Parameter subcomponents cleanly isolate true mechanisms of the model

Interpretive claim that the subcomponents correspond to real functional units.

Source paper

extracted_from

cimcWhitepaper

Neighborhood — ranked by edge-count

Findings (1)

finding

The VPD-based edit has similarly low off-target effects as uninterpretable fine-tuning methods
supports
Performance comparison showing subcomponent editing is comparable to fine-tuning in preserving off-target behavior.

Communities (4)

community

Mechanistic interpretability & model evaluation
members_of
Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
Mechanistic interpretability via parameter decomposition
members_of
Tracing information flow through weight matrices and attention heads using attribution graphs to identify causally important subcomponents in language models.
Causal parameter decomposition in neural networks
members_of
Isolating interpretable, role-specific model subcomponents through causal analysis and targeted edits to understand mechanistic function.
Causal parameter subcomponent isolation
members_of
Identifying model components causally responsible for specific behaviors while removable for irrelevant tasks

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

A good parameter subcomponent is causally important only for specific roles and can be removed from the model without hurting performance on irrelevant promptsclaim0.811
Definitional principle guiding VPD: subcomponents should encode narrow, targeted computational roles rather than distributed, multi-purpose functionality.
can we use these parameter subcomponents to perform clean, targeted changes?question0.787
Implicit question driving the editing experiment.
Parameter subcomponentconcept0.780
One of the simple rank-one matrices resulting from VPD that sums with others to reconstruct the original model weights and has a specific functional role.
are the resulting parameter subcomponents actually interpretable objects?question0.777
First question posed after applying VPD, investigating whether the subcomponents make sense.
Investigating the causal substructure of neural representations is necessary to avoid misidentifying data structures of simpler representations as abstract conceptsclaim0.742
Motivated by the finding that lexical entailment decomposes into word identities.
Direct model editing via parameter subcomponent modification—emoticon eye recognition altered to predict shocked faces with no retrainingfinding0.737
Demonstrated that VPD-discovered subcomponents encode true computational machinery by enabling targeted, predictable behavior changes without gradient-based training.
Decompose parameters, not activationsquote0.728
Core slogan encapsulating the paradigm shift of VPD.
VPD subcomponents are sparse, interpretable, and avoid feature splitting.claim0.728
Assertion about the qualitative advantages of VPD's rank-one decomposition.