ACL Findings2026
RISER: Orchestrating Latent Reasoning Skills for Adaptive Activation Steering
Wencheng Ye, Xiaoyang Yuan, Yi Bin, Hengyu Jin, Liang Peng, Pengpeng Zeng, Heng Tao Shen
Abstract
Recent work on domain-specific reasoning with large language models (LLMs) often relies on training-intensive approaches that require parameter updates. While activation steering has emerged as a parameterefficient alternative, existing methods apply static, manual interventions that fail to adapt to the dynamic nature of complex reasoning. To address this limitation, we propose RISER (Router-based Intervention for Steerable Enhancement of Reasoning), a plugand-play intervention framework that adaptively steers LLM reasoning in activation space. RISER constructs a library of reusable reasoning vectors and employs a lightweight Router to dynamically compose them for each input. The Router is optimized via reinforcement learning under task-level rewards, activating latent cognitive primitives in an emergent and compositional manner. Across seven diverse benchmarks, RISER yields 3.4-6.5% average zero-shot accuracy improvements over the base model while surpassing CoT-style reasoning with 2-3× higher token efficiency and robust accuracy gains. Further analysis shows that RISER can autonomously combine multiple vectors into interpretable, precise control strategies, paving way for more controllable and efficient LLM reasoning.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
Builds on22
- Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah et al.NeurIPS 2020 · 64,255 citations
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- Measuring Massive Multitask Language UnderstandingDan Hendrycks, Collin Burns, Steven Basart, Andy Zou et al.ICLR 2021 · 7,905 citations
- TruthfulQA: Measuring How Models Mimic Human FalsehoodsStephanie Lin, Jacob Hilton, Owain EvansACL 2022 · 3,228 citations
- Sparse Autoencoders Find Highly Interpretable Features in Language ModelsRobert Huben, Hoagy Cunningham, Logan Riggs Smith, Aidan Ewart et al.ICLR 2024 · 1,072 citations
Related papers
- Characterizing and Mitigating Reasoning Drift in Large Language ModelsYufeng Zhang, Xuepeng Wang, Lingxiang Wu, Jinqiao WangICLR 2026
- Activation Steering for Chain-of-Thought CompressionSeyedarmin Azizi, Erfan Baghaei Potraghloo, Souvik Kundu, Massoud PedramACL Findings 2026
- Learning to Reason over Continuous Tokens with Reinforcement LearningYiran Zhao, Yuhui Xu, Doyen Sahoo, Caiming Xiong et al.ICLR 2026 · 1 citation
- Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation SteeringEric Bigelow, Daniel Wurgaft, YingQiao Wang, Noah Goodman et al.ICML 2026
- AdaMix: Adaptive Mixing for Short and Long Reasoning AdaptersHao Luo, Xiao Yan, Xinyan Li, Qiming Zeng et al.ACL 2026
