Safety Mirage: How Spurious Correlations Undermine VLM Safety Fine-Tuning and Can Be Mitigated by Machine Unlearning

Documenti analoghi

• Backdoor Secrets Unveiled: Identifying Backdoor Data with Optimized Scaled Prediction Consistency
  di: Pal, Soumyadeep, et al.
  Pubblicazione: (2024)
• Unlearners Can Lie: Evaluating and Improving Honesty in LLM Unlearning
  di: Gu, Renjie, et al.
  Pubblicazione: (2026)
• SEUF: Is Unlearning One Expert Enough for Mixture-of-Experts LLMs?
  di: Zhuang, Haomin, et al.
  Pubblicazione: (2024)
• Forgetting to Forget: Attention Sink as A Gateway for Backdooring LLM Unlearning
  di: Shang, Bingqi, et al.
  Pubblicazione: (2025)
• RaVL: Discovering and Mitigating Spurious Correlations in Fine-Tuned Vision-Language Models
  di: Varma, Maya, et al.
  Pubblicazione: (2024)