Diagnosing and Repairing Unsafe Channels in Vision-Language Models via Causal Discovery and Dual-Modal Safety Subspace Projection

Ähnliche Einträge

• Learning to Edit Knowledge via Instruction-based Chain-of-Thought Prompting
  von: Fu, Jinhu, et al.
  Veröffentlicht: (2026)
• Bridging the Gap in Vision Language Models in Identifying Unsafe Concepts Across Modalities
  von: Qu, Yiting, et al.
  Veröffentlicht: (2025)
• Safe Inputs but Unsafe Output: Benchmarking Cross-modality Safety Alignment of Large Vision-Language Model
  von: Wang, Siyin, et al.
  Veröffentlicht: (2024)
• Safe Vision-Language Models via Unsafe Weights Manipulation
  von: D'Incà, Moreno, et al.
  Veröffentlicht: (2025)
• Safe Semantics, Unsafe Interpretations: Tackling Implicit Reasoning Safety in Large Vision-Language Models
  von: Cai, Wei, et al.
  Veröffentlicht: (2025)