claim
active
claim:iterative-attacks-like-pgd-more-effectively-exploit-accumulated-interference-in-deeper-representationsIterative attacks like PGD more effectively exploit accumulated interference in deeper representations.
Interpretation of layer4 L0 spike.
Source paper
extracted_from(2025) · Gorton, Liv · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Contrast with PGD.
- Table 2.
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.
- If adversarial attacks exploit interference, they should activate more features than clean inputs.hypothesis0.767Prediction tested via L0.
- Key quote connecting path redundancy to interferometric information encoding.
- Interpretation of Figure 3.
- Performance gains over CAA in steering tasks.
- Iterative refinement substantially increases automated interpretability performance in early experimentsfinding0.732Early experiments toward interpreter agents.
- Epigraph from Feyerabend; warns against over-systematization while introducing a systematic framework—reflects the paper's tension between order and emergence.
- Explains discrepancy.
- Speculation.