finding
active
finding:some-gradient-based-attribution-methods-give-attributions-independent-of-model-and-data-generating-processSome gradient-based attribution methods give attributions independent of model and data-generating process
Adebayo et al. sanity checks.
Source paper
extracted_from(2025) · Lee Sharkey · Bilal I. Chughtai · Joshua Batson · Jack W. Lindsey +25
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Attribution methods are necessary for causal explanations but are often difficult to interpretsupportsOpen problem in attribution.
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 by APD; poor approximations of causal importance.
- Baseline method against which probe-based ranking is compared; more computationally expensive.
- Motivation.
- Backprop of logit-minus-mean-logit loss to surface causal intervention candidates, summed across token positions.
- Feature attribution (gradient-based) correlates 0.8 with ablation effects on the 'John' and 'Kobe' examples.finding0.796Validation of attribution as a fast proxy for causal importance.
- Measure causal importance of upstream variables.
- Linear classifier approach applied to model activations to identify which training datapoints caused undesired behaviors in post-training.
- Quantifies method limitations.