framework
active
framework:attribution-based-parameter-decomposition-apdAttribution-based Parameter Decomposition (APD)
Decomposes parameters into faithful, minimal, simple components.
Neighborhood — ranked by edge-count
Papers (2)
paper
- Stochastic Parameter Decompositionintroducesmentions
Methods (1)
method
- Used by APD; poor approximations of causal importance.
Frameworks (2)
framework
- Linear parameter decomposition with stochastic ablation.
- Linear parameter decompositionimplementsFramework decomposing network parameters into sum of sparsely used vectors in parameter space.
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.
- Core methodological framework introduced in this paper; decomposes weight matrices into rank-one interpretable subcomponents using adversarial ablations.
- Core technique introduced in this paper for decomposing neural network weight matrices into mechanistically simple, interpretable rank-one subcomponents.
- Footnote explaining the name change from APD to VPD.
- Baseline method against which probe-based ranking is compared; more computationally expensive.
- Williams and Beer's decomposition of joint mutual information into unique, redundant, and synergistic components.
- Author interpretation.
- Some gradient-based attribution methods give attributions independent of model and data-generating processfinding0.736Adebayo et al. sanity checks.