framework
active
framework:feature-manifoldsFeature Manifolds
Hypothesized extension of superposition where features may be higher-dimensional manifolds rather than 1D directions
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (2)
concept
- Manifold tilingassociated_withLatents tile a manifold to reduce loss.
- Interpretability dark matterassociated_withStructure SAEs miss.
Frameworks (2)
framework
- Dual to discrete dictionary features; central frame.
- Prior model of superposition where features are discrete 1D objects repelling each other roughly evenly; paper argues this is incomplete
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.
- Discrete features and continuous manifolds are dual descriptions.
- A smooth, potentially curved surface in activation space along which activations vary according to a coherent semantic dimension.
- One-dimensional curved surface in internal activation space; the paper demonstrates alignment with behavior manifold.
- Features may not be strictly one-dimensional objects; higher-dimensional feature manifolds may exist in model representationshypothesis0.801Extension of superposition hypothesis to account for continuous families of features
- Discrete features and feature manifolds are dual descriptions of the same underlying objectclaim0.801Duality.
- Core future-work framing.
- An interpretability approach that describes representations in terms of entire curved manifolds rather than many small features.
- Research thread within About Blank concerning the structure and relational properties of neural network feature representations; covariance pooling tangentially supports this thread.