claim
active
claim:non-linear-representations-are-generally-inefficient-for-neural-networksNon-linear representations are generally inefficient for neural networks
Intuition behind linear representations.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Questions (1)
question
- Open question.
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.
- Linear representation hypothesis: neural networks represent meaningful concepts as directions in their activation spaces.hypothesis0.796Foundation for interpreting features as linear directions.
- Interpretive claim about what linear DAS results actually tell us
- Architectural requirement from machine learning.
- Evidence that evolved machines share biological property of non-optimal modularity, blurring the distinction
- The paper's concluding summary statement asserting the deep interpretive significance of representation geometry.
- Recent work identifying cases where LLM features are not one-dimensionally linear, a caveat to the linearity hypothesis.
- Closing statement.
- Opening sentence framing the paper's core inquiry.