A transformer's Koopman spectrum is provably recoverable from M calibration samples at the M^{-1/2} rate up to permutation, making it the first identified mechanistic-interpretability primitive.
Transcoders find interpretable LLM feature circuits
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Intrinsic Structure: Spectral Identifiability for Mechanistic Interpretability
A transformer's Koopman spectrum is provably recoverable from M calibration samples at the M^{-1/2} rate up to permutation, making it the first identified mechanistic-interpretability primitive.