`包裹,长文用`
`/`
`划分小节。 3. 列表内容用`
- `,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。
正文内容分析:
"As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, machine unlearning (MU) has emerged as a crucial solution for removing sensitive data while preserving model performance. However, existing MU methods typically rely on visual data of the target concepts, which is often unavailable due to strict data retention policies, thus creating a demand for source-free unlearning approaches that operate without access to the target data." -> 这是一个段落,可以加个标题或者直接作为第一段。考虑到是新闻/论文摘要类的,可以加个``作为背景或引言,或者直接用`
`。为了符合“长文用
/
划分小节”,我可以加几个小节标题,但原文没有明显的标题,如果我自己加标题算不算“增删改内容”?规则说“不得增删改内容”,所以最好不加额外的标题文字,或者把现有的句子结构作为段落。不过规则2说“长文用
/
划分小节”,如果不加标题文字,怎么划分?也许可以把关键概念作为标题?不行,“不得增删改内容”。那就直接用`
`包裹每个自然段,或者把描述阶段的部分用列表。
原文拆分:
段落1:As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, machine unlearning (MU) has emerged as a crucial solution for removing sensitive data while preserving model performance. However, existing MU methods typically rely on visual data of the target concepts, which is often unavailable due to strict data retention policies, thus creating a demand for source-free unlearning approaches that operate without access to the target data.
段落2:In this work, we propose Source-free Proxy Anchor Concept Erasure (SPACE), the first source-free unlearning framework specialized for MLLMs.
段落3:SPACE consists of two stages: (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space. (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts.
这里提到了两个阶段,可以用`
- `来展示,但原文是“(1) ... (2) ...”,如果改成列表,是否算修改内容?为了安全,保留原文的编号,或者用列表形式,把(1)和(2)放在li里。
例如:
SPACE consists of two stages:
- (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space.
- (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts.
这样没有增删改内容,只是改变了排版结构,符合规则3。
段落4:DCSI confines updates to the null space of retained knowledge, ensuring structural integrity. We theoretically prove that SPACE strictly bounds the perturbation on retained knowledge and maximizes feature spectral entropy, thereby maintaining the model's performance.
段落5:Furthermore, extensive experiments across six datasets show that SPACE achieves performance comparable to that of state-of-the-art data-dependent methods, validating its effectiveness in source-free MU scenarios. The source code will be released.
关于“长文用/
划分小节”:如果原文没有小节标题,强行加标题会违反“不得增删改内容”。所以这里可能不需要硬加`
/
`,或者可以把某些核心句子用``加粗。规则3提到“重要句子可用”。 比如:
In this work, we propose Source-free Proxy Anchor Concept Erasure (SPACE), the first source-free unlearning framework specialized for MLLMs.
检查规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 -> 好的。 2. 每个自然段落用
包裹,长文用
/
划分小节。 -> 原文不长,且没有现成标题,不加虚构标题。 3. 列表内容用
- ,引用用
,代码用
,重要句子可用。 -> 用- 处理两个阶段,重要名词用。
4. 保留原文所有信息,不得增删改内容。 -> 严格对照。
5. 不要输出//等文档级标签。 -> 好的。
构建HTML:
As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, machine unlearning (MU) has emerged as a crucial solution for removing sensitive data while preserving model performance. However, existing MU methods typically rely on visual data of the target concepts, which is often unavailable due to strict data retention policies, thus creating a demand for source-free unlearning approaches that operate without access to the target data.
In this work, we propose Source-free Proxy Anchor Concept Erasure (SPACE), the first source-free unlearning framework specialized for MLLMs.
SPACE consists of two stages:
- (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space.
- (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts.
DCSI confines updates to the null space of retained knowledge, ensuring structural integrity. We theoretically prove that SPACE strictly bounds the perturbation on retained knowledge and maximizes feature spectral entropy, thereby maintaining the model's performance.
Furthermore, extensive experiments across six datasets show that SPACE achieves performance comparable to that of state-of-the-art data-dependent methods, validating its effectiveness in source-free MU scenarios. The source code will be released.
检查原文是否完全一致:
原文:"SPACE consists of two stages: (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space. (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts."
我的HTML:
SPACE consists of two stages:
- (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space.
- (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts.
这里把冒号后的两句话变成了列表,文字内容没有增删改。为了更贴近原文,也可以保持在同一个里,但规则3要求“列表内容用
- ”,这明显是列表内容,所以用
是合理的。 再检查加粗部分: 原文:"source-free unlearning approaches" -> source-free unlearning approaches 原文:"Source-free Proxy Anchor Concept Erasure (SPACE)" -> Source-free Proxy Anchor Concept Erasure (SPACE) 没有修改文字,只是加了标签。 确保没有多余的换行和空格,输出纯净的HTML。 不需要
/
,因为原文没有小节标题,硬加会违反规则4。如果非要加,可以加个
引言
,但这属于增加内容。所以不用。 最终检查HTML标签闭合和格式。
...
`,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。 正文内容分析: "As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, machine unlearning (MU) has emerged as a crucial solution for removing sensitive data while preserving model performance. However, existing MU methods typically rely on visual data of the target concepts, which is often unavailable due to strict data retention policies, thus creating a demand for source-free unlearning approaches that operate without access to the target data." -> 这是一个段落,可以加个标题或者直接作为第一段。考虑到是新闻/论文摘要类的,可以加个``作为背景或引言,或者直接用`
`。为了符合“长文用
/
划分小节”,我可以加几个小节标题,但原文没有明显的标题,如果我自己加标题算不算“增删改内容”?规则说“不得增删改内容”,所以最好不加额外的标题文字,或者把现有的句子结构作为段落。不过规则2说“长文用
/
划分小节”,如果不加标题文字,怎么划分?也许可以把关键概念作为标题?不行,“不得增删改内容”。那就直接用`
`包裹每个自然段,或者把描述阶段的部分用列表。 原文拆分: 段落1:As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, machine unlearning (MU) has emerged as a crucial solution for removing sensitive data while preserving model performance. However, existing MU methods typically rely on visual data of the target concepts, which is often unavailable due to strict data retention policies, thus creating a demand for source-free unlearning approaches that operate without access to the target data. 段落2:In this work, we propose Source-free Proxy Anchor Concept Erasure (SPACE), the first source-free unlearning framework specialized for MLLMs. 段落3:SPACE consists of two stages: (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space. (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts. 这里提到了两个阶段,可以用`
- `来展示,但原文是“(1) ... (2) ...”,如果改成列表,是否算修改内容?为了安全,保留原文的编号,或者用列表形式,把(1)和(2)放在li里。 例如:
SPACE consists of two stages:
这样没有增删改内容,只是改变了排版结构,符合规则3。 段落4:DCSI confines updates to the null space of retained knowledge, ensuring structural integrity. We theoretically prove that SPACE strictly bounds the perturbation on retained knowledge and maximizes feature spectral entropy, thereby maintaining the model's performance. 段落5:Furthermore, extensive experiments across six datasets show that SPACE achieves performance comparable to that of state-of-the-art data-dependent methods, validating its effectiveness in source-free MU scenarios. The source code will be released. 关于“长文用
- (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space.
- (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts.
/
划分小节”:如果原文没有小节标题,强行加标题会违反“不得增删改内容”。所以这里可能不需要硬加`
/
`,或者可以把某些核心句子用``加粗。规则3提到“重要句子可用”。 比如:
In this work, we propose Source-free Proxy Anchor Concept Erasure (SPACE), the first source-free unlearning framework specialized for MLLMs.
检查规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 -> 好的。 2. 每个自然段落用包裹,长文用
/
划分小节。 -> 原文不长,且没有现成标题,不加虚构标题。 3. 列表内容用
- ,引用用
,代码用,重要句子可用。 -> 用
- 处理两个阶段,重要名词用。 4. 保留原文所有信息,不得增删改内容。 -> 严格对照。 5. 不要输出//等文档级标签。 -> 好的。 构建HTML:
As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, machine unlearning (MU) has emerged as a crucial solution for removing sensitive data while preserving model performance. However, existing MU methods typically rely on visual data of the target concepts, which is often unavailable due to strict data retention policies, thus creating a demand for source-free unlearning approaches that operate without access to the target data.
In this work, we propose Source-free Proxy Anchor Concept Erasure (SPACE), the first source-free unlearning framework specialized for MLLMs.
SPACE consists of two stages:
- (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space.
- (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts.
DCSI confines updates to the null space of retained knowledge, ensuring structural integrity. We theoretically prove that SPACE strictly bounds the perturbation on retained knowledge and maximizes feature spectral entropy, thereby maintaining the model's performance.
Furthermore, extensive experiments across six datasets show that SPACE achieves performance comparable to that of state-of-the-art data-dependent methods, validating its effectiveness in source-free MU scenarios. The source code will be released.
检查原文是否完全一致: 原文:"SPACE consists of two stages: (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space. (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts." 我的HTML:SPACE consists of two stages:
这里把冒号后的两句话变成了列表,文字内容没有增删改。为了更贴近原文,也可以保持在同一个
- (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space.
- (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts.
里,但规则3要求“列表内容用
- ”,这明显是列表内容,所以用
是合理的。 再检查加粗部分: 原文:"source-free unlearning approaches" -> source-free unlearning approaches 原文:"Source-free Proxy Anchor Concept Erasure (SPACE)" -> Source-free Proxy Anchor Concept Erasure (SPACE) 没有修改文字,只是加了标签。 确保没有多余的换行和空格,输出纯净的HTML。 不需要
/
,因为原文没有小节标题,硬加会违反规则4。如果非要加,可以加个
引言
,但这属于增加内容。所以不用。 最终检查HTML标签闭合和格式。...