method
active
method:max-pooling-of-per-token-featuresMax pooling of per-token features
Alternative aggregation giving high discrimination.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Per-Token SAEusesStandard baseline SAE operating on individual token activations.
Methods (1)
method
- Sum pooling of per-token featuresrelated_toAggregation of token-level activations.
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.
- Top discrimination on 2,000 holdout turns.
- §5.5.
- Structured number feature.
- Figure 7.
- Figure 1.
- Prediction from coverage results.
- Motivates nested SAE.
- Turn-averaged features have highest embedding similarity 0.663 at k=3; per-token max lowest at 0.617finding0.721Consistent across k=1,2,3,5.