method
active
method:relu-sae-with-l1-penaltyReLU SAE with L1 penalty
Trained on toy manifolds.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- L1 ReLU SAErelated_toSAE variant trained on ResNet18.
Datasets (1)
dataset
- Training data for L(n) curves.
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.
- Baseline SAE architecture.
- Design justification.
- Architecture compared.
- The objective function combining L2 reconstruction error and L1 penalty scaled by decoder norm, used to train the SAE.
- Exception to ReLU underperformance.
- Fig 1.
- Unlearning not discriminative.
- ReLU underperforms.