Preference Poisoning Attacks on Reward Model Learning

Documenti analoghi

• RLHFPoison: Reward Poisoning Attack for Reinforcement Learning with Human Feedback in Large Language Models
  di: Wang, Jiongxiao, et al.
  Pubblicazione: (2023)
• Verified Safe Reinforcement Learning for Neural Network Dynamic Models
  di: Wu, Junlin, et al.
  Pubblicazione: (2024)
• Learning Linear Utility Functions From Pairwise Comparison Queries
  di: Ge, Luise, et al.
  Pubblicazione: (2024)
• Protecting Language Models Against Unauthorized Distillation through Trace Rewriting
  di: Ma, Xinhang, et al.
  Pubblicazione: (2026)
• Larger or Smaller Reward Margins to Select Preferences for Alignment?
  di: Huang, Kexin, et al.
  Pubblicazione: (2025)