跳到正文
arXiv · cs.AI · Wenyi Wu, Zixuan Song, Kun Zhou, Yifei Shao, Zhiting Hu, Biwei Huang · 2 天前模型评分41

CoMEM:为视觉语言模型构建通用连续记忆

Towards General Continuous Memory for Vision-Language Models

导读

研究者提出 CoMEM,让视觉语言模型(VLM,Vision-Language Model)充当自身的连续记忆编码器,把任意多模态、多语言知识压缩为仅 8 个连续嵌入向量。该方法只需微调 1.2% 的模型参数、使用 15.6K 条自合成样本,推理时原模型保持冻结,记忆模块即插即用。在八个多模态推理基准上,这一设计验证了其有效性。

分类置信度

中 · 待人工复核

  • 只有单一媒体来源,需要人工复核。

来源级别:媒体与个人

事件类型:研究发布

证据摘录

  • “In contrast, we propose using continuous memory, a compact set of dense embeddings to more effectively and efficiently represent multimodal and multilingual knowledge.”
  • “requiring only 1.2% of the model's parameters and a small corpus of 15.6K self-synthesized samples”
  • “Since the inference-time VLM remains frozen, our memory module is plug-and-play and can be flexibly integrated as needed.”

来源:arXiv · cs.AI · arxiv.org