請(qǐng)教一下如何寫學(xué)術(shù)論文（會(huì)議論文，以CVPR等為例）的rebuttal？特別是reviewer說你的方法很ad hoc ? 有成功過的人士可以指點(diǎn)一下嗎？

電光幻影煉金術(shù)（香港中文大學(xué) CS PhD在讀）回答：

目的和預(yù)期成效

rebuttal的目的：首要的目的是說服AC，其次目的是說服中間的審稿人，保住正面審稿人，再次的目的是說服負(fù)面審稿人。

對(duì)于分?jǐn)?shù)不太理想的同學(xué)，rebuttal是通往下次投稿的土壤。我的節(jié)目叫從審稿到中稿，一個(gè)核心的觀點(diǎn)就是中稿的過程是很平滑的，可以逐次提升的。暫時(shí)意見不好不要灰心。

對(duì)于新手而言，不要怕堅(jiān)持投稿/recycle

明確以下是核心問題：Novelty，能否超過baseline，政策問題（抄襲，重投），實(shí)驗(yàn)對(duì)錯(cuò)的問題。總而言之就是一個(gè)硬指標(biāo)：能否推動(dòng)前人的研究成果。

明確一些是次要問題：部分超參數(shù)，比較的一些小模塊，一些懷疑。

2. 態(tài)度和口吻

不要起狀態(tài)，展現(xiàn)出自己是合理和理性的人。

不要質(zhì)疑審稿人的能力。

如果說審稿人不理解/文章有不清楚的地方，解釋就好了。

即便很傻，問到也要回答。

不要總是道歉。但是也是可以道歉的：概念性的問題/寫作的問題。AC往往是比較重視概念的。

不要過分傾向于積極的審稿人。不要把積極的審稿人和消極的審稿人對(duì)立起來。直接針對(duì)最客觀最本質(zhì)的問題回答。不要耍政治斗爭(zhēng)/狼人殺的小聰明。
經(jīng)常出現(xiàn)的情況，積極的審稿人說寫作很好，消極的審稿人說寫作根本看不懂。一定是有則改之，無則加勉。

避免抨擊審稿人。比如說審稿人犯了事實(shí)性的錯(cuò)誤，那可能會(huì)完全失去這個(gè)審稿人。可以直接把事實(shí)講出來。

避免說審稿人說錯(cuò)了之類的，直接把正確的東西說出來就對(duì)了。

盡量多引用一些文章，引經(jīng)據(jù)典。

3、關(guān)于寫作

要有一個(gè)清晰的結(jié)構(gòu)，很容易定位信息。

短小精悍。

直面問題，不要指代。反面例子：審稿人問：novelty是什么？作者回答：novelty在方法部分。

正確的回答：就直接說novelty是XX模塊，別人沒用過。或者說novelty是提出并解決了XX問題，別人沒做過。

反面例子：審稿人問：實(shí)驗(yàn)為什么這么設(shè)置？作者回答：看Sec3.5。
正確的回答：把來龍去脈用兩三句話講一下，我這么設(shè)置是因?yàn)閯e人的文章也這么做了。或者說在驗(yàn)證集中比較好的參數(shù)。

了解審稿人的隱含義。比如說上面這個(gè)例子，審稿人的隱含義：別人的參數(shù)都不這么設(shè)置的。

請(qǐng)人幫忙讀一下/改一下rebuttal。

Grammarly。一般來講grammarly改完之后比人要好。

要把負(fù)面的東西正面化。

共性的問題可以提煉出來回答。

不要說final version，要說revised version。

不要說we acknolwledge you / we inform you，要說we agree

盡量回答所有的問題，至少要清晰的回答主要的問題，并且不違背其他原則。

審稿人的標(biāo)號(hào)要清晰。

老問題：是按照審稿人組織還是按照問題組織。我的經(jīng)驗(yàn)是先回答重要的共性的問題，然后按照審稿人組織。

不要有太多的結(jié)論和前綴。

把容易勝利的點(diǎn)和重要的點(diǎn)放在前面。

寫作完后最后過一遍。

保證每句話都是self-contained的。

4. 關(guān)于實(shí)驗(yàn)這塊

rebuttal想翻盤全靠實(shí)驗(yàn)，原地干拔（指干吵架，空口解釋）根本行不通。

補(bǔ)充的實(shí)驗(yàn)不理想怎么辦？不理想不怕，看對(duì)你文章是不是致命打擊。

如果審稿人提的實(shí)驗(yàn)已經(jīng)有，可以指出，但是要有概括。不能說你看XX地方去，要把設(shè)計(jì)實(shí)驗(yàn)->發(fā)現(xiàn)-->排除其他原因>結(jié)論的邏輯捋清楚。

5. 常見的坑點(diǎn)

不要說看起來容易做起來難。

不要發(fā)牢騷。

盡量不要舉報(bào)審稿人 (ac letter)。（1）AC本身就是負(fù)面的審稿人。（2）負(fù)面審稿人是AC邀請(qǐng)的。除非很明顯的問題：deep learning is just some tricks/ is not good as some arxiv paper.

避免想象，要有事實(shí)依據(jù)。比如說我的文章十年之后會(huì)有影響力/我的文章之后會(huì)解決這個(gè)問題。Don't do science fiction。

不要作不合理的推測(cè)。比如審稿人是瞧不起中國(guó)人/瞧不起做這個(gè)領(lǐng)域的人/這個(gè)審稿人之前審過我的文章。

6. 外部資源-模板-示范

http://www.siggraph.org/s2009/submissions/technical_papers/faqs.php#rebuttal

Rebuttal 模版

We sincerely appreciate the constructive suggestions from reviewers.

General response: First, new experiments on Atari found that our methods can improve both A3C and TreeQN on 37 (32 added) out of 50 (44 added) games while achieving comparable results on other games. Second, our main contribution is to show that one can abstract routines from a single demonstration to help policy learning. The major difference between LfD approaches and ours is typical LfD methods do not learn reusable and interpretable routines. Third, the most similar work in the LfD domain is ComPILE (Kipf et al. 2019), but we significantly outperform them in data efficiency. Furthermore, we added experiments to generalize our approach to the continuous-action domain.

1. Experiments on Atari games (R2, R4, R5)

We tested on 44 more Atari games using the setting in Sec 4.2 and outperformed the baselines. The relative percentage improvement brought by routines for TreeQN / A3C is (calculated by (our score - baseline score) / baseline score * 100):

Ami:13/21, Ass:8/18, Asterix:12/-3, Asteroids:8/2, Atl:-1/18, Ban:4/-3, Bat:10/-1, Bea:23/7, Bow:3/0, Box:-4/19, Bre:16/6, Cen:-9/10, Cho:13/17, Dem:19/7, Dou:20/0, Fish:0/5, Fre:8/0, Fro:9/17, Gop:11/5, Gra:2/4, Hero:11/7, Ice:-1/0, James:21/15, Kan:18/-7, Kung-Fu:10/26, Monte:3/2, Name:17/13, Pitfall:-4/-11, Pong:4/9, Private:19/15, River:21/30, Road:15/19, Robot:1/0, Seaquest:-9/-3, Space:24/19, Star:29/19, Ten:0/-1, Time:23/9, Tutan:18/6, UpN:7/16, Vent:9/17, Video:23/26, Wizard:13/9, Zax:-1/-7.

2. Baselines of LfDs (R2, R5)

We included an LfD baseline, behavior cloning, in Sec 4.1.2. We further compared ours to ComPILE based on their open-source implementation. We trained a ComPILE model on one single demonstrationon Atari and tested under the Hierarchical RL setting (other details are the same as their paper’s Sec 4.3). We found that their approach deterioratesthe A3C baseline on 50 Atari games from 5% to 34% (13% on average). We speculate that this is because their model relieson a VAE to regenerate the demonstration, and the model over-fits to the few states appeared in the available demonstration. Even on the simple grid-world cases presented in their paper, they need 1024 instances of demonstrationsto train the VAE, which reveals the data inefficiency of their approach.

3. Continuous-action environments (R4)

We have extended our methods to high-dimensional continuous-action cases. Specifically, we conducted experiments on the challenging car-racing environment TORCS. We define routines as components in the programmatic policy (refer to PROPEL from Verma et al. 2019). In the routine proposal phase, we do not use Sequitur but use an off-the-shelf program synthesis introduced by Verma et al. 2018. We also evaluate routines by frequency and length. Note that too similar routines would be counted as the same when calculating frequency. During routine usage, those abstracted routines can not only accelerate search but also improve Bayesian optimization. On the most challenging track AALBORG, by combining our proposed routine learning policy, our car could run 16% faster than the SOTA PROPELPROG (from 147 sec/lap to 124 sec/lap) and reduce the crashing ratio from 0.12 to 0.10.

4. Size of action space vs. training speed and performance (R2, R4)

By controlling the number of selected routines (~3 routines), training could still be accelerated, not slowed. We have demonstrated this in Supplementary Sec 3.1.3.

5. CrazyClimber for A3C (R2, R4)

In practice, one may use small budget training (e.g., train PPO for 1M steps) to decide which routines to add to the action space. We found that this technique can help improve the A3C baseline for 5% on CrazyClimber.

6. Prior knowledge on when to use routines (R2)

We identified all routines from the demonstration and added a loss to map the corresponding state to routine (as used in DQfD). In Qbert, Krull, and CrazyClimber, the relative percentage improvement of this is -6, 10, -4. We will add a paragraph to illustrate this, and we are looking forward to better ways to learn when to use routines.

7. Show usage of routine scores (R5)

We have included examples of routine abstraction in Supplementary Table 1. We would specify scoring examples in routine abstraction in the revision.

8. Baseline PbE (R4)

Although the PbE does not directly fetch proposals from demonstrations, it leverages the demonstration to evaluate routines and choose the best ones. We will provide detailed routine scores to illustrate this.

9. Figure/text suggestions (R4)

Thanks. We will revise the teaser figure, texts, and captions accordingly.

涂存超（清華大學(xué)?計(jì)算機(jī)科學(xué)PhD）回答：

作為一名高年級(jí)博士生，會(huì)議rebuttal的經(jīng)歷頗為豐富，也在這個(gè)過程中總結(jié)出了些許經(jīng)驗(yàn)，供各位參考。

1. Rebuttal的基本格式

一般rebuttal都有比較嚴(yán)格的篇幅要求，比如不能多于500或600個(gè)詞。所以rebuttal的關(guān)鍵是要在有限的篇幅內(nèi)盡可能清晰全面的回應(yīng)數(shù)個(gè)reviewer的關(guān)注問題，做到釋義清楚且廢話少說。目前我的rebuttal的格式一般如下所示：

其中，不同reviewer提出的同樣的問題可以不用重復(fù)回答，可以直接"Please refer to A2 to reviewer#1"。結(jié)構(gòu)清晰的rebuttal能夠?qū)eviewer和area chair提供極大的便利，也便于理解。

2. Rebuttal的內(nèi)容

Rebuttal一定要著重關(guān)注reviewer提出的重點(diǎn)問題，這些才是決定reviewer的態(tài)度的關(guān)鍵，不要嘗試去回避這種問題。回答這些問題的時(shí)候要直接且不卑不亢，保持尊敬的同時(shí)也要敢于指出reviewer理解上的問題。根據(jù)我的審稿經(jīng)驗(yàn)，那些明顯在回避一些問題的response只會(huì)印證自己的負(fù)面想法；而能夠直面reviewer問題，有理有據(jù)指出reviewer理解上的偏差的response則會(huì)起到正面的效果。（PS: 如果自己的工作確實(shí)存在reviewer提出的一些問題，不妨表示一下贊同，并把針對(duì)這個(gè)問題的改進(jìn)列為future work)

面對(duì)由于reviewer理解偏差造成全部reject的情況，言辭激烈一點(diǎn)才有可能引起Area Chair的注意，有最后一絲機(jī)會(huì)，當(dāng)然，最基本的禮貌還是要有，不過很有可能有負(fù)面的效果，參考今年ICLR LipNet論文rebuttal https://openreview.net/forum?id=BkjLkSqxg。

3. Rebuttal的意義

大家都知道通過rebuttal使reviewer改分的概率很低，但我認(rèn)為rebuttal是一個(gè)盡人事的過程，身邊也確實(shí)有一些從reject或borderline通過rebuttal最終被錄用的例子。尤其像AAAI/IJCAI這種AI大領(lǐng)域的會(huì)議，最近兩年投稿動(dòng)則三四千篇，這么多reviewer恰好是自己小領(lǐng)域同行的概率很低，難免會(huì)對(duì)工作造成一些理解上的偏差甚至錯(cuò)誤，此時(shí)的rebuttal就顯得特別重要。所以對(duì)于處于borderline或者由于錯(cuò)誤理解造成低分的論文，一定!一定!一定!要寫好rebuttal!

---

最后貼一下LeCun在CVPR2012發(fā)給pc的一封withdrawal rebuttal鎮(zhèn)樓（該rebuttal被pc做了匿名處理），據(jù)說促成了ICLR的誕生，希望自己以后也有寫這種rebuttal的底氣：）

Hi Serge,

We decided to withdraw our paper #[ID no.] from CVPR "[Paper Title]" by [Author Name] et al.
We posted it on ArXiv: http://arxiv.org/ [Paper ID] .

We are withdrawing it for three reasons: 1) the scores are so low, and the reviews so ridiculous, that I don't know how to begin writing a rebuttal without insulting the reviewers; 2) we prefer to submit the paper to ICML where it might be better received; 3) with all the fuss I made, leaving the paper in would have looked like I might have tried to bully the program committee into giving it special treatment.

Getting papers about feature learning accepted at vision conference has always been a struggle, and I've had more than my share of bad reviews over the years. Thankfully, quite a few of my papers were rescued by area chairs.

This time though, the reviewers were particularly clueless, or negatively biased, or both. I was very sure that this paper was going to get good reviews because: 1) it has two simple and generally applicable ideas for segmentation ("purity tree" and "optimal cover"); 2) it uses no hand-crafted features (it's all learned all the way through. Incredibly, this was seen as a negative point by the reviewers!); 3) it beats all published results on 3 standard datasetsfor scene parsing; 4) it's an order of magnitude faster than the competing methods.

If that is not enough to get good reviews, I just don't know what is.

So, I'm giving up on submitting to computer vision conferences altogether. ?CV reviewers are just too likely to be clueless or hostile towards our brand of methods. Submitting our papers is just a waste of everyone's time (and incredibly demoralizingto my lab members)

I might come back in a few years, if at least two things change:
- Enough people in CV become interested in feature learning that the ?probability of getting a non-clueless and non-hostile reviewer is more ?than 50% (hopefully [Computer Vision Researcher]'s tutorial on the topic at CVPR will have some positive effect).
- CV conference proceedings become open access.

We intent to resubmit the paper to ICML, where we hope that it will fall in the hands of more informed and less negatively biased reviewers (not that ML reviewers are generally more informed or less biased, but they are just more informed about our kind of stuff). Regardless, I actually have a keynote talk at [Machine Learning Conference], where I'll be talking about the results in this paper.

Be assured that I am not blaming any of this on you as the CVPR program chair. I know you are doing your best within the traditional framework of CVPR.

I may also submit again to CV conferences if the reviewing process is fundamentally reformed so that papers are published before they get reviewed.

You are welcome to forward this message to whoever you want.

I hope to see you at NIPS or ICML.

Cheers,

-- [Author]

熊風(fēng)（香港科技大學(xué)?計(jì)算機(jī)）回答：

前面的各個(gè)回答已經(jīng)說得很好了。我結(jié)合自己這次ICCV (2017)的經(jīng)歷補(bǔ)充一下看法。

之前看到結(jié)果是borderline, weak reject, borderline的時(shí)候, 心都涼了。覺得這次肯定沒戲了，一個(gè)accept也沒有。自己當(dāng)時(shí)心態(tài)上都有點(diǎn)破罐子破摔，rebuttal寫的比較隨便。但是@啥都不會(huì)

學(xué)長(zhǎng)還是很認(rèn)真地一字一句幫我改，指導(dǎo)我應(yīng)該補(bǔ)充哪些地方，強(qiáng)調(diào)哪些地方。最后rebuttal果然奏效，paper中了。而且reviewer對(duì)我們的rebuttal給了好評(píng): “The authors provided a sufficient rebuttal to solve most of my concerns.” “Two of the three reviewers recommend accept as a poster and agree that the rebuttal addresses their concerns.”

我覺得首先應(yīng)該要認(rèn)清reviewer的思路。

其實(shí)reviewer想拒你的理由，也不一定就是他給出的那些comments. 有時(shí)候可能就是做的東西不符合他的口味，就是想拒，但又找不出明顯的硬傷，然后就挑一堆有的沒的問題。讓這種reviewer改意見感覺挺難的，他心目中對(duì)你論文的看法早就定了。

我這次投的ICCV，給weak reject的那個(gè)reviewer就是這種情況。提的一些問題根本不在點(diǎn)上，基本沒有評(píng)論我的方法本身，凈挑一些dataset, evaluation metrics的問題，盡管我們用的dataset, evaluation metrics是大家普遍都在用的。這個(gè)reviewer到最后也沒修改意見，還是給了weak reject.

但其他兩個(gè)給borderline的reviewer，提的一些問題就靠譜了很多。有的確實(shí)是我們論文的問題。而且從一些話里也可以感受到他們的態(tài)度也是搖擺不定的，比如“some issues in the paper should be fixed before being fully accepted”. ?遇到這種情況，就要果斷爭(zhēng)取。我們r(jià)ebuttal主要就是盡可能地把這兩個(gè)reviewer的concerns解釋清楚。最后這兩個(gè)reviewer也修改了意見。

如果是論文沒寫好，比如遺漏了一些關(guān)鍵細(xì)節(jié)。大方承認(rèn)，并且在rebuttal里面把這些細(xì)節(jié)展現(xiàn)出來，承諾在論文之后的版本會(huì)加上。

如果是reviewer對(duì)你的論文理解有問題，那就在rebuttal里面更加詳細(xì)地解釋。

如果確實(shí)指出了自己模型/實(shí)驗(yàn)的一些瑕疵，可以強(qiáng)調(diào)自己論文的亮點(diǎn)和contribution. 比如有個(gè)reviewer說"the proposed method cannot beat other state of the arts", ?我們可以強(qiáng)調(diào)我們并沒有像其他論文那樣用ensemble的方法來提升performance，我們的focus是XXX. ?比起performance的一些細(xì)微提升，XXX更重要。

上面就是自己的一些經(jīng)驗(yàn)，分享給大家看看就好。可能也不一定正確，可能就是這次rebuttal我們運(yùn)氣不錯(cuò)而已。

最后祝大家都好運(yùn)。

文章轉(zhuǎn)載自知乎，著作權(quán)歸屬原作者

——The ?End——

分享

收藏

點(diǎn)贊

在看

總結(jié)

以上是生活随笔為你收集整理的如何写学术论文的rebuttal？的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。