1
ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models
多模态大模型遗忘新方法ASRU，结合激活引导与强化学习，提升遗忘后生成质量，更符合实际需求。
arXiv:2605.15687v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) may memorize sensitive cross-modal information during pretr…