Imagen de Portada

ROAST: Rollout-based On-distribution Activation Steering Technique

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Su, Xuanbo, Luo, Hao, Zhang, Yingfang, Zhang, Lijun
Formato:	Preprint
Publicado:	2026
Materias:	Machine Learning Computation and Language
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Beyond Steering Vector: Flow-based Activation Steering for Inference-Time Intervention
por: Jin, Zehao, et al.
Publicado: (2026)

Mistake Notebook Learning: Batch-Clustered Failures for Training-Free Agent Adaptation
por: Su, Xuanbo, et al.
Publicado: (2025)

ROAST: Review-level Opinion Aspect Sentiment Target Joint Detection for ABSA
por: Chebolu, Siva Uday Sampreeth, et al.
Publicado: (2024)

Sparse-RL: Breaking the Memory Wall in LLM Reinforcement Learning via Stable Sparse Rollouts
por: Luo, Sijia, et al.
Publicado: (2026)

Steering Language Models With Activation Engineering
por: Turner, Alexander Matt, et al.
Publicado: (2023)

HyperSteer: Activation Steering at Scale with Hypernetworks
por: Sun, Jiuding, et al.
Publicado: (2025)

Steer Like the LLM: Activation Steering that Mimics Prompting
por: Heyman, Geert, et al.
Publicado: (2026)

Mem-T: Densifying Rewards for Long-Horizon Memory Agents
por: Yue, Yanwei, et al.
Publicado: (2026)

SPEC-RL: Accelerating On-Policy Reinforcement Learning with Speculative Rollouts
por: Liu, Bingshuai, et al.
Publicado: (2025)

Spherical Steering: Geometry-Aware Activation Rotation for Language Models
por: You, Zejia, et al.
Publicado: (2026)

Guiding Giants: Lightweight Controllers for Weighted Activation Steering in LLMs
por: Hegazy, Amr, et al.
Publicado: (2025)

Steering MoE LLMs via Expert (De)Activation
por: Fayyaz, Mohsen, et al.
Publicado: (2025)

Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning
por: Xu, Yixuan Even, et al.
Publicado: (2025)

Do Not Waste Your Rollouts: Recycling Search Experience for Efficient Test-Time Scaling
por: Wang, Xinglin, et al.
Publicado: (2026)

Programming Refusal with Conditional Activation Steering
por: Lee, Bruce W., et al.
Publicado: (2024)

SAKE: Steering Activations for Knowledge Editing
por: Scialanga, Marco, et al.
Publicado: (2025)

Understanding How CodeLLMs (Mis)Predict Types with Activation Steering
por: Lucchetti, Francesca, et al.
Publicado: (2024)

FishBack: Pullback Fisher Geometry for Optimal Activation Steering in Transformers
por: Wang, Sihan, et al.
Publicado: (2026)

HINT: Helping Ineffective Rollouts Navigate Towards Effectiveness
por: Wang, Xinyi, et al.
Publicado: (2025)

Faithful Bi-Directional Model Steering via Distribution Matching and Distributed Interchange Interventions
por: Bao, Yuntai, et al.
Publicado: (2026)

Endogenous Resistance to Activation Steering in Language Models
por: McKenzie, Alex, et al.
Publicado: (2026)

Activation Steering for Synthetic Data Generation: The Role of Diversity in Downstream Safety Detection
por: Deshpande, Vijeta, et al.
Publicado: (2026)

$V_{0.5}$: Generalist Value Model as a Prior for Sparse RL Rollouts
por: Zhang, Yi-Kai, et al.
Publicado: (2026)

Extracting Unlearned Information from LLMs with Activation Steering
por: Seyitoğlu, Atakan, et al.
Publicado: (2024)

Steering Llama 2 via Contrastive Activation Addition
por: Panickssery, Nina, et al.
Publicado: (2023)

Activation Steering via Generative Causal Mediation
por: Sankaranarayanan, Aruna, et al.
Publicado: (2026)

VSPO: Vector-Steered Policy Optimization for Behavioral Control
por: Zhang, Xuechen, et al.
Publicado: (2026)

Learning Self-Correction in Vision-Language Models via Rollout Augmentation
por: Ding, Yi, et al.
Publicado: (2026)

Extending Activation Steering to Broad Skills and Multiple Behaviours
por: van der Weij, Teun, et al.
Publicado: (2024)

Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimization
por: Cao, Yuanpu, et al.
Publicado: (2024)

Improving Instruction-Following in Language Models through Activation Steering
por: Stolfo, Alessandro, et al.
Publicado: (2024)

Accelerating RL Post-Training Rollouts via System-Integrated Speculative Decoding
por: Iso, Hayate, et al.
Publicado: (2026)

SteerConf: Steering LLMs for Confidence Elicitation
por: Zhou, Ziang, et al.
Publicado: (2025)

Knowledgeable Agents by Offline Reinforcement Learning from Large Language Model Rollouts
por: Pang, Jing-Cheng, et al.
Publicado: (2024)

ContextFocus: Activation Steering for Contextual Faithfulness in Large Language Models
por: Anand, Nikhil, et al.
Publicado: (2026)

Multi-property Steering of Large Language Models with Dynamic Activation Composition
por: Scalena, Daniel, et al.
Publicado: (2024)

Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation Steering
por: Bigelow, Eric, et al.
Publicado: (2025)

Interpretable Steering of Large Language Models with Feature Guided Activation Additions
por: Soo, Samuel, et al.
Publicado: (2025)

Conceptors for Semantic Steering
por: Triantafyllopoulos, Ilias, et al.
Publicado: (2026)

Jet-RL: Enabling On-Policy FP8 Reinforcement Learning with Unified Training and Rollout Precision Flow
por: Xi, Haocheng, et al.
Publicado: (2026)