发现多模态模型编辑会因输入拆分失效，提出新方法精准定位并更新跨模态知识。
Correct When Paired, Wrong When Split: Decoupling and Editing Modality-Specific Neurons in MLLMs
- 分离模态特异神经元，实现针对文本与图像路径的精准知识编辑。
- 在多模态输入下编辑成功率超90%，单模态输入下仍保持有效更新。
- 适合需跨模态知识维护的研究者，尤其关注模型可解释性与编辑可靠性。
尽管知识编辑为多模态大语言模型（MLLMs）提供了高效的更新机制，但当前范式仍存在一个重要却未被充分探索的问题：编辑解耦失败。即当模型接收图文联合查询时，实体知识可被成功更新，但在输入被拆分为单一模态时，知识往往恢复到旧有预编辑状态。深入的实证分析表明，MLLM中的实体知识并非以统一表征存储，而是分布于解耦的模态特异性路径中。因此，偏向多模态查询的更新无法有效传播至单模态回路。为此，我们提出DECODE，显式解耦并定位模态特异性神经元组，实现目标知识的精准编辑。大量实验表明，DECODE在不同模态触发条件下均能持续实现有效知识更新，显著缓解了编辑解耦失败问题。
原文摘要 · Abstract (English)
Although Knowledge Editing provides an efficient mechanism for updating the knowledge of Multimodal Large Language Models (MLLMs), we find that current paradigms still suffer from an important yet remain underexplored issue : editing decoupling failure, where entity-related knowledge can be updated when the model is triggered by multimodal inputs (text--image query pairs), however, it often reverts to outdated pre-edit facts when the paired inputs are split into unimodal ones. Our in-depth empirical analysis reveals that the entity knowledge in MLLMs is not stored as a unified representation, but is instead distributed across disentangled modality-specific pathways. As a result, updates biased toward multimodal queries fail to propagate effectively to unimodal circuits. To bridge this gap, we propose DECODE, which explicitly disentangles and localizes modality-specific neuron groups for targeted knowledge. Extensive experiments demonstrate that DECODE consistently achieves effective knowledge updates under different modality triggers, thereby mitigating editing decoupling failures.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。