Metraipolis › archive › dc_352cf099b922
Hidden not Deleted: How Networks Suppress Entangled Features — arXiv:2609.27593
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.27593
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T09:12:48.227Z
- SHA-256 of the archived text
4aefc520d257809a9f8f1a7a2adf48b96707fa8d636d03642bb67863aa79d107- Stable URL of this page
https://metraipolis.com/documento/dc_352cf099b922
How to cite
Hidden not Deleted: How Networks Suppress Entangled Features — arXiv:2609.27593 — arXiv (en), https://arxiv.org/abs/2609.27593 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_352cf099b922 (Metraipolis archive dc_352cf099b922, sha256 4aefc520d257809a9f8f1a7a2adf48b96707fa8d636d03642bb67863aa79d107)
BibTeX
@misc{dc352cf099b922,
title = {Hidden not Deleted: How Networks Suppress Entangled Features — arXiv:2609.27593},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.27593},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_352cf099b922; sha256 4aefc520d257809a9f8f1a7a2adf48b96707fa8d636d03642bb67863aa79d107},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.27593. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Concept erasure methods that operate via linear projection assume that features occupy separable subspaces. We show this assumption fails under dense superposition: when two features are forced into an antipodal pair sharing a single subspace, state-of-the-art linear erasure destroys both, not just the target. Networks trained with gradient descent instead solve this problem non-linearly, but not uniformly: they converge to one of two distinct circuit-level solutions depending on initialization, which we call mirror and shadow solutions. We map this bifurcation as a function of feature entanglement, show it reflects a stable attractor structure rather than an artifact of our setup, and use targeted causal interventions to demonstrate that both solutions leave a substantial, measurable trace of the erased feature's representation intact, recoverable through a single scalar patch rather than requiring any further training. This mirrors a failure mode recently observed empirically in LLM unlearning, where suppression rather than deletion allows forgotten knowledge to resurface; our results offer a mechanistic, causally-validated account of why that failure mode occurs. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-24 in cs.AI). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.27593 — 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 2026-10-08T08:13:26.694Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-24?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).