method
active
method:standard-l1-saeStandard L1 SAE
Architecture compared.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Do Sparse Autoencoders Capture Concept Manifolds?citesmentions
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.
- SAE variant trained on ResNet18.
- Trained on toy manifolds.
- Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.
- The objective function combining L2 reconstruction error and L1 penalty scaled by decoder norm, used to train the SAE.
- Key limitation that prevents tracing inter-layer dynamics or how steering propagates through model depth
- Design justification.
- Hybrid with separate turn-averaged and residual per-token SAEs.
- Standard baseline SAE operating on individual token activations.