finding
active
finding:at-k-64-bsf-concept-maps-are-roughly-five-times-smoother-total-variation-at-b-4-than-at-b-1At k=64, BSF concept maps are roughly five times smoother (total variation) at b=4 than at b=1.
Spatial coherence.
Source paper
extracted_from(2026) · Thomas Fel · Matthew Kowal · Mozes Jacobs · Dron Hazra +21
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Motivation for lifting atoms to blocks.
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.
- BSF codes surpass the SAE in best single-concept F1 on ImageNet-1k at every sparsity, gap widening with b.finding0.765Class selectivity.
- Concurrent work result showing emergent misalignment occurs in small models
- If BSFs capture concepts, moving within a block should manipulate behavior according to concept geometry.hypothesis0.749Motivates diffusion steering.
- Evidence for twisting.
- BSF codes improve on the basic probe for classification and approach it for segmentation and depth.finding0.737Linear probing.
- Figure 4 left.
- MDL result at 20% distortion.
- At S=200 variation across runs (TVD 0.0451) nearly equals sampling variation within a run (TVD 0.0442).finding0.730Supports multinomial model.