引用: David Salesin 谈如何写论文

Part 1: Choosing Research Topic
1) Choose the right topic is the most important thing. It's like when you are making an animation, the story is the most important thing.
Right Topic > Research > Writing the Paper
2) Leave the Main Stream
There are scientists who stay in eddies of the flow, refining and iterating on the existed problems, and spend most of their career life in eddies.
If you stay in these eddies, your will endanger your work at larger risks to be scooped, which means you are doing very similar works.
3) Choose something new, develop new areas
4) Get a BIG problem, something that can make big changes.
5) Asking, is it an Important Problem, will it make a big change if I solved it?
6) If it's important, asking another question, do I Have a Good Idea to solve it?
If the answer is yes, which means you have confidence in solving the problem. Also, the problem can not be too easy to solve, or it will be a trivial problem.
7) How original is the problem?
8) How well-defined is the problem?
9) How interested am I in it?
10) When facing a problem, do you have a “Secret Weapon”, or advantages, compared with others? David's formal artistic study serves as his Secret Weapon.
11) Answer the above questions in WRITING, which is very important.
12) Pick up DIFFICULT problems, which are now RIPE problems.
They may be unsolvable problem in the past, but not now, as the research work in the whole area went ahead.
13) Think about this, why hasn't this research problem been solved or even thought before?
14) Don't be afraid that you are not an “EXPERT”, you have the beginner's view, and that's a big advantage.

Part 2: Doing the Research
1) Well define the problem, and split it into sub-problems.
2) Make the work going
In the beginning,
a. Use the Synthetic Data (with little noise)
b.Lower the Dimention
c. Make it working first
3) Middle Game:
a) If it goes on well, problem has legs, and you are on a flow, just let it guide you!
b) If not, doing code maintenance, and write down your progress and problems.
4) End Game:
a) Generating & Polishing the results
b) Brainstorming about the real “DEMO KILLERS”
c) Getting rid of the hacks that you don't have to write about.
5) During the Whole Game,
Keep a Notebook, and write up you work as you go alone
A Log will help you keep track of all the problems on the way.
6) Managing your emotions
The ups and downs, this is research…
7) Managing your advisor (^_^)
Write up the discussion with your mentor immediately after the talk.
You may have several advisors, if they guide you into different directions, do not worry, it will provide you more choices and possibilities.

Part 3: Writing the Paper
1)     Start the writing early, it helps you identify the “hacks”, and helps you form the craft.
2)     Give the reader an opportunity to appreciate the difficulty depth of the work, don't make it plain.
3)     The paper, as short as possible, and easy to understand
4)     Don't bore the readers.
5)     Use a story board for your paper, of how it goes.
6)     所有的公式和符号，要有清晰明了的解释。
7)     即使将文章中所有的公式去掉，读者仍然可以明白你的工作
8)     关于文章的语气，要使用We, 说We的时候，让对方感到说得就是“The Reader and I”, 比“The authors”要好，要带动读者，让他跟随你的思路
9)     避免被动语气，多用主动句式。
10)   论文的标题十分重要
Make sure the Title cover on two things:
Your work and the difference of your work with others
11) 作者署名
David选择co-author的几个原则：
a.     People has regular meetings with me discussing the problem
b.     Coder who paid huge efforts
c.      Regular but not so frequent meetings
12)   写论文的Abstract
Make it the essence, make it tense!
第一句点题，讲清楚这篇论文的重点
第二句，summarize your approach
第三句，mention your contributions
最后，一句做 take away
13) 论文的Introduction
The first sentence of Introduction triggers the readers' imagination…
Introduce your work in both ways：1. plain and informal 2. formal
结构:
1．          The difficulty of the problem
2．          Basic approach
3．          Contribution
4．          Applications of your approach
另要注意欲扬先抑，文章的开始部分不要过分渲染方法的出色，保持朴素节制的介绍，把最吸引人的部分放在结果部分中展示。如果顺序相反，则很可能让读者失望, ^_^。
14）论文的Related Word部分
1. Failings. 讲讲以前工作所无法解决的问题。但是不要批评前人的工作。
2. Complimentary work. 介绍具有互补性质的工作
3. Work as Building Blocks. 介绍一下前人的基础工作
15）Minimize jargon, make the paper easy to understand, even to readers from other research area.
16) List the Truth. But not praise your own work, say, “We've implemented an elegant solution…” :P
That's all the notes I took today. :)       
Salesin先生的Talk和Slides做的非常尽心，许多细节上的东西，听了很受启发。
讲座中间还发生了一个小插曲。当时Salesin先生正在解释Title对于一篇文章是多么的重要，忽然“哐”的一声，一块天花板垂下来做钟摆运动，而单词Title，就正好打在那块板上面，大家都乐翻了。Salesin利索的跳上桌子，试图把板儿装回去。他很瘦，蓄着大胡子，看上像个远航船上的希腊水手。一番努力未果，只好任那块凑热闹的天花板悬着，晃晃悠悠。
我检索了他的名字，除了华盛顿大学的主页，IMDB上也能找到，出现在三部影片的档案中。