Metraipolis › archive › dc_ce221c9bf98c

Intrinsic Structure: Spectral Identifiability for Mechanistic Interpretability — arXiv:2608.10172

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.10172
License (as archived)
CC0-1.0
Archived
2026-10-19T06:19:16.185Z
SHA-256 of the archived text
45bc4aa30b361c0bf010cb2853d009f23c41638d0eb6a1c14536c6dc91b89601
Stable URL of this page
https://metraipolis.com/documento/dc_ce221c9bf98c

How to cite

Intrinsic Structure: Spectral Identifiability for Mechanistic Interpretability — arXiv:2608.10172 — arXiv (en), https://arxiv.org/abs/2608.10172 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ce221c9bf98c (Metraipolis archive dc_ce221c9bf98c, sha256 45bc4aa30b361c0bf010cb2853d009f23c41638d0eb6a1c14536c6dc91b89601)
BibTeX
@misc{dcce221c9bf98c,
  title = {Intrinsic Structure: Spectral Identifiability for Mechanistic Interpretability — arXiv:2608.10172},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.10172},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ce221c9bf98c; sha256 45bc4aa30b361c0bf010cb2853d009f23c41638d0eb6a1c14536c6dc91b89601},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.10172. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Mechanistic interpretability explains models by identifying circuits inside them, but has no way to tell whether a circuit is a property of the model or an artifact of the method that found it. Sparse autoencoders illustrate the problem: different seeds and widths recover materially different features from the same activations, and no theory says whether that variability is incidental or structural. We put dictionary learning for interpretability on an identifiability footing. Treating the forward pass as a controlled dynamical system with depth as time and lifting it with the Koopman operator yields a finite linear realisation whose \emph{spectrum} is a coordinate-free property of the model. We prove the spectrum is recoverable from $M$ calibration samples at rate $M^{-1/2}$ up to permutation - to our knowledge the first identifiability theorem for a mechanistic-interpretability primitive, with a matching minimax lower bound, a median-of-means variant for heavy-tailed activations, and a dissociation theorem: whenever the realisation is non-normal, the directions carrying activation variance and the directions carrying information across depth cannot coincide. The identifiable object and the legible object are not the same object. On GPT-2 small, Gemma-2-2B and Qwen3-8B-Base the spectrum converges everywhere and attains the predicted exponent on Qwen3-8B-Base ($0.506 \pm 0.031$); shortfalls collapse onto one curve against each cell's sample threshold. Koopman modes beat random directions but lose to principal components on indirect-object identification, with the gap decaying $4.1\times$ in depth-distance, as the theorem predicts. The Koopman spectrum is an identifiable, model-intrinsic fingerprint with a stated error bar, not a legible decomposition.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-12 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.10172 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-19T06:11:45.067Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-08-12?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).