Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection

Documenti analoghi

• Online Merging Optimizers for Boosting Rewards and Mitigating Tax in Alignment
  di: Lu, Keming, et al.
  Pubblicazione: (2024)
• Mitigating the Alignment Tax of RLHF
  di: Lin, Yong, et al.
  Pubblicazione: (2023)
• Domain Generalizable Continual Learning
  di: Yan, Hongwei, et al.
  Pubblicazione: (2025)
• Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax
  di: Su, Zeli, et al.
  Pubblicazione: (2026)
• Noise Contrastive Alignment of Language Models with Explicit Rewards
  di: Chen, Huayu, et al.
  Pubblicazione: (2024)