method
active
method:target-weighted-era-tweraTarget-weighted ERA (TWERA)
Target-weighted variant of ERA introduced in prior work.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Characterizing interference weights in a tiny language modelintroducesmentions
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.
- Prompt-independent feature interaction estimates for traversing harm features.
- ~27,000 point comparison.
- Prior coactivation-based proxy for effectiveness.
- Novel multiclass generalization of SCR.
- Baseline that minimizes sum of task losses with equal weights.
- Loss balancing based on learning speed.
- The ideal anatomical pattern (e.g., smiling face) that cells aim to achieve.
- Previous work viewing fast weights as alternative to LSTM; noted as prior context for TEM's fast weight memory.