Metraipolis › archive › dc_e540620150ca
Information-Theoretic Design of Sparse Autoencoder-Based Steering — arXiv:2611.01378
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.01378
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T18:22:49.307Z
- SHA-256 of the archived text
fd002aaa79019971bd98c48af5838dd16df57394b46f1492119f517d013a29af- Stable URL of this page
https://metraipolis.com/documento/dc_e540620150ca
How to cite
Information-Theoretic Design of Sparse Autoencoder-Based Steering — arXiv:2611.01378 — arXiv (en), https://arxiv.org/abs/2611.01378 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e540620150ca (Metraipolis archive dc_e540620150ca, sha256 fd002aaa79019971bd98c48af5838dd16df57394b46f1492119f517d013a29af)
BibTeX
@misc{dce540620150ca,
title = {Information-Theoretic Design of Sparse Autoencoder-Based Steering — arXiv:2611.01378},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.01378},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e540620150ca; sha256 fd002aaa79019971bd98c48af5838dd16df57394b46f1492119f517d013a29af},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.01378. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Sparse autoencoder (SAE)-based activation steering is a lightweight approach for modifying the behavior of large language models (LLMs) by adding small vectors to hidden states during inference. However, designing effective steering interventions remains challenging. In this work, we develop an information-theoretic framework for analyzing SAE activations through empirical feature-firing distributions. The central object of the framework is the Jensen--Shannon divergence (JSD) between such firing distributions, which quantifies how far a generated or intervened state deviates from a dataset-induced target distribution. Based on an analytic divergence response for virtually shifted states, we construct steering vectors toward the target firing distribution. The same divergence can also be used to monitor deviation from another empirical distribution, such as a general valid-language distribution. We demonstrate the effectiveness of the steering in a question-mode generation task study. We further use teacher-forced pulse experiments to show that local interventions induce structured downstream JSD responses in later layers, and apply the framework to a safety-related target--contrast response setting for feature analysis. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2611.01378 — 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 2026-11-03T18:10:07.924Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).