community
active
corpus: papers
community:neural-geometryNeural Geometry
15 members. Each node is clickable.
Loading graph…
Bridges (7)
Other communities that share members with this one — cross-cutting threads or papers that sit at the seam between two themes.
- LLM Introspection2 shared
- Manifold geometry for steering networks1 shared
- Christopher Alexander's 15 properties1 shared
- LLM Interpretability & Behavioral Analysis1 shared
- Manifold geometry for neural network steering1 shared
- Geometric structure in neural representations1 shared
- LLM interpretability & self-awareness1 shared
Concepts (11)
- Manifold SteeringCentral framework: steering neural networks by intervening along the curved manifold where a concept lives, rather than in straight lines through activation space.
- Activation-Behavior Manifold Isometry
- Base-10 Addition Mechanism in Llama
- Causal Abstraction over Cyclic Tasks
- Fourier Features for Numerical Computation
- Mountain Car Case StudyBenchmark demonstrating that car position encodes as a 1D manifold; linear interventions fail by crossing voids; following the geometric curve produces smooth control.
- Neural Geometries
- Parameter Decomposition (vs Activation)
- Pullback Geometry (Behavior-Aware Metric)
- SAE Critique (Manifold Shattering)
- VPD (adVersarial Parameter Decomposition)Core methodological framework introduced in this paper; decomposes weight matrices into rank-one interpretable subcomponents using adversarial ablations.