method
active
method:context-projection-variance-explained-analysisContext projection variance explained analysis
Projects token representation onto context subspace.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Time-dependent correlation structure in activations.
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.
- Efficient estimation strategy for projection difference using a random subset of training data to reduce computational cost
- Up to 80% of variance of x_t is explained by a 500-token preceding context on Pile samples.finding0.737Fig. 2d,h.
- Metric for pre-finetuning data screening: difference between average projection of training responses and base model natural responses onto a persona direction
- Reports phase-like breakpoints and geometry changes as context scales; UCCT provides measurable predictor
- The process by which a form responds to its unique local conditions, leading to uniqueness.
- Feature that activates across all tokens within a specific context (e.g., DNA sequences, base64 strings)
- Graph structure inferred purely in-context.