MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering
This addresses privacy and safety concerns for users deploying MLLMs, though it is an incremental improvement over existing unlearning methods.
The paper tackles the problem of unlearning memorized private data, outdated knowledge, and harmful content in multimodal large language models (MLLMs) by proposing MLLMEraser, a training-free framework using activation steering, which achieves stronger forgetting performance with lower computational cost and minimal utility degradation compared to state-of-the-art baselines.
Multimodal large language models (MLLMs) have demonstrated remarkable capabilities across vision-language tasks, yet their large-scale deployment raises pressing concerns about memorized private data, outdated knowledge, and harmful content. Existing unlearning approaches for MLLMs typically adapt training-based strategies such as gradient ascent or preference optimization, but these methods are computationally expensive, irreversible, and often distort retained knowledge. In this work, we propose MLLMEraser, an input-aware, training-free framework for test-time unlearning. Our approach leverages activation steering to enable dynamic knowledge erasure without parameter updates. Specifically, we construct a multimodal erasure direction by contrasting adversarially perturbed, knowledge-recall image-text pairs with knowledge-erasure counterparts, capturing both textual and visual discrepancies. To prevent unnecessary interference, we further design an input-aware steering mechanism that adaptively determines when and how the erasure direction should be applied, preserving utility on retained knowledge while enforcing forgetting on designated content. Experiments on LLaVA-1.5 and Qwen-2.5-VL demonstrate that MLLMEraser consistently outperforms state-of-the-art MLLM unlearning baselines, achieving stronger forgetting performance with lower computational cost and minimal utility degradation.