Circuit Motif

A recurring, abstract pattern found in circuits (e.g., equivariance, unioning over cases), inspired by circuit motifs in systems biology

Neighborhood — ranked by edge-count

paper

thinker

Gabriel Goh
studies
Reported multimodal neurons in vision models including region neurons and person detector neurons

concept

Superposition
extends
Phenomenon where models represent more features than dimensions via almost-orthogonal directions.
Equivariant Circuit
extends
A circuit exhibiting symmetry such that the weights rotate with the orientation of the feature they detect
Unioning Over Cases
extends
A circuit motif where the network separately detects multiple cases and takes a union to create invariant neurons; exemplified by the oriented dog head detection circuit

book

An Introduction to Systems Biology: Design Principles of Biological Circuits
cites
Alon's systems biology text that provides the concept of circuit motifs adopted by the circuits agenda

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.

motifconcept0.837
The systematically repeating unit in a pattern
Circuits Frameworkframework0.781
Mechanistic interpretability framework for understanding neural network computation as circuits of features
Circuits Threadframework0.778
An open scientific collaboration hosted on Distill slack studying the inner workings of neural networks via zoomed-in mechanistic investigation
Minimal Circuit Descriptionconcept0.771
Hypothesis that discretization finds minimum-size circuits equivalent to minimal algorithmic descriptions of patterns
Circuit Findingmethod0.762
Interpretability technique for identifying functional sub-circuits in neural networks, supported by pyvene
Circuit Analysisconcept0.758
Fine-grained approach to identifying specific network components responsible for reflection, mentioned as future direction.
Circuit Interpretabilityconcept0.747
Advantage of DiffLogic CA over NCA — learned rules are pure binary logic circuits that can be visualized and analyzed
character-circuit overlapconcept0.740
The overlap between circuits used for self-model and for modeling fictional characters; self-character is represented differently from fiction.