claim
active
claim:mere-correlation-of-feature-directions-as-used-in-prior-work-need-not-suffice-to-find-manifoldsMere correlation of feature directions, as used in prior work, need not suffice to find manifolds
Contrast with Engels et al.
Source paper
extracted_from(2026) · Usha Bhalla · Thomas Fel · Can Rager · Sheridan Feucht +8
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Comparison of five similarity metrics.
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.
- Future-work perspective.
- Explicitly identified research gap: anecdotal evidence exists but rigorous characterization is absent
- Mechanism for null result.
- How can we reliably distinguish causal from merely correlated features when probing neural networks?question0.766Probing open question.
- Proposes that nonlinear geometric structure is superior to linear feature spaces for capturing semantic content.
- Features may not be strictly one-dimensional objects; higher-dimensional feature manifolds may exist in model representationshypothesis0.758Extension of superposition hypothesis to account for continuous families of features
- Implication for interpretability.