concept
active
concept:push-pull-geometry-in-representation-spacePush-Pull Geometry in Representation Space
Geometric property enforced by contrastive learning to ensure discriminative facet directions in SAE latent space
Neighborhood — ranked by edge-count
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.
- The bidirectional correspondence M_h ↔ M_y, indicating that geometry in representation is not incidental but causally shapes behavior.
- Author’s interpretive claim that the shared geometry is general and robust.
- A vector in activation space aligned with a behavioral concept; core object manipulated by RepE methods
- The idea that features are encoded as directions in activation space.
- One-dimensional curved surface in internal activation space; the paper demonstrates alignment with behavior manifold.
- The finding that steering along M_h yields M_y behavior, and optimizing for M_y paths recovers M_h trajectories.
- The central question of whether representational geometry implies corresponding computational structure
- Core finding: the structure models use internally (representations) is precisely reflected in their external behavior (outputs).