question
active
question:is-there-a-statistical-test-for-catching-superpositionIs there a statistical test for catching superposition?
Open question.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Concepts (1)
concept
- SuperpositioncitesPhenomenon where models represent more features than dimensions via almost-orthogonal directions.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Attentional features spread across multiple heads.
- Specific phrases or sequences memorized via binary features in superposition, enabling narrow pattern matching despite few neurons
- Core theoretical framework: neural networks represent more features than neurons by encoding features as directions in superposition
- Most pragmatically important question.
- Theoretical model of how neural networks encode more features than dimensions, informing linear representation work.
- Virtual weights between features forced into superposition by low-dimensional projections.
- Low-rank matrices W^TW.
- Adversarial example transfer may be predicted by correlated arrangement of features in superpositionhypothesis0.737Speculative.