finding
active
finding:moderate-l0-of-50-150-balances-metrics-lower-l0-helps-interpretability-higher-l0-improves-reconstruction-ravel-tpp-and-absorptionModerate L0 of 50-150 balances metrics; lower L0 helps interpretability; higher L0 improves reconstruction, RAVEL, TPP and absorption
Task-dependent optimal sparsity.
Source paper
extracted_from(2025) · Adam Karvonen · Can Rager · Johnny Lin · Curt Tigges +11
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Sparsity trade-off.
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.
- Caveat on metric.
- Unexpected result.
- Test-time probe metrics.
- Demonstrates that theoretical credence weighting can drastically lower the assessed probability for the same evidence.
- Fig. 22.
- Mixed LR probe achieves 0.386 to 0.826 accuracy against 6.7% chance, but fails on natural documentsfinding0.746No evidence of chronic emotion representation.
- Theoretical interpretation of antipodal alignment and misalignment phenomena in PCA visualizations
- Core cross-modal empirical result: larger and better language models align better with vision models