finding
active
finding:relu-hidden-layer-model-n-10-m-5-shifts-from-monosemantic-to-polysemantic-neurons-as-sparsity-increasesReLU hidden layer model (n=10, m=5) shifts from monosemantic to polysemantic neurons as sparsity increases
Neuron-level phase change.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Claims (2)
claim
- Toy model predictions are consistent with empirical observations on polysemanticity in real networkssupportsValidation; geometry and dynamics less certain.
- The phase diagram explains why neurons are sometimes monosemantic and sometimes polysemanticsupportsTheory contribution.
Hypotheses (1)
hypothesis
- Major prediction checked against InceptionV1.
Questions (1)
question
- Motivating 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.
- ReLU hidden layer model circumvents the ReLU by using positive biases or simulating biasfinding0.845Limitation of that toy model.
- Synthetic theoretical example showing pernicious divergence via hidden pathway activation
- Counter-example disproving that architectural sparsity alone can prevent polysemanticity
- Core demonstration.
- Toy model with privileged basis.
- Figure 10 discrepancy.
- Without superposition.
- Frontier ranking differs from other metrics.