首页 > AI前沿 > MemDoctor: Stage-wise Error Attribution and Repair for MLLM Agent Memory

MemDoctor: Stage-wise Error Attribution and Repair for MLLM Agent Memory

arXiv自然语言 2026-11-16 20:38 15 阅读 查看原文
用户希望将一段纯文本的新闻(实际上是学术论文摘要)排版成语义化的HTML片段。 规则要求: 1. 只输出HTML片段本身,不要Markdown围栏,不要解释。 2. 每个自然段落用`

`包裹,长文用`

`/`

`划分小节。 3. 列表用`
  • `,引用用`
    `,代码用`
    `,重要句子可用``。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出``/``/``等文档级标签。
    
    分析原文:
    "Multimodal large language models (MLLMs) now assist users across sessions of images and text, carried by an external memory of what was said and shown. When such an assistant answers wrongly, the failure usually began earlier, in how a record was written, updated or retrieved, and the answer alone does not say where. We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them. A stage-resolved analysis on Mem-Gallery traces most failures to a source stage, and most of those lie before the answer. Injecting known errors against fact-preserving control edits, we then follow each tainted record downstream. Early errors survive to the answer as apparent evidence, and hurt most when no other record holds the fact. A diagnosis that trusts the quoted evidence therefore blames the stage where the error surfaces, not the one where it began. We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures. It traces each needed fact backward through the four stages, accepts a diagnosis only when the cited evidence is found in memory, and repairs that stage before answering again. On a 4B backbone, MemDoctor improves accuracy on both Mem-Gallery and MemLens, outperforming all evaluated 8B baselines, and the same loop also improves MuRAG and A-Mem."
    
    原文是一整段,没有明显的小节标题,也没有列表、引用或代码。
    按照规则,每个自然段落用`

    `包裹。因为只有一段,所以直接用一个`

    `包裹即可。 为了体现“重要句子可用``”,可以将核心贡献或结论加粗,比如“We propose MemDoctor...”或者“We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them.”。 不过规则说“不得增删改内容”,加粗不改变文本内容,只是增加样式。 让我们仔细看一下原文结构,其实可以拆分成几个自然段落,或者就保持一段。通常摘要就是一段。如果保持一段:

    Multimodal large language models (MLLMs) now assist users across sessions of images and text, carried by an external memory of what was said and shown. When such an assistant answers wrongly, the failure usually began earlier, in how a record was written, updated or retrieved, and the answer alone does not say where. We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them. A stage-resolved analysis on Mem-Gallery traces most failures to a source stage, and most of those lie before the answer. Injecting known errors against fact-preserving control edits, we then follow each tainted record downstream. Early errors survive to the answer as apparent evidence, and hurt most when no other record holds the fact. A diagnosis that trusts the quoted evidence therefore blames the stage where the error surfaces, not the one where it began. We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures. It traces each needed fact backward through the four stages, accepts a diagnosis only when the cited evidence is found in memory, and repairs that stage before answering again. On a 4B backbone, MemDoctor improves accuracy on both Mem-Gallery and MemLens, outperforming all evaluated 8B baselines, and the same loop also improves MuRAG and A-Mem.

    检查规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 -> 满足 2. 每个自然段落用

    包裹,长文用

    /

    划分小节。 -> 这里只有一段,不需要h2/h3。 3. 列表内容用
    • ,引用用
      ,代码用
      ,重要句子可用。 -> 没有列表、引用、代码。重要句子用了。
      4. 保留原文所有信息,不得增删改内容。 -> 满足,文本完全一致。
      5. 不要输出//等文档级标签。 -> 满足。
      
      再检查一下文本是否有拼写或格式问题。
      "Multimodal large language models (MLLMs) now assist users across sessions of images and text, carried by an external memory of what was said and shown. When such an assistant answers wrongly, the failure usually began earlier, in how a record was written, updated or retrieved, and the answer alone does not say where. We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them. A stage-resolved analysis on Mem-Gallery traces most failures to a source stage, and most of those lie before the answer. Injecting known errors against fact-preserving control edits, we then follow each tainted record downstream. Early errors survive to the answer as apparent evidence, and hurt most when no other record holds the fact. A diagnosis that trusts the quoted evidence therefore blames the stage where the error surfaces, not the one where it began. We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures. It traces each needed fact backward through the four stages, accepts a diagnosis only when the cited evidence is found in memory, and repairs that stage before answering again. On a 4B backbone, MemDoctor improves accuracy on both Mem-Gallery and MemLens, outperforming all evaluated 8B baselines, and the same loop also improves MuRAG and A-Mem."
      
      直接输出这个 `

      ` 标签即可。 ```html

      Multimodal large language models (MLLMs) now assist users across sessions of images and text, carried by an external memory of what was said and shown. When such an assistant answers wrongly, the failure usually began earlier, in how a record was written, updated or retrieved, and the answer alone does not say where. We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them. A stage-resolved analysis on Mem-Gallery traces most failures to a source stage, and most of those lie before the answer. Injecting known errors against fact-preserving control edits, we then follow each tainted record downstream. Early errors survive to the answer as apparent evidence, and hurt most when no other record holds the fact. A diagnosis that trusts the quoted evidence therefore blames the stage where the error surfaces, not the one where it began. We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures. It traces each needed fact backward through the four stages, accepts a diagnosis only when the cited evidence is found in memory, and repairs that stage before answering again. On a 4B backbone, MemDoctor improves accuracy on both Mem-Gallery and MemLens, outperforming all evaluated 8B baselines, and the same loop also improves MuRAG and A-Mem.

      ``` 注意:规则要求“不要Markdown围栏”,所以直接输出 `

      ...

      `,不带 ```html 和 ```。 确认没有遗漏任何字符。 原文:"We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures." 加粗后:"We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures." 没问题。 或者也可以把 "We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them." 加粗,不过加粗哪句都可以,只要不改变文字。就加粗提出 MemDoctor 这句,比较符合“重要句子”的定义。 最终输出:

      Multimodal large language models (MLLMs) now assist users across sessions of images and text, carried by an external memory of what was said and shown. When such an assistant answers wrongly, the failure usually began earlier, in how a record was written, updated or retrieved, and the answer alone does not say where. We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error