本文中的閃圖複製自北嶺加州州立大學（California State University, Northridge）
網頁https://www.csun.edu/~vcpsy00h/creativity/define.htm
注：文末附交流羣，最近趕ACL，比較忙，很多同學加了沒有回過期了，可以重新加一下，備註好的一定會回覆，敬請諒解。
這是我在1993年寫給一名學生的信，內容涉及他的論文初稿。2003年，我修改了一下這封信，刪除了與該學生相關的具體內容，並將修改後的這封信作爲對所有研究生的額外知識要求。
I wrote this in 1993 as a letter to a student concerning a draft of his dissertation. in 2003 I edited it to remove some specific references to the student and present it as a small increment to the information available to my grad students. --spaf
先讓我以一些看似顯而易見的事情開始。Let me start by reviewing some things that may seem obvious:
第一，你的論文是獲得博士學位要求的一部分。除此之外，研究、理論、實驗等本身也是你獲得博士學位的條件。一個人不會試圖在自己的論文中面面俱到地寫下所有這些。
Your dissertation is part of the requirements for a PhD. The research, theory, experimentation, et al. also contribute. One does not attempt to capture everything in one's dissertation.
第二，博士論文是一種專業（technical）文檔，用於記錄和提出論點的證明。它面向專業領域的讀者，必須清晰而完整，但不一定詳盡無遺。還要注意——實驗數據，如果使用的話，不是證明（proof）——它只是是證據（evidence）。證明應作爲分析和批判性陳述來呈現。一般來說，你論文的每一項陳述要麼是有專業領域文獻支撐的常識，要麼是由學位候選人（您）能提供證明的原創結果。這些陳述必須直接與論點的證明有關，否則就不需要。
The dissertation is a technical work used to document and set forth proof of one's thesis. It is intended for a technical audience, and it must be clear and complete, but not necessarily exhaustively comprehensive. Also note -- experimental data, if used, is not the proof -- it is evidence. The proof is presented as analysis and critical presentation. As a general rule, every statement in your dissertation must be common knowledge, supported by citation to technical literature, or else original results proved by the candidate (you). Each of those statements must directly relate to the proof of the thesis or else they are not needed.
第三，學位論文（dissertation）並不等同於論點（thesis）。一個人的論點只是一個主張——一個假設。論文應詳細描述如何證明這個假設（或者，很少，反駁這一聲明，並呈現出其他重要結果）。
The dissertation is not the thesis. One's thesis is a claim -- a hypothesis. The dissertation describes, in detail, how one proves the hypothesis (or, rarely, disproves the claim and shows other important results).
讓我們重溫一下論點本身這一概念。論點是一個假設，一個猜想，一個定理。論文是一份正式的、程式化的文件，用來爲你的論點辯護。論點必須具有重要的原創性（即未曾有人對它的真實性進行過證明），它必須能拓展科學知識的現有狀態。
Let's revisit the idea of the thesis itself. It is a hypothesis, a conjecture, a theorem. The dissertation is a formal, stylized document used to argue your thesis. The thesis must be significant, original (no one has yet demonstrated it to be true), and it must extend the state of scientific knowledge.
你需要做的第一件事是想出不超過三個句子來表達你的論點。你對論點的陳述既要得到委員會的認同，也要令自己滿意。
The first thing you need to do is to come up with no more than three sentences that express your thesis. Your committee must agree that your statements form a valid thesis statement. You too must be happy with the statement -- it should be what you will tell anyone if they ask you what your thesis is (few people will want to hear an hour presentation as a response).
一旦你有論點陳述，你就可以開始展開論文寫作了。例如，摘要應該是你論點及其如何證明的一頁紙描述。摘要應總結論文成果，強調對科學的貢獻。
Once you have a statement of thesis, you can begin to develop the dissertation. The abstract, for instance, should be a one-page description of your thesis and how you present the proof of it. The abstract should summarize the results of the thesis and should stress the contributions to science made thereby.
以下是一些在編寫/校對時要記住的更一般的提示（原文有不少與計算機專業相關的條目，不具有普適性，因此這裏沒有翻譯——譯者注）：Here are some more general hints to keep in mind as you write/edit:
1. 一般不應該使用副詞——相反，使用精確的語言。例如，不要說某事"發生得很快"。與此類似，應避免使用"快"，"慢"，"完美"，"很快"，"理想"，"很多"等等詞語。
Adverbs should generally not be used -- instead, use something precise. For example, do not say that something "happens quickly." How fast is quickly? Is it relative to CPU speeds? Network speeds? Does it depend on connectivity, configuration, programming language, OS release, etc? What is the standard deviation? As per the above, use of the words "fast", "slow", "perfect", "soon", "ideal", "lots of" and related should all be avoided. So should "clearly", "obviously", "simple", "like", "few", "most", "large", et al.
2. 你所寫的是科學事實。對美學、倫理、個人偏好等的判斷的描述應該放在結論章節中， 如果它們必須出現的話。避免說"事實上"，"實際上"，"實際情況是"，和任何類似的結構 - 你寫的一切必須是事實，所以沒有必要再陳述你寫的是事實。
What you are writing is scientific fact. Judgments of aesthetics, ethics, personal preference, and the like should be in the conclusions chapter if they should be anywhere at all. With that in mind, avoid use of words such as "good", "bad", "best", and any similar discussion. Also avoid stating "In fact," "Actually," "In reality," and any similar construct -- everything you are writing must be factual, so there is no need to state such things. If you feel compelled to use one of these constructs, then carefully evaluate what you are saying to be certain you are not injecting relative terms, opinions, value judgements, or other items that are inappropriate for a dissertation.
3. 避免提及時間和環境。如果特定的時間或間隔很重要，請說明確，如"1905 年至 1920 年之間"，而不是"過去 15 年"。
Avoid mention of time and environment. "Today's computers" are antiques far sooner than you think. Your thesis should still be true many years from now. If a particular time or interval is important, then be explicit about it, as in "Between 1905 and 1920" rather than "Over the last 15 years." (See the difference, given some distance in time?)
4. 確保你聲稱作爲證明的東西都會被任何科學家或數學家認可。Be sure that something you claim as a proof would be recognized as such by any scientist or mathematician.
5. 您和您的論文應該是您討論的主題的最終（當前）權威。因此，不應有"以我們所知"或"就我們所知"的例子。要麼你知道肯定，要麼你不知道——如果你不知道，你就不應該說！
You and your dissertation are supposed to be the ultimate (current) authority on the topic you are covering. Thus, there should be no instance of "to the best of our knowledge" or "as far as we can tell." Either you know for certain, or you don't -- and if you don't know, you shouldn't state it!
6. 重點關注結果而不是研究方法。雖然研究方法應被明確描述，但不應在第三章和第四章中成爲您討論的中心主題。
Focus on the results and not the methodology. Methodology should be clearly described, but not the central topic of your discussion in chapters III & IV
請記住，你——博士學位候選人——有望成爲你研究主題的世界級專家。這一主題不應過於寬泛，但又必須足夠大才有意義。你的導師和委員會成員不應該比你更瞭解這個主題。你的論文應該解釋你的發現，並和答辯一起證明你精通某一領域且你現在是一流的專家。這並不意味着要寫你所知道的一切——它意味着你將最重要的觀點寫得足以讓其他人同意你的結論。Keep in mind that you -- the Ph.D. candidate -- are expected to become the world's foremost expert on your topic area. That topic area should not be unduly broad, but must be big enough to be meaningful. Your advisor and committee members are not supposed to know more about the topic than you do -- not individually, at least. Your dissertation is supposed to explain your findings and, along with the defense, demonstrate your mastery of the area in which you are now the leading expert. That does not mean writing everything you know -- it means writing enough about the most important points that others can agree with your conclusions.
最後，不要落入這樣一個束縛了許多候選人並導致他們中的一些人在完成任務前就被淘汰了的陷阱：你的論點不需要具有革命性。它只需要是該領域的一個漸進式進步。很少有博士論文對該領域有顯著的影響。相反，作者的出版物和產品集可能會改變該領域。
Last of all, don't fall into the trap that ties up many a candidate, and causes some of them to flame out before completion: your thesis does not need to be revolutionary. It simply needs to be an incremental advancement in the field. Few Ph.D. dissertations have ever had a marked impact on the field. Instead, it is the set of publications and products of the author that may change the field.
如果你的論文是像大多數其它論文，那麼它的讀者對象只會是你的委員會和一些尋求在你的工作的基礎上推進一步的其他博士候選人。因此，它不需要是文學的傑作，也不需要解決一個長期存在的問題。它只需要正確，在委員會看來足夠有意義，且需要完整。當你畢業後改變世界時，我們都會鼓掌。在這一點，你會發現，許多著名的科學家在他們的職業生涯裏的工作與博士論文主題完全不同。博士論文用以證明你有能力找到並呈現原創結果；畢業後你的事業和生活將告訴你爲了產生影響還需要考慮哪些其它事情。
If your dissertation is like most, it will only be read by your committee and some other Ph.D. candidates seeking to build on your work. As such, it does not need to be a masterwork of literature, nor does it need to solve a long-standing problem in computing. It merely needs to be correct, to be significant in the judgement of your committee, and it needs to be complete. We will all applaud when you change the world after graduation. And at that you will find that many well-known scientists in CS have made their careers in areas different from their dissertation topic. The dissertation is proof that you can find and present original results; your career and life after graduation will demonstrate the other concerns you might have about making an impact.
所以開始工作吧！So get to work!
節選自：Sparf, What is a Ph.D. Dissertation? https://spaf.cerias.purdue.edu/StudentInfo/spaf.html
機器學習/深度學習算法/自然語言處理交流羣
已建立機器學習算-自然語言處理微信交流羣！想要進交流羣進行學習的同學，可以直接加我的微信號：HIT_NLP。加的時候備註一下：知乎+學校+暱稱 （不加備註不會接受同意，望諒解），想進pytorch羣，備註知乎+學校+暱稱+Pytorch即可。然後我們就可以拉你進羣了。羣裏已經有非得多國內外高校同學，交流氛圍非常好。
強烈推薦大家關注機器學習算法與自然語言處理賬號和機器學習算法與自然語言處理微信公衆號，可以快速瞭解到最新優質的乾貨資源。
推薦閱讀
GPT“高仿”系列開源了！最大可達GPT-3大小，還能自主訓練
我，大學沒畢業，在OpenAI搞AI，想教教你如何提升“研究品味”
賽爾原創@AAAI 2021 | 糾結於聯合學習中的建模方法？快來看看圖網絡顯式建模!
如何提高PyTorch“煉丹”速度？這位小哥總結了17種方法，可直接上手更改的那種
斯坦福CS224W《圖機器學習》2021開課！Jure Leskovec大牛主講，附課程PPT下載
第二十屆中國計算語言學大會（CCL 2021）技術評測任務徵集
IJCAI 2020今日開幕，傑出論文獎、卓越研究獎、約翰·麥卡錫獎等8項大獎公佈！
全面回顧2020年圖機器學習進展，12位大神論道、寄望2021年大爆發！
2020年arXiv十大熱門論文來了！不止GPT-3、SimCLR、YOLOv4...
超過500個附代碼的AI/機器學習/深度學習/計算機視覺/NLP項目
Awesome Transformer for Vision Resources List庫
2020 Top10計算機視覺論文總結：論文，代碼，解讀，還有demo視頻！
156個參考文獻！Visual Transformer 調研survey
加拿大蒙特利爾大學助理教授劉邦招收2021/2022年博士生