concept
active
concept:feature-disentanglementFeature Disentanglement
Independent concepts are encoded by separate latents.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (3)
method
- Novel metric extending SHIFT via zero ablation.
- RAVEL evaluationaboutDisentanglement metric via latent interventions with Cause and Isolation metrics.
- Novel multiclass generalization of SCR.
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.
- Related research agenda seeking representations that separate conceptually distinct factors; contrasted with superposition approach
- CIMC's characterization of part of the solution to the Hard Problem: insight into the structural necessities of phenomenal representation
- Adjacent ML literature on separating independent factors of variation; related to but distinct from the polysemanticity problem studied here
- Phenomenon where a feature in a small SAE splits into multiple finer features in a larger SAE.
- Explains inverse scaling.
- SAE-derived representation space where individual dimensions activate for distinct concepts, enabling cleaner steering
- Problems suggesting sparsity breaks down as interpretability proxy.
- Property that features activate on only a small fraction of inputs; enables compressed sensing and is what allows superposition to work