Metraipolis › archive › dc_db018d6bdec5
LocUS: Head Selection and Subspace Projection for Targeted Activation Steering — arXiv:2609.31122
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.31122
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T02:04:06.972Z
- SHA-256 of the archived text
76ce3056c5690061e87e0f5af5e5745f494f9a0b473e44212bac7b3c4559bb26- Stable URL of this page
https://metraipolis.com/documento/dc_db018d6bdec5
How to cite
LocUS: Head Selection and Subspace Projection for Targeted Activation Steering — arXiv:2609.31122 — arXiv (en), https://arxiv.org/abs/2609.31122 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_db018d6bdec5 (Metraipolis archive dc_db018d6bdec5, sha256 76ce3056c5690061e87e0f5af5e5745f494f9a0b473e44212bac7b3c4559bb26)
BibTeX
@misc{dcdb018d6bdec5,
title = {LocUS: Head Selection and Subspace Projection for Targeted Activation Steering — arXiv:2609.31122},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.31122},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_db018d6bdec5; sha256 76ce3056c5690061e87e0f5af5e5745f494f9a0b473e44212bac7b3c4559bb26},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.31122. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Activation steering is a powerful training-free paradigm for controlling large language models at inference time. However, standard approaches estimate a per-layer steering direction from contrastive data and apply it on the layer's entire representation space, which may couple the intervention to off-target properties present in the contrastive data and degrade unrelated capabilities. To mitigate this issue, we introduce LocUS (Localized Unembedding Steering), a method which grounds activation steering to the model's own output vocabulary subspace. By identifying a property-specific linear subspace within the unembedding matrix, LocUS enforces a geometric constraint that restricts the steering transformation to a specific subspace and at the same time localizes its application to a sparse subset of attention heads. Extensive evaluations across three model families on toxicity mitigation, sentiment redirection and sycophancy suppression show that LocUS matches or outperforms state-of-the-art baselines while intervening on under 6% of parameters and better preserving general capability. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-28 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.31122 — 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 2026-10-08T01:38:49.458Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).