claim
active
claim:spd-is-more-scalable-and-robust-to-hyperparameters-than-apdSPD is more scalable and robust to hyperparameters than APD
Central claim.
Source paper
extracted_from(2025) · Lucius Bushnaq · Dan Braun · Lee Sharkey
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (6)
finding
- SPD finds 102 W_in components and a rank-51 W_out component spanning all three layers in the 3-layer modelassociated_withsupportsAPD struggled beyond two layers.
- Ground-truth columns recovered with negligible shrinkage.
- New models harder for APD.
- Recovery with hidden identity.
- Robustness.
- Figure 10 discrepancy.
Hypotheses (1)
hypothesis
- Future direction.
Questions (2)
question
- Open problem.
- Open problem.
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.
- Author interpretation.
- SPD's robustness comes from trained importance functions, gradient flow, no simplicity loss, and direct causal importancehypothesis0.792Four hypothesized sources.
- SPD avoids shrinkage.
- SPD avoids shrinkage because importance norm penalizes mask probability, not singular valuesclaim0.776Mechanism.
- Source of robustness.
- Section 4.
- Motivation.
- SPD finds 100 W_in components and a single rank-50 W_out component spanning both layers in the 2-layer modelfinding0.751Multi-layer mechanisms.