claim
active
claim:the-large-harmful-in-utions-weight-implies-weight-superposition-since-it-would-have-no-reason-to-be-large-if-virtual-weights-were-trained-directlyThe large harmful IN→utions weight implies weight superposition since it would have no reason to be large if virtual weights were trained directly
Motivating example.
Source paper
extracted_fromNicholas L. Turner · Jeffrey Wu · Joshua Batson
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Largest IN Tokens→Logits weight votes 'utions', which never follows IN in training; Fisher effectiveness ~3 orders below the top weights.
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.
- Helpful and harmful weights are scattered across the entire range of virtual weight magnitudesfinding0.817IN Tokens→Logits shows weak VW-helpfulness relationship.
- Large virtual weights are not inherently misleading but not guaranteed to be helpful or effectiveclaim0.802Caution for practitioners.
- Discussion.
- Discussion implication.
- Interpretation of motivating example.
- Three regimes by effectiveness.
- Core mechanism statement.
- Superposition may be a major contributing factor, or even the primary cause, of adversarial examples.hypothesis0.735Hypothesis under test.