claim
active
claim:block-sparsity-is-the-prior-matched-to-a-generative-model-in-which-a-representation-is-a-sparse-sum-of-low-dimensional-manifoldsBlock sparsity is the prior matched to a generative model in which a representation is a sparse sum of low-dimensional manifolds.
Main theoretical claim, Prop. 1.
Source paper
extracted_from(2026) · Thomas Fel · Matthew Kowal · Mozes Jacobs · Dron Hazra +21
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (3)
finding
- Toy superposition recovery.
- SMixAE as published reaches per-block R² ≈0.71; with signed code and block projection ≈0.85.supportsRepair of SMixAE.
- Baseline failure.
Claims (2)
claim
- Rotation-invariance argument.
- Central framing.
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.
- Expected limitation.
- Support-term savings.
- Rationale for manifold steering.
- Caption of Figure 1 summarizing the central empirical finding of the paper
- Central thesis.
- Better bases may give sparsity.
- Main proposal.
- SAEs assume sparsity (number of concepts) is time-invariant, drawn i.i.d. from a fixed distribution.claim0.723Corollary 4.1.1.