Metraipolis › archive › dc_7adf6d970943

Representation Redistribution in Sparse Autoencoders — arXiv:2610.21240

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.21240
License (as archived)
CC0-1.0
Archived
2026-10-16T09:14:17.179Z
SHA-256 of the archived text
6f57f280234f52570ad62d3dfd2d557d532feecc9d82c23cdfe926f25e3fcddc
Stable URL of this page
https://metraipolis.com/documento/dc_7adf6d970943

How to cite

Representation Redistribution in Sparse Autoencoders — arXiv:2610.21240 — arXiv (en), https://arxiv.org/abs/2610.21240 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7adf6d970943 (Metraipolis archive dc_7adf6d970943, sha256 6f57f280234f52570ad62d3dfd2d557d532feecc9d82c23cdfe926f25e3fcddc)
BibTeX
@misc{dc7adf6d970943,
  title = {Representation Redistribution in Sparse Autoencoders — arXiv:2610.21240},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.21240},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7adf6d970943; sha256 6f57f280234f52570ad62d3dfd2d557d532feecc9d82c23cdfe926f25e3fcddc},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.21240. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Sparse autoencoders (SAEs) decompose language model activations into sparse features, but prompt transformations designed to preserve the underlying request can redistribute activity across them. A feature change observed on one transformed prompt need not recur, and even a recurring change need not affect a downstream computation. We introduce the redistribution sensitivity score (RSS), which identifies large feature responses that recur in a consistent direction, and evaluate the resulting supports on variants excluded from selection. Across 27 pretrained SAEs spanning eight language models, the tested prompt transformations produce a reproducible redistribution profile on these excluded variants. We then separate recurring representation change from signed readout contribution and decision failure. For a fixed linear readout, an exact decomposition shows how response magnitude, readout coefficients, directional alignment, and the clean margin determine downstream consequences. In two frozen safety readouts, combining RSS support erosion with the full clean margin yields recurrent failure AUROCs of 0.935 and 0.958 using median supports of 20 and 19 features, compared with 0.965 and 0.964 using full erosion. RSS therefore provides a way to identify recurring sparse representation changes and quantify their contribution to a downstream readout without equating representation change with failure. Code is available at this https URL

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.21240 — 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 2026-10-16T08:44:17.922Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).