method
active
method:automated-circuit-discovery-acdcAutomated Circuit DisCovery (ACDC)
Automates identifying task-relevant subgraphs.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Open Problems in Mechanistic Interpretabilityintroducesmentions
Frameworks (1)
framework
- Circuit discovery pipelineimplementsFive-step pipeline: task definition, decomposition, description, description-validation loop, final validation.
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.
- Interpretability technique for identifying functional sub-circuits in neural networks, supported by pyvene
- Will circuit discovery provide insights into arbitrary tasks, or only where tasks can be crisply defined?question0.738Circuit discovery scope.
- Lists issues: concept-based tasks, flawed decomposition, low faithfulness.
- Circuits active by default, e.g. can't-answer, suppressed by contrary evidence.
- Fine-grained approach to identifying specific network components responsible for reflection, mentioned as future direction.
- Advantage of DiffLogic CA over NCA — learned rules are pure binary logic circuits that can be visualized and analyzed
- Anthropic technique revealing computational pathways in Claude.
- A recurring, abstract pattern found in circuits (e.g., equivariance, unioning over cases), inspired by circuit motifs in systems biology