concept
active
concept:attention-head-superpositionAttention head superposition
Possibility that functional units are distributed across heads.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Attention superpositionrelated_toAttentional features spread across multiple heads.
Questions (1)
question
- Distributed functional units.
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.
- Transformer attention heads that could be recruited to extract different kinds of information (text vs. thoughts).
- Analysis measuring whether each attention head's maximum attention increase points to the correct injected sentence
- Phenomenon where models represent more features than dimensions via almost-orthogonal directions.
- Mathematical equivalence enabling independent analysis of each attention head
- Specific phrases or sequences memorized via binary features in superposition, enabling narrow pattern matching despite few neurons
- The composition of two attention heads via V-composition, forming a new entity with its own attention pattern A^h2 * A^h1 and OV matrix W_OV^h2 * W_OV^h1
- Suspected but no confirmed example in real LLM.
- A head can implement multiple behaviors via high-rank output without superposition.