 @default.
• W3014077729 cites W2764024122 @default.
• W3014077729 cites W2769421449 @default.
• W3014077729 cites W2798626461 @default.
• W3014077729 cites W2806956262 @default.
• W3014077729 cites W2810075754 @default.
• W3014077729 cites W2883162033 @default.
• W3014077729 cites W2902227449 @default.
• W3014077729 cites W2945976633 @default.
• W3014077729 cites W2951004968 @default.
• W3014077729 cites W2962851944 @default.
• W3014077729 cites W2962858109 @default.
• W3014077729 cites W2962921245 @default.
• W3014077729 cites W2963203586 @default.
• W3014077729 cites W2963420658 @default.
• W3014077729 cites W2963446712 @default.
• W3014077729 cites W2963483561 @default.
• W3014077729 cites W2963749936 @default.
• W3014077729 cites W2964184826 @default.
• W3014077729 cites W2964303497 @default.
• W3014077729 cites W2964306570 @default.
• W3014077729 cites W2970971581 @default.
• W3014077729 cites W2978238437 @default.
• W3014077729 cites W2979454998 @default.
• W3014077729 cites W2979997102 @default.
• W3014077729 cites W3080865084 @default.
• W3014077729 cites W3104310207 @default.
• W3014077729 cites W3118608800 @default.
• W3014077729 cites W3137125108 @default.
• W3014077729 hasPublicationYear "2020" @default.
• W3014077729 type Work @default.
• W3014077729 sameAs 3014077729 @default.
• W3014077729 citedByCount "2" @default.
• W3014077729 countsByYear W30140777292020 @default.
• W3014077729 countsByYear W30140777292021 @default.
• W3014077729 crossrefType "posted-content" @default.
• W3014077729 hasAuthorship W3014077729A5014628588 @default.
• W3014077729 hasAuthorship W3014077729A5015317495 @default.
• W3014077729 hasAuthorship W3014077729A5016080094 @default.
• W3014077729 hasAuthorship W3014077729A5017100408 @default.
• W3014077729 hasAuthorship W3014077729A5039152554 @default.
• W3014077729 hasAuthorship W3014077729A5067413001 @default.
• W3014077729 hasAuthorship W3014077729A5071593501 @default.
• W3014077729 hasAuthorship W3014077729A5074759940 @default.
• W3014077729 hasAuthorship W3014077729A5082634513 @default.
• W3014077729 hasConcept C123657996 @default.
• W3014077729 hasConcept C142362112 @default.
• W3014077729 hasConcept C153349607 @default.
• W3014077729 hasConcept C154945302 @default.
• W3014077729 hasConcept C162324750 @default.
• W3014077729 hasConcept C177264268 @default.
• W3014077729 hasConcept C184337299 @default.
• W3014077729 hasConcept C187736073 @default.
• W3014077729 hasConcept C199360897 @default.
• W3014077729 hasConcept C2776214188 @default.
• W3014077729 hasConcept C2776760102 @default.
• W3014077729 hasConcept C2780451532 @default.
• W3014077729 hasConcept C2781238097 @default.
• W3014077729 hasConcept C2984842247 @default.
• W3014077729 hasConcept C41008148 @default.
• W3014077729 hasConcept C50644808 @default.
• W3014077729 hasConceptScore W3014077729C123657996 @default.
• W3014077729 hasConceptScore W3014077729C142362112 @default.
• W3014077729 hasConceptScore W3014077729C153349607 @default.
• W3014077729 hasConceptScore W3014077729C154945302 @default.
• W3014077729 hasConceptScore W3014077729C162324750 @default.
• W3014077729 hasConceptScore W3014077729C177264268 @default.
• W3014077729 hasConceptScore W3014077729C184337299 @default.
• W3014077729 hasConceptScore W3014077729C187736073 @default.
• W3014077729 hasConceptScore W3014077729C199360897 @default.
• W3014077729 hasConceptScore W3014077729C2776214188 @default.
• W3014077729 hasConceptScore W3014077729C2776760102 @default.
• W3014077729 hasConceptScore W3014077729C2780451532 @default.
• W3014077729 hasConceptScore W3014077729C2781238097 @default.

• next