Matches in SemOpenAlex for { <https://semopenalex.org/work/W4385571067> ?p ?o ?g. }
Showing items 1 to 73 of
73
with 100 items per page.
- W4385571067 abstract "Despite the popularity of Shapley Values in explaining neural text classification models, computing them is prohibitive for large pretrained models due to a large number of model evaluations. In practice, Shapley Values are often estimated with a small number of stochastic model evaluations. However, we show that the estimated Shapley Values are sensitive to random seed choices – the top-ranked features often have little overlap across different seeds, especially on examples with longer input texts. This can only be mitigated by aggregating thousands of model evaluations, which on the other hand, induces substantial computational overheads. To mitigate the trade-off between stability and efficiency, we develop an amortized model that directly predicts each input feature’s Shapley Value without additional model evaluations. It is trained on a set of examples whose Shapley Values are estimated from a large number of model evaluations to ensure stability. Experimental results on two text classification datasets demonstrate that our amortized model estimates Shapley Values accurately with up to 60 times speedup compared to traditional methods. Further, our model does not suffer from stability issues as inference is deterministic. We release our code at https://github.com/yangalan123/Amortized-Interpretability." @default.
- W4385571067 created "2023-08-05" @default.
- W4385571067 creator A5000542080 @default.
- W4385571067 creator A5001906708 @default.
- W4385571067 creator A5025690046 @default.
- W4385571067 creator A5078193661 @default.
- W4385571067 creator A5086654042 @default.
- W4385571067 creator A5087096372 @default.
- W4385571067 date "2023-01-01" @default.
- W4385571067 modified "2023-09-23" @default.
- W4385571067 title "Efficient Shapley Values Estimation by Amortization for Text Classification" @default.
- W4385571067 doi "https://doi.org/10.18653/v1/2023.acl-long.483" @default.
- W4385571067 hasPublicationYear "2023" @default.
- W4385571067 type Work @default.
- W4385571067 citedByCount "0" @default.
- W4385571067 crossrefType "proceedings-article" @default.
- W4385571067 hasAuthorship W4385571067A5000542080 @default.
- W4385571067 hasAuthorship W4385571067A5001906708 @default.
- W4385571067 hasAuthorship W4385571067A5025690046 @default.
- W4385571067 hasAuthorship W4385571067A5078193661 @default.
- W4385571067 hasAuthorship W4385571067A5086654042 @default.
- W4385571067 hasAuthorship W4385571067A5087096372 @default.
- W4385571067 hasBestOaLocation W43855710671 @default.
- W4385571067 hasConcept C112972136 @default.
- W4385571067 hasConcept C11413529 @default.
- W4385571067 hasConcept C119857082 @default.
- W4385571067 hasConcept C138885662 @default.
- W4385571067 hasConcept C142417499 @default.
- W4385571067 hasConcept C144237770 @default.
- W4385571067 hasConcept C154945302 @default.
- W4385571067 hasConcept C162319229 @default.
- W4385571067 hasConcept C177142836 @default.
- W4385571067 hasConcept C199022921 @default.
- W4385571067 hasConcept C199360897 @default.
- W4385571067 hasConcept C2776401178 @default.
- W4385571067 hasConcept C2781067378 @default.
- W4385571067 hasConcept C33923547 @default.
- W4385571067 hasConcept C41008148 @default.
- W4385571067 hasConcept C41895202 @default.
- W4385571067 hasConcept C80444323 @default.
- W4385571067 hasConceptScore W4385571067C112972136 @default.
- W4385571067 hasConceptScore W4385571067C11413529 @default.
- W4385571067 hasConceptScore W4385571067C119857082 @default.
- W4385571067 hasConceptScore W4385571067C138885662 @default.
- W4385571067 hasConceptScore W4385571067C142417499 @default.
- W4385571067 hasConceptScore W4385571067C144237770 @default.
- W4385571067 hasConceptScore W4385571067C154945302 @default.
- W4385571067 hasConceptScore W4385571067C162319229 @default.
- W4385571067 hasConceptScore W4385571067C177142836 @default.
- W4385571067 hasConceptScore W4385571067C199022921 @default.
- W4385571067 hasConceptScore W4385571067C199360897 @default.
- W4385571067 hasConceptScore W4385571067C2776401178 @default.
- W4385571067 hasConceptScore W4385571067C2781067378 @default.
- W4385571067 hasConceptScore W4385571067C33923547 @default.
- W4385571067 hasConceptScore W4385571067C41008148 @default.
- W4385571067 hasConceptScore W4385571067C41895202 @default.
- W4385571067 hasConceptScore W4385571067C80444323 @default.
- W4385571067 hasLocation W43855710671 @default.
- W4385571067 hasOpenAccess W4385571067 @default.
- W4385571067 hasPrimaryLocation W43855710671 @default.
- W4385571067 hasRelatedWork W2966218717 @default.
- W4385571067 hasRelatedWork W3003792107 @default.
- W4385571067 hasRelatedWork W3006943036 @default.
- W4385571067 hasRelatedWork W3163334550 @default.
- W4385571067 hasRelatedWork W4200511449 @default.
- W4385571067 hasRelatedWork W4206534706 @default.
- W4385571067 hasRelatedWork W4229079080 @default.
- W4385571067 hasRelatedWork W4299487748 @default.
- W4385571067 hasRelatedWork W4300434475 @default.
- W4385571067 hasRelatedWork W4324106546 @default.
- W4385571067 isParatext "false" @default.
- W4385571067 isRetracted "false" @default.
- W4385571067 workType "article" @default.