finding
active
finding:k-fac-edit-on-vit-base-drops-memorized-train-accuracy-from-81-6-to-3-5-gt-accuracy-to-66-5-svd-58-9K-FAC edit on ViT-Base drops memorized train accuracy from 81.6% to 3.5%, GT accuracy to 66.5% (SVD 58.9%)
Table 2.
Source paper
extracted_from(2025) · Merullo, Jack · Vatsavaya, Srihita · Bushnaq, Lucius · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (2)
claim
- Basis for editing.
- Section 4.
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.
- Regularization effect.
- K-FAC edit reduces OLMo-7B Dolma strict accuracy from 99.9% to 3.4% (BSN 6.0%, SVD 3.0%)finding0.796Table 1.
- Vision disentanglement.
- Better transfer to unseen memorized data.
- Demonstrates VS-generated diverse negative examples improve downstream model performance in offline RL
- Best performing VS variant for math synthetic data generation with GPT-4.1
- Shows VS not only maintains but can slightly improve factual accuracy compared to baseline methods
- Strong empirical evidence that VS recovers pretraining distribution while direct prompting collapses