ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models

arXiv:2605.15687v2 Announce Type: replace-cross Abstract: Multimodal large language models (MLLMs) may memorize sensitive cross-modal information during pretraining, making machine unlearning (MU) crucial. Existing methods typically evaluate unlearning effectiveness based on output deviations, while overlooking the generation quality after...

Læs hele artiklen hos kilden.