concept
active
concept:curvature-basis-k-fac-eigenbasisCurvature basis (K-FAC eigenbasis)
Weight components ordered high to low curvature.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Approximates FIM as G⊗A to give dataset-average curvature eigenbasis.
Concepts (1)
concept
- Curvature massassociated_withProduct of eigenvalues used to select eigenvector pairs.
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.
- Basis for editing.
- Section 4.
- K-FAC eigenvector bands show sorted activation strength top10%>10-25%>25-50%>bottom50%, unlike SVDfinding0.743Supports curvature basis interpretability.
- Keeps top eigenvector pairs by curvature mass to suppress memorization.
- If curvature measures memorization, eigenvector bands will activate differently on memorized vs clean datahypothesis0.735Section 4 hypothesis.
- Vision disentanglement.
- Disentanglement evidence.
- Program research vector: evidence that neural networks discover underlying geometry of the world, supporting universality hypothesis.