concept
active
concept:off-manifold-regionsoff-manifold regions
Regions outside the natural activation manifold M_h, where linear steering ventures, leading to unnatural outputs.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Hypothesized extension of superposition where features may be higher-dimensional manifolds rather than 1D directions
- A smooth, potentially curved surface in activation space along which activations vary according to a coherent semantic dimension.
- The type of manifold fitted to the cyclic concept structure in both activation and behavior space — a path along which steering moves the model.
- An interpretability approach that describes representations in terms of entire curved manifolds rather than many small features.
- The low-dimensional geometric structure discovered in neural activation space; contrasted with linear/Euclidean geometry.
- Technique used to fit M_h and M_y from data; enables manifold steering.
- A smoothly varying lower-dimensional surface in activation space that captures a concept better than a straight linear direction.