method
active
method:sae-feature-clustering-leiden-spherical-k-meansSAE feature clustering (Leiden / spherical k-means)
Cluster features by activation statistics to get higher-level behaviors.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (2)
method
- Finds data clusters where response concept is chosen/rejected leaning.
- Finds response-delta clusters shifted for prompt clusters.
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.
- Extension of mechanistic interpretability findings to the metacognitive domain
- Unsupervised feature-finding method using cluster centroid difference as feature direction
- Core critique of sparse autoencoders: they break the geometric structure of representations, making it harder to see the big picture.
- Feature cluster examples from the UMAP.
- Claim that feature grounding enables interpretability metrics.
- Features act as local patch detectors.
- A promising property for interpretability analysis off-distribution.
- Quantifies how J-space aligned an SAE feature is.