`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。
分析正文内容:
正文是一段连续的学术文本,没有明显的列表或引用、代码。需要将其合理划分为段落,并适当添加标题(``或`
`)以及加粗(``)重要句子。
原文:
Multimodal continual instruction tuning requires multimodal large language models to sequentially acquire new capabilities without catastrophic forgetting. Mixture-of-experts (MoE) architectures offer a promising way to mitigate task interference through expert specialization and combination. However, while existing MoE-based methods avoid overwriting shared parameters by allocating separate experts for each task, this progressive expansion incurs severe cross-task redundancy and linear parameter growth. Specifically, experts across tasks often learn overlapping functions alongside task-specific capabilities. For instance, ScienceQA and TextVQA both locate relevant image regions. To curb parameter redundancy without inducing forgetting, we propose Consolidating and Preserving MoE (CoPre-MoE). For each incoming task, CoPre-MoE first learns task-specific MoE to capture specialized capabilities. To eliminate redundancy and reduce parameter growth across tasks, we design a shared-and-private expert architecture. Specifically, we first evaluate functional overlap to selectively consolidate overlapping experts across tasks into a fixed-capacity generic expert pool. We then factorize the remaining task-specific experts into shared projection bases and compact task-specific cores within a private expert pool. To mitigate forgetting, we construct the router by pairing each expert with learnable retrieval keys, where historical keys remain frozen during continual updates while newly added keys are regularized with an orthogonality penalty. At inference, input-dependent routing dynamically combines both pools without test-time task labels. Extensive experiments demonstrate that CoPre-MoE achieves superior performance while substantially reducing parameter overhead.
划分小节和段落:
可以加一个总标题,比如 `研究背景与挑战
`,然后第一段讲背景和问题。
第二段讲提出的方法。
第三段讲实验结果。
或者直接按自然逻辑分段,加适当的标题。
草稿:
研究背景与现有挑战
Multimodal continual instruction tuning requires multimodal large language models to sequentially acquire new capabilities without catastrophic forgetting. Mixture-of-experts (MoE) architectures offer a promising way to mitigate task interference through expert specialization and combination. However, while existing MoE-based methods avoid overwriting shared parameters by allocating separate experts for each task, this progressive expansion incurs severe cross-task redundancy and linear parameter growth. Specifically, experts across tasks often learn overlapping functions alongside task-specific capabilities. For instance, ScienceQA and TextVQA both locate relevant image regions.
CoPre-MoE 方法设计
To curb parameter redundancy without inducing forgetting, we propose Consolidating and Preserving MoE (CoPre-MoE). For each incoming task, CoPre-MoE first learns task-specific MoE to capture specialized capabilities.
To eliminate redundancy and reduce parameter growth across tasks, we design a shared-and-private expert architecture. Specifically, we first evaluate functional overlap to selectively consolidate overlapping experts across tasks into a fixed-capacity generic expert pool. We then factorize the remaining task-specific experts into shared projection bases and compact task-specific cores within a private expert pool.
To mitigate forgetting, we construct the router by pairing each expert with learnable retrieval keys, where historical keys remain frozen during continual updates while newly added keys are regularized with an orthogonality penalty. At inference, input-dependent routing dynamically combines both pools without test-time task labels.
实验结果
Extensive experiments demonstrate that CoPre-MoE achieves superior performance while substantially reducing parameter overhead.
检查是否增删改内容:
原文1: Multimodal continual instruction tuning requires multimodal large language models to sequentially acquire new capabilities without catastrophic forgetting. Mixture-of-experts (MoE) architectures offer a promising way to mitigate task interference through expert specialization and combination. However, while existing MoE-based methods avoid overwriting shared parameters by allocating separate experts for each task, this progressive expansion incurs severe cross-task redundancy and linear parameter growth. Specifically, experts across tasks often learn overlapping functions alongside task-specific capabilities. For instance, ScienceQA and TextVQA both locate relevant image regions.
原文2: To curb parameter redundancy without inducing forgetting, we propose Consolidating and Preserving MoE (CoPre-MoE). For each incoming task, CoPre-MoE first learns task-specific MoE to capture specialized capabilities.
原文3: To eliminate redundancy and reduce parameter growth across tasks, we design a shared-and-private expert architecture. Specifically, we first evaluate functional overlap to selectively consolidate overlapping experts across tasks into a fixed-capacity generic expert pool. We then factorize the remaining task-specific experts into shared projection bases and compact task-specific cores within a private expert pool.
原文4: To mitigate forgetting, we construct the router by pairing each expert with learnable retrieval keys, where historical keys remain frozen during continual updates while newly added keys are regularized with an orthogonality penalty. At inference, input-dependent routing dynamically combines both pools without test-time task labels.
原文5: Extensive experiments demonstrate that CoPre-MoE achieves superior performance while substantially reducing parameter overhead.
所有文字内容完全一致,没有增删改。
使用了``划分小节,`
`包裹段落,``突出重要句子。 没有Markdown围栏,没有解释,没有文档级标签。 确认格式:
研究背景与现有挑战
Multimodal continual instruction tuning requires...
等等,用户说“不要输出//等文档级标签”,并且“只输出HTML片段本身”。
标题文字(如“研究背景与现有挑战”)算不算“增删改内容”?
规则4说“保留原文所有信息,不得增删改内容”。增加小标题的文字算不算增加内容?
为了严格遵守“不得增删改内容”,我最好不使用带有额外文字的``/`
`,或者用原文中已有的短语作为标题,或者干脆不用标题,只分段。 但是规则2说“长文用
/
划分小节”。如果不加标题文字,可以用`
`吗?这不太语义化。
或者使用原文中的句子作为标题?比如:
Multimodal continual instruction tuning
但这也改变了原文的段落结构。
最安全的做法是:把原文的自然段用``包裹,如果必须用`
`/`
`,可以使用原文中的关键词作为标题,或者不添加额外文字,只用标签包裹原文的某些句子作为标题(但这会改变段落结构)。 其实,规则说“长文用
/
划分小节”,我可以把原文的首句或者核心概念作为`
`,但这样可能会被认为是“改内容”。 为了绝对安全,不添加任何原文没有的文字作为标题,可以将原文按逻辑分成几个`
`,如果不需要强制加`
`,就不加,或者用原文中已有的短语。 比如:
Multimodal continual instruction tuning
requires multimodal... (这样截断句子不好) 还是不用`
`了,只用`
`和``,因为原文并不算特别“长”(只有一段,大概200词),可以拆分成几个自然段。
`,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。 分析正文内容: 正文是一段连续的学术文本,没有明显的列表或引用、代码。需要将其合理划分为段落,并适当添加标题(``或`
`)以及加粗(``)重要句子。 原文: Multimodal continual instruction tuning requires multimodal large language models to sequentially acquire new capabilities without catastrophic forgetting. Mixture-of-experts (MoE) architectures offer a promising way to mitigate task interference through expert specialization and combination. However, while existing MoE-based methods avoid overwriting shared parameters by allocating separate experts for each task, this progressive expansion incurs severe cross-task redundancy and linear parameter growth. Specifically, experts across tasks often learn overlapping functions alongside task-specific capabilities. For instance, ScienceQA and TextVQA both locate relevant image regions. To curb parameter redundancy without inducing forgetting, we propose Consolidating and Preserving MoE (CoPre-MoE). For each incoming task, CoPre-MoE first learns task-specific MoE to capture specialized capabilities. To eliminate redundancy and reduce parameter growth across tasks, we design a shared-and-private expert architecture. Specifically, we first evaluate functional overlap to selectively consolidate overlapping experts across tasks into a fixed-capacity generic expert pool. We then factorize the remaining task-specific experts into shared projection bases and compact task-specific cores within a private expert pool. To mitigate forgetting, we construct the router by pairing each expert with learnable retrieval keys, where historical keys remain frozen during continual updates while newly added keys are regularized with an orthogonality penalty. At inference, input-dependent routing dynamically combines both pools without test-time task labels. Extensive experiments demonstrate that CoPre-MoE achieves superior performance while substantially reducing parameter overhead. 划分小节和段落: 可以加一个总标题,比如 `
研究背景与挑战
`,然后第一段讲背景和问题。 第二段讲提出的方法。 第三段讲实验结果。 或者直接按自然逻辑分段,加适当的标题。 草稿:研究背景与现有挑战
Multimodal continual instruction tuning requires multimodal large language models to sequentially acquire new capabilities without catastrophic forgetting. Mixture-of-experts (MoE) architectures offer a promising way to mitigate task interference through expert specialization and combination. However, while existing MoE-based methods avoid overwriting shared parameters by allocating separate experts for each task, this progressive expansion incurs severe cross-task redundancy and linear parameter growth. Specifically, experts across tasks often learn overlapping functions alongside task-specific capabilities. For instance, ScienceQA and TextVQA both locate relevant image regions.
CoPre-MoE 方法设计
To curb parameter redundancy without inducing forgetting, we propose Consolidating and Preserving MoE (CoPre-MoE). For each incoming task, CoPre-MoE first learns task-specific MoE to capture specialized capabilities.
To eliminate redundancy and reduce parameter growth across tasks, we design a shared-and-private expert architecture. Specifically, we first evaluate functional overlap to selectively consolidate overlapping experts across tasks into a fixed-capacity generic expert pool. We then factorize the remaining task-specific experts into shared projection bases and compact task-specific cores within a private expert pool.
To mitigate forgetting, we construct the router by pairing each expert with learnable retrieval keys, where historical keys remain frozen during continual updates while newly added keys are regularized with an orthogonality penalty. At inference, input-dependent routing dynamically combines both pools without test-time task labels.
实验结果
Extensive experiments demonstrate that CoPre-MoE achieves superior performance while substantially reducing parameter overhead.
检查是否增删改内容: 原文1: Multimodal continual instruction tuning requires multimodal large language models to sequentially acquire new capabilities without catastrophic forgetting. Mixture-of-experts (MoE) architectures offer a promising way to mitigate task interference through expert specialization and combination. However, while existing MoE-based methods avoid overwriting shared parameters by allocating separate experts for each task, this progressive expansion incurs severe cross-task redundancy and linear parameter growth. Specifically, experts across tasks often learn overlapping functions alongside task-specific capabilities. For instance, ScienceQA and TextVQA both locate relevant image regions. 原文2: To curb parameter redundancy without inducing forgetting, we propose Consolidating and Preserving MoE (CoPre-MoE). For each incoming task, CoPre-MoE first learns task-specific MoE to capture specialized capabilities. 原文3: To eliminate redundancy and reduce parameter growth across tasks, we design a shared-and-private expert architecture. Specifically, we first evaluate functional overlap to selectively consolidate overlapping experts across tasks into a fixed-capacity generic expert pool. We then factorize the remaining task-specific experts into shared projection bases and compact task-specific cores within a private expert pool. 原文4: To mitigate forgetting, we construct the router by pairing each expert with learnable retrieval keys, where historical keys remain frozen during continual updates while newly added keys are regularized with an orthogonality penalty. At inference, input-dependent routing dynamically combines both pools without test-time task labels. 原文5: Extensive experiments demonstrate that CoPre-MoE achieves superior performance while substantially reducing parameter overhead. 所有文字内容完全一致,没有增删改。 使用了``划分小节,`
`包裹段落,``突出重要句子。 没有Markdown围栏,没有解释,没有文档级标签。 确认格式:
研究背景与现有挑战
Multimodal continual instruction tuning requires...
等等,用户说“不要输出//等文档级标签”,并且“只输出HTML片段本身”。 标题文字(如“研究背景与现有挑战”)算不算“增删改内容”? 规则4说“保留原文所有信息,不得增删改内容”。增加小标题的文字算不算增加内容? 为了严格遵守“不得增删改内容”,我最好不使用带有额外文字的``/`
`,或者用原文中已有的短语作为标题,或者干脆不用标题,只分段。 但是规则2说“长文用
/
划分小节”。如果不加标题文字,可以用`
`吗?这不太语义化。 或者使用原文中的句子作为标题?比如:Multimodal continual instruction tuning
但这也改变了原文的段落结构。 最安全的做法是:把原文的自然段用``包裹,如果必须用`
`/`
`,可以使用原文中的关键词作为标题,或者不添加额外文字,只用标签包裹原文的某些句子作为标题(但这会改变段落结构)。 其实,规则说“长文用
/
划分小节”,我可以把原文的首句或者核心概念作为`
`,但这样可能会被认为是“改内容”。 为了绝对安全,不添加任何原文没有的文字作为标题,可以将原文按逻辑分成几个`
`,如果不需要强制加`
`,就不加,或者用原文中已有的短语。 比如:
Multimodal continual instruction tuning
requires multimodal... (这样截断句子不好) 还是不用`
`了,只用`
`和``,因为原文并不算特别“长”(只有一段,大概200词),可以拆分成几个自然段。