SPD uses stochastic masking and a learned causal importance function to decompose neural network parameters into sparsely active rank-one subcomponents, recovering ground-truth mechanisms in toy models where APD struggled.
Towards monosemanticity: Decomposing language models with dictionary learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Stochastic Parameter Decomposition
SPD uses stochastic masking and a learned causal importance function to decompose neural network parameters into sparsely active rank-one subcomponents, recovering ground-truth mechanisms in toy models where APD struggled.