method
active
method:neuron-weight-stack-plotNeuron weight stack plot
Visualizes polysemantic neurons.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Frameworks (1)
framework
- Toy model with privileged basis.
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.
- Equations 11-12 measuring neuron role per input feature.
- Periodically reinitializing dead autoencoder neurons using high-loss data points to improve feature coverage
- Neuroscientific theory suggesting empathy is mediated by neural self-other overlap; motivates SOO approach
- Hebrew feature is effectively invisible in the neuron basis
- Method used to identify and partition the 28 MLP neurons into disjoint clusters by Fourier period
- Neuroscience view of neurons as unit; analog to early interpretability.
- Input-independent metric for stages of inference from Gurnee et al., applied to both feedforward and looped models
- Shows base64 feature is polysemantic at neuron level but monosemantic as learned feature