Matches in SemOpenAlex for { <https://semopenalex.org/work/W4317464138> ?p ?o ?g. }
Showing items 1 to 69 of
69
with 100 items per page.
- W4317464138 abstract "How language models process complex input that requires multiple steps of inference is not well understood. Previous research has shown that information about intermediate values of these inputs can be extracted from the activations of the models, but it is unclear where that information is encoded and whether that information is indeed used during inference. We introduce a method for analyzing how a Transformer model processes these inputs by focusing on simple arithmetic problems and their intermediate values. To trace where information about intermediate values is encoded, we measure the correlation between intermediate values and the activations of the model using principal component analysis (PCA). Then, we perform a causal intervention by manipulating model weights. This intervention shows that the weights identified via tracing are not merely correlated with intermediate values, but causally related to model predictions. Our findings show that the model has a locality to certain intermediate values, and this is useful for enhancing the interpretability of the models." @default.
- W4317464138 created "2023-01-20" @default.
- W4317464138 creator A5044214762 @default.
- W4317464138 creator A5048821605 @default.
- W4317464138 creator A5060910700 @default.
- W4317464138 creator A5073408576 @default.
- W4317464138 date "2023-01-17" @default.
- W4317464138 modified "2023-10-14" @default.
- W4317464138 title "Tracing and Manipulating Intermediate Values in Neural Math Problem Solvers" @default.
- W4317464138 doi "https://doi.org/10.48550/arxiv.2301.06758" @default.
- W4317464138 hasPublicationYear "2023" @default.
- W4317464138 type Work @default.
- W4317464138 citedByCount "0" @default.
- W4317464138 crossrefType "posted-content" @default.
- W4317464138 hasAuthorship W4317464138A5044214762 @default.
- W4317464138 hasAuthorship W4317464138A5048821605 @default.
- W4317464138 hasAuthorship W4317464138A5060910700 @default.
- W4317464138 hasAuthorship W4317464138A5073408576 @default.
- W4317464138 hasBestOaLocation W43174641381 @default.
- W4317464138 hasConcept C111472728 @default.
- W4317464138 hasConcept C11413529 @default.
- W4317464138 hasConcept C124101348 @default.
- W4317464138 hasConcept C138673069 @default.
- W4317464138 hasConcept C138885662 @default.
- W4317464138 hasConcept C154945302 @default.
- W4317464138 hasConcept C199360897 @default.
- W4317464138 hasConcept C2776214188 @default.
- W4317464138 hasConcept C2779808786 @default.
- W4317464138 hasConcept C2780009758 @default.
- W4317464138 hasConcept C2780586882 @default.
- W4317464138 hasConcept C2781067378 @default.
- W4317464138 hasConcept C33923547 @default.
- W4317464138 hasConcept C41008148 @default.
- W4317464138 hasConcept C41895202 @default.
- W4317464138 hasConcept C75291252 @default.
- W4317464138 hasConcept C80444323 @default.
- W4317464138 hasConceptScore W4317464138C111472728 @default.
- W4317464138 hasConceptScore W4317464138C11413529 @default.
- W4317464138 hasConceptScore W4317464138C124101348 @default.
- W4317464138 hasConceptScore W4317464138C138673069 @default.
- W4317464138 hasConceptScore W4317464138C138885662 @default.
- W4317464138 hasConceptScore W4317464138C154945302 @default.
- W4317464138 hasConceptScore W4317464138C199360897 @default.
- W4317464138 hasConceptScore W4317464138C2776214188 @default.
- W4317464138 hasConceptScore W4317464138C2779808786 @default.
- W4317464138 hasConceptScore W4317464138C2780009758 @default.
- W4317464138 hasConceptScore W4317464138C2780586882 @default.
- W4317464138 hasConceptScore W4317464138C2781067378 @default.
- W4317464138 hasConceptScore W4317464138C33923547 @default.
- W4317464138 hasConceptScore W4317464138C41008148 @default.
- W4317464138 hasConceptScore W4317464138C41895202 @default.
- W4317464138 hasConceptScore W4317464138C75291252 @default.
- W4317464138 hasConceptScore W4317464138C80444323 @default.
- W4317464138 hasLocation W43174641381 @default.
- W4317464138 hasOpenAccess W4317464138 @default.
- W4317464138 hasPrimaryLocation W43174641381 @default.
- W4317464138 hasRelatedWork W1516360963 @default.
- W4317464138 hasRelatedWork W1821941829 @default.
- W4317464138 hasRelatedWork W1966731546 @default.
- W4317464138 hasRelatedWork W2369494673 @default.
- W4317464138 hasRelatedWork W2952027461 @default.
- W4317464138 hasRelatedWork W2964205372 @default.
- W4317464138 hasRelatedWork W3104846002 @default.
- W4317464138 hasRelatedWork W3115763169 @default.
- W4317464138 hasRelatedWork W4299281608 @default.
- W4317464138 hasRelatedWork W4313057118 @default.
- W4317464138 isParatext "false" @default.
- W4317464138 isRetracted "false" @default.
- W4317464138 workType "article" @default.