framework
active
framework:linear-parameter-decompositionLinear parameter decomposition
Framework decomposing network parameters into sum of sparsely used vectors in parameter space.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Stochastic Parameter Decompositionintroducesmentions
Frameworks (3)
framework
- Stochastic Parameter Decomposition (SPD)implementsLinear parameter decomposition with stochastic ablation.
- Decomposes parameters into faithful, minimal, simple components.
- Three desiderata for parameter components defining mechanisms.
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.
- Claimed by Braun et al.; demonstrated empirically here.
- Correlative technique measuring the type of information encoded in distributed representations via linear predictability.
- The idea that features are encoded as directions in activation space.
- The core idea of decomposing weight matrices into components for interpretability.
- Used to evaluate representation quality across VTAB tasks
- IID mass-mean probing coincides with LDA when covariance is known; used to derive the corrected probe formula
- The central object of study — the idea that a concept like truth is encoded as a direction in the LLM's latent space