framework
active
framework:topk-sae

TopK SAE

Evaluated architecture (Gao et al.).

Neighborhood — ranked by edge-count

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

  • Sparse dictionary learning method used to extract interpretable features from EEG transformer embeddings.
  • Hierarchical architecture that excels on disentanglement and scales positively.
  • Nested SAEframework0.761
    Hybrid SAE jointly training turn-averaged inner and per-token outer partitions.
  • SAE featuresconcept0.760
    The individual, supposedly monosemantic directions learned by SAEs; argued here to fragment manifolds into disconnected pieces.
  • Internal concepts a model uses, extracted as human-interpretable components.
  • Matryoshka SAEmethod0.745
    Architecture compared.
  • Per-Token SAEframework0.741
    Standard baseline SAE operating on individual token activations.
  • Decoupled SAEframework0.739
    Hybrid with separate turn-averaged and residual per-token SAEs.