claim
active
claim:the-curvature-basis-is-more-interpretable-and-accurate-than-svd-for-describing-the-memorized-non-memorized-spectrumThe curvature basis is more interpretable and accurate than SVD for describing the memorized/non-memorized spectrum
Section 4.
Source paper
extracted_from(2025) · Merullo, Jack · Vatsavaya, Srihita · Bushnaq, Lucius · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Table 2.
- K-FAC eigenvector bands show sorted activation strength top10%>10-25%>25-50%>bottom50%, unlike SVDsupportsSupports curvature basis interpretability.
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.
- If curvature measures memorization, eigenvector bands will activate differently on memorized vs clean datahypothesis0.819Section 4 hypothesis.
- Consistent with prior work.
- Weight components ordered high to low curvature.
- Automated interpretability and specificity ratings show SAE features are clearer than MLP neurons.
- Central claim.
- Quantitative comparison supporting SAE utility.
- Section 6.
- Load-bearing theoretical claim providing the conceptual foundation for DAS.