Test-time Backdoor Mitigation for Black-Box Large Language Models with Defensive Demonstrations

Ähnliche Einträge

• Mitigating Backdoor Threats to Large Language Models: Advancement and Challenges
  von: Liu, Qin, et al.
  Veröffentlicht: (2024)
• Instructions as Backdoors: Backdoor Vulnerabilities of Instruction Tuning for Large Language Models
  von: Xu, Jiashu, et al.
  Veröffentlicht: (2023)
• From Shortcuts to Triggers: Backdoor Defense with Denoised PoE
  von: Liu, Qin, et al.
  Veröffentlicht: (2023)
• RLHFPoison: Reward Poisoning Attack for Reinforcement Learning with Human Feedback in Large Language Models
  von: Wang, Jiongxiao, et al.
  Veröffentlicht: (2023)
• FATH: Authentication-based Test-time Defense against Indirect Prompt Injection Attacks
  von: Wang, Jiongxiao, et al.
  Veröffentlicht: (2024)