framework
active
framework:additive-mixture-of-manifolds-modelAdditive Mixture of Manifolds model
Data-generating process: activation is sparse sum of low-dimensional manifold points.
Neighborhood — ranked by edge-count
Papers (2)
paper
- Do Sparse Autoencoders Capture Concept Manifolds?analogous_tointroduces
Frameworks (1)
framework
- The hypothesis that models internalize concepts as approximately linear directions in representation space; used to interpret MDS injection behavior
Questions (1)
question
- Acknowledged open issue in footnote.
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.
- Stated limitation.
- One-dimensional curved surface in internal activation space; the paper demonstrates alignment with behavior manifold.
- A smooth, potentially curved surface in activation space along which activations vary according to a coherent semantic dimension.
- Hypothesized extension of superposition where features may be higher-dimensional manifolds rather than 1D directions
- Technique used to fit M_h and M_y from data; enables manifold steering.
- Synthetic benchmark with known ground-truth manifolds.
- Combined causal model estimating next-token distributions.
- The type of manifold fitted to the cyclic concept structure in both activation and behavior space — a path along which steering moves the model.