concept
active
concept:featuresFeatures
Interpretable concept representations hypothesized as basic units of computation.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Frames model interpretability like biology, features as cells, tools as microscopes.
Claims (1)
claim
Hypotheses (1)
hypothesis
- Core framing hypothesis; analogy not to be taken literally.
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.
- Metaphor treating each system feature or function as a separate application that can be independently loaded and managed.
- Domain of techniques for constructing informative features from raw data; covariance pooling is a feature engineering method for token sequences.
- Extra feature representing unexplained error to make isomorphism exact.
- Frequency-weighted spectral value πkλkj determining which features are retained.
- Internal concepts a model uses, extracted as human-interpretable components.
- Research thread within About Blank concerning the structure and relational properties of neural network feature representations; covariance pooling tangentially supports this thread.
- Synthesizing inputs maximizing component activation.
- Proposed attention analog of sparse monosemantic features for decomposing attention layers.