Aligned but Fragile: Enhancing LLM Safety Robustness via Zeroth-Order Optimization

Documenti analoghi

• Robust and Efficient Zeroth-Order LLM Fine-Tuning via Adaptive Bayesian Subspace Optimizer
  di: Feng, Jian, et al.
  Pubblicazione: (2026)
• Localized Zeroth-Order Prompt Optimization
  di: Hu, Wenyang, et al.
  Pubblicazione: (2024)
• Towards Mitigating Excessive Forgetting in LLM Unlearning via Entanglement-Guidance with Proxy Constraint
  di: Liu, Zhihao, et al.
  Pubblicazione: (2025)
• AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization
  di: Cemri, Mert, et al.
  Pubblicazione: (2026)
• OAT-Rephrase: Optimization-Aware Training Data Rephrasing for Zeroth-Order LLM Fine-Tuning
  di: Long, Jikai, et al.
  Pubblicazione: (2025)