在计算机顶会投稿中，收到“Rebuttal”（作者反驳/申辩）机会意味着你的论文还没被判死刑。但窗口期通常只有 3-7天，且篇幅极短（通常限制在1页PDF或5000字符以内）。
如何在螺蛳壳里做道场？本文将拆解顶会Rebuttal的底层逻辑，并提供可直接套用的审稿人回复模板。
底层逻辑：Rebuttal不是吵架，是“精准拆弹”
审稿人每天要处理几十篇Rebuttal。你的目标不是说服审稿人“我比爱因斯坦牛”，而是消除误解、补全漏洞、给AC（领域主席）一个录取你的理由。
核心策略：
- 二八定律： 80%的篇幅回应最致命的Weakness（通常是Reject的那位审稿人），20%回应Minor Issues。
- 分类处理： 将审稿人意见分为“误解类”、“确实缺失类”、“恶意挑刺类”。
- 可视化： 能用表格对比的，绝不用大段文字。
审稿人回复通用结构模板（直接复制修改）
模板 1：回应“创新性不足” (Novelty Concern)
审稿人意见： The novelty is limited. The proposed method is a simple combination of A and B.
回复模板：
We respectfully disagree. While [Method A] and [Method B] are known, the key insight of our work is [你的核心洞察，例如：发现A在B场景下的梯度消失问题]。 Unlike prior works that [简述前人做法], our method [你的独特做法] achieves [具体效果].
New Evidence: We have added Table X in the revised version, comparing with [最接近的SOTA方法]. Our method improves [指标] by [具体数值] with [更少的参数量/更快的速度]. This demonstrates the non-trivial technical contribution beyond simple combination.
模板 2：回应“实验不充分/缺少对比” (Missing Baselines/Experiments)
审稿人意见： The paper lacks comparison with recent methods like [Method X] and [Method Y].
回复模板：
We thank the reviewer for this constructive suggestion. We have now conducted experiments on [数据集] comparing with [Method X] and [Method Y].
Result: As shown in Table R1 below, our method outperforms [Method X] by +2.3% and [Method Y] by +1.8% in terms of [指标]. Due to the page limit, we will include the full table in the camera-ready version.
(注：如果没时间跑完，就贴出训练曲线图或部分结果，并承诺Camera-ready补上)
模板 3：回应“写作差/语法错误” (Clarity Issues)
审稿人意见： The paper is hard to follow. There are many typos.
回复模板：
We apologize for the confusion. We have carefully revised the entire manuscript to improve clarity. Specifically:
1. Section 3: We rephrased the definition of [公式/概念] to avoid ambiguity.
2. Figure 2: We enlarged the font and added clearer captions.
3. Typos: We have fixed all typos pointed out by the reviewer.
We hope the revised version meets the standard of clarity.
模板 4：回应“审稿人误解了你的方法” (Misunderstanding)
审稿人意见： The authors claim X, but their method actually does Y.
回复模板：
We appreciate the reviewer for pointing this out. However, we would like to clarify a misunderstanding. In our paper, X is defined as [精确定义]. The reviewer might be referring to [另一个概念].
To avoid future confusion, we have added a footnote/paragraph in Section 2.2 to explicitly distinguish between [概念A] and [概念B]. Our core claim remains valid because [简短逻辑链].
模板 5：回应“审稿人要求补充无法完成的实验” (Impossible Requests)
审稿人意见： Please run experiments on a 1000-layer model.
回复模板：
We thank the reviewer for the suggestion. We agree this is an interesting direction. However, training a 1000-layer model requires [具体资源：如 512 TPUs for 30 days], which is infeasible within the rebuttal period.
Alternative Evidence: To address the reviewer's underlying concern about scalability, we instead conducted a scaling study on [较小规模]. As shown in Figure R2, our method shows a consistent trend. We will leave the extreme-scale validation as future work.
高分Rebuttal的排版技巧（黄金公式）
不要写成纯文本！AC看的是证据。
推荐格式（Markdown/LaTeX）：
Q1: [复制审稿人问题核心句]
A1: [一句话总结回复]
Evidence: [表格/公式/引用]
Changes: [一句话说明修改位置]
示例排版：
Q2.1: Comparison with [Method X] is missing.
A: We have added the comparison.
Table R1: Comparison on ImageNet.
| Method | Top-1 Acc | Params |
|---|
| Method X | 78.5% | 50M |
| Ours | 80.2% | 45M |
Change: Table 3 in main paper.
绝对不能踩的“自杀式”雷区
1. 不要怼审稿人智商。
- 错误： "The reviewer clearly doesn't understand our method."
- 正确： "We apologize for the unclear presentation. We will clarify that..."
2. 不要只回复“We will add this in the final version”。
- 除非你附上了部分结果或详细的修改计划，否则这就是空头支票。顶会Rebuttal是结果导向的。
3. 不要忽略“评分最低”的审稿人。
- AC往往看的是最低分是否被拉起来。集中火力回应给Weak Reject/Oral的审稿人。
4. 不要使用情绪化词汇。
- 避免： ridiculous, absurd, obviously wrong.
- 使用： respectfully, we humbly argue, it seems there is a misunderstanding.
