Matches in SemOpenAlex for { <https://semopenalex.org/work/W2972862635> ?p ?o ?g. }
- W2972862635 abstract "Recent research towards understanding neural networks probes models in a top-down manner, but is only able to identify model tendencies that are known a priori. We propose Susceptibility Identification through Fine-Tuning (SIFT), a novel abstractive method that uncovers a model's preferences without imposing any prior. By fine-tuning an autoencoder with the gradients from a fixed classifier, we are able to extract propensities that characterize different kinds of classifiers in a bottom-up manner. We further leverage the SIFT architecture to rephrase sentences in order to predict the opposing class of the ground truth label, uncovering potential artifacts encoded in the fixed classification model. We evaluate our method on three diverse tasks with four different models. We contrast the propensities of the models as well as reproduce artifacts reported in the literature." @default.
- W2972862635 created "2019-09-19" @default.
- W2972862635 creator A5024983536 @default.
- W2972862635 creator A5027450194 @default.
- W2972862635 creator A5037310413 @default.
- W2972862635 creator A5077791956 @default.
- W2972862635 date "2019-09-10" @default.
- W2972862635 modified "2023-09-27" @default.
- W2972862635 title "What do Deep Networks Like to Read" @default.
- W2972862635 cites W1832693441 @default.
- W2972862635 cites W1840435438 @default.
- W2972862635 cites W193524605 @default.
- W2972862635 cites W1986926488 @default.
- W2972862635 cites W2113459411 @default.
- W2972862635 cites W2130942839 @default.
- W2972862635 cites W2143017621 @default.
- W2972862635 cites W2250473257 @default.
- W2972862635 cites W2250539671 @default.
- W2972862635 cites W2251939518 @default.
- W2972862635 cites W2282821441 @default.
- W2972862635 cites W2296076036 @default.
- W2972862635 cites W2547875792 @default.
- W2972862635 cites W2766801974 @default.
- W2972862635 cites W2788403449 @default.
- W2972862635 cites W2798819017 @default.
- W2972862635 cites W2886424491 @default.
- W2972862635 cites W2888329843 @default.
- W2972862635 cites W2889436406 @default.
- W2972862635 cites W2906152891 @default.
- W2972862635 cites W2949227999 @default.
- W2972862635 cites W2951714314 @default.
- W2972862635 cites W2962736243 @default.
- W2972862635 cites W2962776659 @default.
- W2972862635 cites W2962843521 @default.
- W2972862635 cites W2963029978 @default.
- W2972862635 cites W2963078909 @default.
- W2972862635 cites W2963120843 @default.
- W2972862635 cites W2963233086 @default.
- W2972862635 cites W2963259903 @default.
- W2972862635 cites W2963343509 @default.
- W2972862635 cites W2963503967 @default.
- W2972862635 cites W2964159778 @default.
- W2972862635 cites W2964204621 @default.
- W2972862635 cites W2964287726 @default.
- W2972862635 cites W2964308564 @default.
- W2972862635 hasPublicationYear "2019" @default.
- W2972862635 type Work @default.
- W2972862635 sameAs 2972862635 @default.
- W2972862635 citedByCount "5" @default.
- W2972862635 countsByYear W29728626352019 @default.
- W2972862635 countsByYear W29728626352020 @default.
- W2972862635 countsByYear W29728626352021 @default.
- W2972862635 crossrefType "posted-content" @default.
- W2972862635 hasAuthorship W2972862635A5024983536 @default.
- W2972862635 hasAuthorship W2972862635A5027450194 @default.
- W2972862635 hasAuthorship W2972862635A5037310413 @default.
- W2972862635 hasAuthorship W2972862635A5077791956 @default.
- W2972862635 hasConcept C101738243 @default.
- W2972862635 hasConcept C111472728 @default.
- W2972862635 hasConcept C119857082 @default.
- W2972862635 hasConcept C138885662 @default.
- W2972862635 hasConcept C153083717 @default.
- W2972862635 hasConcept C153180895 @default.
- W2972862635 hasConcept C154945302 @default.
- W2972862635 hasConcept C2776502983 @default.
- W2972862635 hasConcept C2984842247 @default.
- W2972862635 hasConcept C41008148 @default.
- W2972862635 hasConcept C50644808 @default.
- W2972862635 hasConcept C52622490 @default.
- W2972862635 hasConcept C61265191 @default.
- W2972862635 hasConcept C75553542 @default.
- W2972862635 hasConcept C95623464 @default.
- W2972862635 hasConceptScore W2972862635C101738243 @default.
- W2972862635 hasConceptScore W2972862635C111472728 @default.
- W2972862635 hasConceptScore W2972862635C119857082 @default.
- W2972862635 hasConceptScore W2972862635C138885662 @default.
- W2972862635 hasConceptScore W2972862635C153083717 @default.
- W2972862635 hasConceptScore W2972862635C153180895 @default.
- W2972862635 hasConceptScore W2972862635C154945302 @default.
- W2972862635 hasConceptScore W2972862635C2776502983 @default.
- W2972862635 hasConceptScore W2972862635C2984842247 @default.
- W2972862635 hasConceptScore W2972862635C41008148 @default.
- W2972862635 hasConceptScore W2972862635C50644808 @default.
- W2972862635 hasConceptScore W2972862635C52622490 @default.
- W2972862635 hasConceptScore W2972862635C61265191 @default.
- W2972862635 hasConceptScore W2972862635C75553542 @default.
- W2972862635 hasConceptScore W2972862635C95623464 @default.
- W2972862635 hasLocation W29728626351 @default.
- W2972862635 hasOpenAccess W2972862635 @default.
- W2972862635 hasPrimaryLocation W29728626351 @default.
- W2972862635 hasRelatedWork W2105728138 @default.
- W2972862635 hasRelatedWork W2153332911 @default.
- W2972862635 hasRelatedWork W2529194139 @default.
- W2972862635 hasRelatedWork W2578212641 @default.
- W2972862635 hasRelatedWork W2770877945 @default.
- W2972862635 hasRelatedWork W2794887779 @default.
- W2972862635 hasRelatedWork W2920808855 @default.
- W2972862635 hasRelatedWork W2963341956 @default.
- W2972862635 hasRelatedWork W2963586005 @default.
- W2972862635 hasRelatedWork W2977603687 @default.