finding
active
finding:per-token-max-pooled-features-reach-95-0-on-10-way-matchingPer-token max-pooled features reach 95.0% on 10-way matching
Top discrimination on 2,000 holdout turns.
Source paper
extracted_from(2026) · Kevin Der · Harish Kamath · Thompson, Ben
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Figure 1.
- Alternative aggregation giving high discrimination.
- Figure 1.
- Aggregation of token-level activations.
- Turn-averaged features have highest embedding similarity 0.663 at k=3; per-token max lowest at 0.617finding0.770Consistent across k=1,2,3,5.
- Figure 1.
- Pairwise coverage.
- Figure 6.