Cover Image

Principled Steering via Null-space Projection for Jailbreak Defense in Vision-Language Models

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Zhu, Xingyu, Zhu, Beier, Wang, Shuo, Fang, Junfeng, Zhao, Kesen, Zhang, Hanwang, He, Xiangnan
Format:	Preprint
Published:	2026
Subjects:	Computer Vision and Pattern Recognition
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Enhancing CLIP Robustness via Cross-Modality Alignment
by: Zhu, Xingyu, et al.
Published: (2025)

Hierarchical Semantic Alignment for Image Clustering
by: Zhu, Xingyu, et al.
Published: (2025)

Mitigating Hallucinations in Large Vision-Language Models without Performance Degradation
by: Zhu, Xingyu, et al.
Published: (2026)

Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation
by: Zhu, Xingyu, et al.
Published: (2026)

Unsupervised Visual Chain-of-Thought Reasoning via Preference Optimization
by: Zhao, Kesen, et al.
Published: (2025)

Selective Vision-Language Subspace Projection for Few-shot CLIP
by: Zhu, Xingyu, et al.
Published: (2024)

GuardAlign: Test-time Safety Alignment in Multimodal Large Language Models
by: Zhu, Xingyu, et al.
Published: (2026)

Dynamic Multimodal Prototype Learning in Vision-Language Models
by: Zhu, Xingyu, et al.
Published: (2025)

Thinking with Images as Continuous Actions: Numerical Visual Chain-of-Thought
by: Zhao, Kesen, et al.
Published: (2026)

Enhancing Zero-Shot Vision Models by Label-Free Prompt Distribution Learning and Bias Correcting
by: Zhu, Xingyu, et al.
Published: (2024)

MuSteerNet: Human Reaction Generation from Videos via Observation-Reaction Mutual Steering
by: Zhou, Yuan, et al.
Published: (2026)

Adapting Point Cloud Analysis via Multimodal Bayesian Distribution Learning
by: Zhu, Xingyu, et al.
Published: (2026)

Robust Fine-tuning of Zero-shot Models via Variance Reduction
by: Zhu, Beier, et al.
Published: (2024)

Reducing Class-Wise Performance Disparity via Margin Regularization
by: Zhu, Beier, et al.
Published: (2026)

Streaming Drag-Oriented Interactive Video Manipulation: Drag Anything, Anytime!
by: Zhou, Junbao, et al.
Published: (2025)

Project-Probe-Aggregate: Efficient Fine-Tuning for Group Robustness
by: Zhu, Beier, et al.
Published: (2025)

Debiased Fine-Tuning for Vision-language Models by Prompt Regularization
by: Zhu, Beier, et al.
Published: (2023)

Real-Time Motion-Controllable Autoregressive Video Diffusion
by: Zhao, Kesen, et al.
Published: (2025)

Generalized Logit Adjustment: Calibrating Fine-tuned Models by Removing Label Bias in Foundation Models
by: Zhu, Beier, et al.
Published: (2023)

Activation Steering Meets Preference Optimization: Defense Against Jailbreaks in Vision Language Models
by: Wu, Sihao, et al.
Published: (2025)

Distilling Parallel Gradients for Fast ODE Solvers of Diffusion Models
by: Zhu, Beier, et al.
Published: (2025)

Boosting Few-Shot Learning via Attentive Feature Regularization
by: Zhu, Xingyu, et al.
Published: (2024)

Defense-to-Attack: Bypassing Weak Defenses Enables Stronger Jailbreaks in Vision-Language Models
by: Zhao, Yunhan, et al.
Published: (2025)

Prompt-aligned Gradient for Prompt Tuning
by: Zhu, Beier, et al.
Published: (2022)

VLS: Steering Pretrained Robot Policies via Vision-Language Models
by: Liu, Shuo, et al.
Published: (2026)

TAIJI: Textual Anchoring for Immunizing Jailbreak Images in Vision Language Models
by: Yin, Xiangyu, et al.
Published: (2025)

Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lens
by: Jiang, Zhangqi, et al.
Published: (2024)

Parallel Diffusion Solver via Residual Dirichlet Policy Optimization
by: Wang, Ruoyu, et al.
Published: (2025)

Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks
by: Wang, Han, et al.
Published: (2024)

Jailbreak Large Vision-Language Models Through Multi-Modal Linkage
by: Wang, Yu, et al.
Published: (2024)

DAMA: Data- and Model-aware Alignment of Multi-modal LLMs
by: Lu, Jinda, et al.
Published: (2025)

3D Question Answering via only 2D Vision-Language Models
by: Wang, Fengyun, et al.
Published: (2025)

UniVLR: Unifying Text and Vision in Visual Latent Reasoning for Multimodal LLMs
by: Jiang, Houcheng, et al.
Published: (2026)

Classes Are Not Equal: An Empirical Study on Image Recognition Fairness
by: Cui, Jiequan, et al.
Published: (2024)

E$^2$AT: Multimodal Jailbreak Defense via Dynamic Joint Optimization for Multimodal Large Language Models
by: Lu, Liming, et al.
Published: (2025)

Jailbreaks on Vision Language Model via Multimodal Reasoning
by: Noheria, Aarush, et al.
Published: (2026)

DragNeXt: Rethinking Drag-Based Image Editing
by: Zhou, Yuan, et al.
Published: (2025)

Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization
by: Li, Jinghan, et al.
Published: (2026)

JaiLIP: Jailbreaking Vision-Language Models via Loss Guided Image Perturbation
by: Mia, Md Jueal, et al.
Published: (2025)

Jailbreak Vision Language Models via Bi-Modal Adversarial Prompt
by: Ying, Zonghao, et al.
Published: (2024)