Toy Models of Superposition
Toy Models of Superposition
About this item
Full title
Toy Models of Superposition
Author / Creator
Publisher
Ithaca: Cornell University Library, arXiv.org
Journal title
arXiv.org, 2022-09
Language
English
Formats
Publication information
Publisher
Ithaca: Cornell University Library, arXiv.org
Subjects
More information
Scope and Contents
Contents
Neural networks often pack many unrelated concepts into a single neuron - a puzzling phenomenon known as 'polysemanticity' which makes interpretability much more challenging. This paper provides a toy model where polysemanticity can be fully understood, arising as a result of models storing additional sparse features in "superposition." We demonstr...
Alternative Titles
Full title
Toy Models of Superposition
Authors, Artists and Contributors
Identifiers
Primary Identifiers
Record Identifier
TN_cdi_proquest_journals_2717199437
Permalink
https://devfeature-collection.sl.nsw.gov.au/record/TN_cdi_proquest_journals_2717199437
Other Identifiers
E-ISSN
2331-8422
How to access this item
Log in as a Library member
