concept
active
concept:superposition-in-residual-stream

Superposition in Residual Stream

The phenomenon where the residual stream communicates many more features than its dimensionality by encoding information across overlapping subspaces

Neighborhood — ranked by edge-count

Concepts (1)

concept
  • The finite dimensional capacity of the residual stream for storing and communicating information between layers; conceptualized as being under high demand

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.

  • Residual Streamconcept0.845
    Proposed pathway flowing through layers at each position; calculates K/V values that feed horizontal information flow.
  • The state in which a dialogue agent maintains multiple possible characters simultaneously, refined as the conversation proceeds
  • Technique to localize causally implicated hidden states by swapping residual stream activations between a true and false input and measuring downstream log-probability changes
  • Core activation intervention: add scaled vector to residual stream at layer l during completion
  • Superpositionconcept0.793
    Phenomenon where models represent more features than dimensions via almost-orthogonal directions.
  • The intermediate representations in transformer layers whose activations are patched and probed for truth information
  • Architectural observation enabling the entire mathematical framework; the residual stream is purely a sum of linear projections
  • Theoretical model of how neural networks encode more features than dimensions, informing linear representation work.