Paper

{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.17560v1</id>\n    <title>ODESteer: A Unified ODE-Based Steering Framework for LLM Alignment</title>\n    <updated>2026-02-19T17:13:44Z</updated>\n    <link href='https://arxiv.org/abs/2602.17560v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.17560v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Activation steering, or representation engineering, offers a lightweight approach to align large language models (LLMs) by manipulating their internal activations at inference time. However, current methods suffer from two key limitations: \\textit{(i)} the lack of a unified theoretical framework for guiding the design of steering directions, and \\textit{(ii)} an over-reliance on \\textit{one-step steering} that fail to capture complex patterns of activation distributions. In this work, we propose a unified ordinary differential equations (ODEs)-based \\textit{theoretical} framework for activation steering in LLM alignment. We show that conventional activation addition can be interpreted as a first-order approximation to the solution of an ODE. Based on this ODE perspective, identifying a steering direction becomes equivalent to designing a \\textit{barrier function} from control theory. Derived from this framework, we introduce ODESteer, a kind of ODE-based steering guided by barrier functions, which shows \\textit{empirical} advancement in LLM alignment. ODESteer identifies steering directions by defining the barrier function as the log-density ratio between positive and negative activations, and employs it to construct an ODE for \\textit{multi-step and adaptive} steering. Compared to state-of-the-art activation steering methods, ODESteer achieves consistent empirical improvements on diverse LLM alignment benchmarks, a notable $5.7\\%$ improvement over TruthfulQA, $2.5\\%$ over UltraFeedback, and $2.4\\%$ over RealToxicityPrompts. Our work establishes a principled new view of activation steering in LLM alignment by unifying its theoretical foundations via ODEs, and validating it empirically through the proposed ODESteer method.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-02-19T17:13:44Z</published>\n    <arxiv:comment>Accepted by ICLR 2026</arxiv:comment>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Hongjue Zhao</name>\n    </author>\n    <author>\n      <name>Haosen Sun</name>\n    </author>\n    <author>\n      <name>Jiangtao Kong</name>\n    </author>\n    <author>\n      <name>Xiaochang Li</name>\n    </author>\n    <author>\n      <name>Qineng Wang</name>\n    </author>\n    <author>\n      <name>Liwei Jiang</name>\n    </author>\n    <author>\n      <name>Qi Zhu</name>\n    </author>\n    <author>\n      <name>Tarek Abdelzaher</name>\n    </author>\n    <author>\n      <name>Yejin Choi</name>\n    </author>\n    <author>\n      <name>Manling Li</name>\n    </author>\n    <author>\n      <name>Huajie Shao</name>\n    </author>\n  </entry>"
}