method
active
method:ising-based-unsupervised-manifold-discovery-pipelineIsing-based unsupervised manifold discovery pipeline
Fits Ising over binarized codes, Louvain clustering, PCA spectral-gap validation.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Do Sparse Autoencoders Capture Concept Manifolds?introducesmentions
Frameworks (1)
framework
- Ising modelusesPrototypical spin system; baseline for phase transition arguments
Methods (2)
method
- Procedure for fitting Ising couplings.
- Partitions atoms by |J| into manifold groups.
Artifacts (1)
artifact
- goodfire-ai/sae-manifoldimplementsCode repository for the 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.
- Hypothesis-generating discovery.
- Technique used to fit M_h and M_y from data; enables manifold steering.
- Five-step pipeline: task definition, decomposition, description, description-validation loop, final validation.
- Post hoc recovery works synthetically.
- Related to point-based simplicial limit.
- Manifold geometry provides a practical blueprint for steering model behavior across diverse tasks and modalities.hypothesis0.721The generalizing predictive claim that manifold steering is a broadly applicable framework beyond the days-of-week case study.
- One-dimensional curved surface in internal activation space; the paper demonstrates alignment with behavior manifold.
- Cross-modality result from the full paper demonstrating that representation-behavior geometry alignment is not limited to language models.