method
active
method:dimensions-per-feature-metric-d-m-w-f-2Dimensions-per-feature metric D*=m/||W||_F^2
Measures features learned.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionintroducesmentions
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.
- Measure of superposition in toy models.
- Resembles fractional quantum Hall effect.
- Fraction of a dimension given to a feature.
- Fraction of training tokens on which a given feature has nonzero activation; used as proxy metric for autoencoder quality
- Metric of features discovered.
- Dual to discrete dictionary features; central frame.
- Warm-up derivation.
- General principle supported tangentially by covariance pooling work; relates to feature co-occurrence structure.