Metraipolis › archive › dc_0115dcf59ec7

High-probability guarantees for linear accessibility in feature superposition — arXiv:2609.09556

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.09556
License (as archived)
CC0-1.0
Archived
2026-10-16T07:25:44.315Z
SHA-256 of the archived text
b7c0260cbde9b33039658628265eb2cabee57b22bc715da5cbaa819626c537fb
Stable URL of this page
https://metraipolis.com/documento/dc_0115dcf59ec7

How to cite

High-probability guarantees for linear accessibility in feature superposition — arXiv:2609.09556 — arXiv (en), https://arxiv.org/abs/2609.09556 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0115dcf59ec7 (Metraipolis archive dc_0115dcf59ec7, sha256 b7c0260cbde9b33039658628265eb2cabee57b22bc715da5cbaa819626c537fb)
BibTeX
@misc{dc0115dcf59ec7,
  title = {High-probability guarantees for linear accessibility in feature superposition — arXiv:2609.09556},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.09556},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0115dcf59ec7; sha256 b7c0260cbde9b33039658628265eb2cabee57b22bc715da5cbaa819626c537fb},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.09556. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Neural networks can leverage feature superposition to encode more concepts than dimensions, but cross-feature interference constrains the linear accessibility of simultaneously active features. By framing linear accessibility as a compressed sensing problem, we derive high-probability bounds for fixed supports under subgaussian noise, proving the sufficient dimension scales linearly ($d=O_{\varepsilon}(k \log m)$) rather than prior worst-case quadratic limits. We characterize the asymmetry between active and inactive interference and the trade-off between interference and observation-noise budgets. We then validate these bounds across system parameters through Gaussian-tail approximations. We also introduce IHT-SAE, which uses learned iterative refinement to improve feature recovery beyond the limits of linear availability. These results quantify the geometric constraints of the linear representation hypothesis, providing a framework for evaluating sparse autoencoders, compositional generalization, and neural interpretability.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-10 in cs.AI).
Categorie arXiv: Machine Learning (stat.ML) ; Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG); Probability (math.PR).
Pagina del lavoro: https://arxiv.org/abs/2609.09556 — 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 2026-10-16T07:00:16.321Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-10?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).