framework
active
framework:batchtopk-saeBatchTopK SAE
Evaluated architecture.
Neighborhood — ranked by edge-count
Papers (4)
paper
- Do Sparse Autoencoders Capture Concept Manifolds?citesmentions
Frameworks (2)
framework
- Matryoshka BatchTopK SAEextendsrelated_toHierarchical architecture that excels on disentanglement and scales positively.
- TopK SAEextendsEvaluated architecture (Gao et al.).
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.
- Sparsity activation used with k=200.
- Hybrid SAE jointly training turn-averaged inner and per-token outer partitions.
- Sparse dictionary learning method used to extract interpretable features from EEG transformer embeddings.
- Hybrid with separate turn-averaged and residual per-token SAEs.
- Evaluated architecture.
- Interpretable features extracted by sparse autoencoders used as steering targets in this study
- Unsupervised recovery of known manifolds.
- Buffer, shuffle, and write activations to avoid autocorrelation between batches.