method
active
method:contrastive-feature-retrieval-pipelineContrastive Feature Retrieval Pipeline
A pipeline employing controlled semantic oppositions to distill monosemantic functional features from sparse activation spaces.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- The main framework proposed for retrieving and steering high-order semantic features in LLMs via sparse autoencoders.
Concepts (1)
concept
- Technique using semantically opposite prompts to contrast and identify features.
Methods (2)
method
- Validation method that uses text generation to confirm semantic control.
- Component of the contrastive retrieval pipeline analyzing activation statistics.
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.
- Rank features by activation ratio on positive vs negative rollouts.
- Method for obtaining concept vectors by subtracting activations from two contrasting prompts.
- Unsupervised probing method from Burns et al. 2023 that identifies directions along which contrast pair representations are far apart
- Synthesizing inputs maximizing component activation.
- Method comparing brain activity in conscious vs. unconscious conditions.
- Scores explanations by distinguishing activating from distractor samples; weakly penalises polysemanticity.
- Activation-triggered examples plus direct logit attribution.
- Supervised learning framework where system learns by observing contrast between current response and nudged improved response; requires weak additional forces from supervisor