method
active
method:elhage-analytic-optimal-feature-attackElhage analytic optimal feature attack
Analytically constructed per-feature attack.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Used instead of gradient descent attacks due to gradient masking.
- Appendix replication.
- Motivates nested SAE.
- If adversarial attacks exploit interference, they should activate more features than clean inputs.hypothesis0.698Prediction tested via L0.
- Compares features across SAEs.
- Core concern.
- Reduces superposition but needs 80% L2 norm.
- Method of optimizing activation-space interventions to produce behavioral paths along M_y, then measuring whether the resulting activation trajectories trace M_h curvature