method
active
method:mlp-gain-metricMLP gain metric
Measures MLP amplification of directions.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Crude measure of excess explanation of p over constituent dimensions, summed across features.
- Feed-forward neural network with hidden layers, capable of representing non-linearly separable functions.
- Gains on unsupervised proxy metrics do not reliably translate to better practical performanceclaim0.683Core thesis.
- Used to measure general capability preservation after steering interventions
- The sparse set of 28 neurons at layer 18 identified as responsible for Fourier feature computation across all cyclic tasks
- what metrics can reliably tell us if dictionary learning has successfully extracted high quality features?question0.665Central methodological gap: current metrics (loss, density histograms, manual inspection) are inadequate
- Measures features learned.
- Task signal in representation normalized to random baseline.