#artificial intelligence
Oct 2025
MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering
This work proposes MLLMEraser, an input-aware, training-free framework for test-time unlearning that consistently outperforms state-of-the-art MLLM unlearning baselines, achieving stronger forgetting performance with lower computational cost and minimal utility degradation.
Chen-Lu Ding, Jian-Can Wu, Leheng Sheng et al.
· arXiv.org · 11 citations