Retrievals Can Be Detrimental: Unveiling the Backdoor Vulnerability of Retrieval-Augmented Diffusion Models

Documenti analoghi

• Neural Antidote: Class-Wise Prompt Tuning for Purifying Backdoors in CLIP
  di: Kong, Jiawei, et al.
  Pubblicazione: (2025)
• Grounding Language with Vision: A Conditional Mutual Information Calibrated Decoding Strategy for Reducing Hallucinations in LVLMs
  di: Fang, Hao, et al.
  Pubblicazione: (2025)
• Seeing Through the Chain: Mitigate Hallucination in Multimodal Reasoning Models via CoT Compression and Contrastive Preference Optimization
  di: Fang, Hao, et al.
  Pubblicazione: (2026)
• Rank Matters: Understanding and Defending Model Inversion Attacks via Low-Rank Feature Filtering
  di: Yu, Hongyao, et al.
  Pubblicazione: (2024)
• BadCLIP: Trigger-Aware Prompt Learning for Backdoor Attacks on CLIP
  di: Bai, Jiawang, et al.
  Pubblicazione: (2023)