concept
active
concept:novel-componentNovel component
Fast-moving residual orthogonal to context.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Proposed objective decomposing activations into predictable and novel components.
Methods (1)
method
- Activation-triggered examples plus direct logit attribution.
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.
- New term in expected free energy representing information gain about the likelihood mapping; drives ignorance resolution
- Slow-moving context-derived part of activation.
- Table 3.
- Table 3.
- Residuals of the novel component are assumed i.i.d. over time, which is still an assumption.claim0.703Acknowledged limitation.
- Expected KL divergence between updated and current posterior over model parameters.
- Linear-model solution that trades off with superposition.