method
active
method:sae-activation-harvest-shuffle-load-pipelineSAE activation harvest-shuffle-load pipeline
Buffer, shuffle, and write activations to avoid autocorrelation between batches.
Neighborhood — ranked by edge-count
Papers (1)
paper
Artifacts (1)
artifact
- ReLU SAE trained on residual stream layer 19 with MSE+L1 loss.
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.
- Token-level analysis of OTD and backtracking latent activations aligned at correction points across episodes
- Evaluated architecture.
- Out-of-distribution generalization of SAE features.
- Standard interpretability approach that VPD critiques and proposes an alternative to.
- Evaluated architecture.
- Hybrid with separate turn-averaged and residual per-token SAEs.
- Design property of the intervention method.
- Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.