My personal notes on writing. Keep updating…

This is a quick summary of Prof. Baochun Li’s talk How to Write Perfect Papers. Another sharing of writing tips is also quite useful, will summarize that later.

Writing Perfect Papers

Story

1. It’s about how we advance the state of the art.
   1. What is the problem.
      1. Following the trend? Could be incremental.
      2. Contrarian, against the trend
   2. What’s new.
      1. First, briefly introduce the state-of-the-art in the literature, solving the same problem you wish to solve.
      2. Talk about what is new in your work.
      3. If the problem is different from previous work - different assumptions for example, provide justifications why this assumption is important and necessary.
   3. How is it better.
      1. theoretical properties or better experimental results.
2. What does a good paper look like?
   1. Rule #1: Be as clear as possible
      1. A small misunderstanding can get your paper rejected.
      2. you need to guide the reviewer into your work, and anticipate what will be difficult to understand or could lead to misunderstanding.
      3. and don’t obfuscate
   2. Rule #2: Convince the reader your work is worth publishing (advancing the state of the art)

Work

1. Read
   1. Become an expert on the problem we want to solve.
   2. How?
      1. Start from a single paper.
      2. Do an expanded-ring search: BFS
         1. cite, cited, same authors
2. Write
   1. write about what you understood
   2. write every day
   3. writing is the best way to force yourself to think clearly and be focused
   4. idea⇒write a paper⇒do research
      1. write related works ⇒ what’s new
      2. write what’s challenging.
      3. write why the problem is important
      4. write what needs to be fixed
   5. Crystallized that you don’t quite understand yet
   6. Writing opens a dialogue, for others to read what you wrote.
3. Feedback loop
   1. You do not sequentially finish your paper from title, abstr, intro, …, conclusion.
   2. Write⇒ get feedback⇒ write ⇒ get feedback …
   3. The importance of feedback
      1. a small misunderstanding can get your paper rejected
      2. it is difficult to read your own paper with the reviewer’s mindset
   4. Each reader can only read your paper for the first time once! Use them carefully.
   5. If reviewers did not understand something, it is because you did not explain it clearly enough.

Art

1. Your paper is the only thing the reviewers (and the readers) see of your work.
2. Pay attention to the title, abstract, intro, and flow of the ideas
   1. Title, 2 line, 10000 readers
      1. as simple and short as possible
      2. the one that is the easiest to understand at a glance
      3. E.g., local sgd converges fast and communicates little.
3. Abstract.
   1. One sentence to state the background.
   2. One sentence to state ⇒ what the problem is? (an interesting and unsolved problem)
   3. Two to four sentences to state the original contributions in the paper⇒what’s new (here is my idea, and my idea works)
   4. One sentence to state how the solution is better, validated using analyses, simulations or experiments.
4. Intro
   1. First impressions are very important.
      1. After reading the intro, the reviewer is likely to have already decided (maybe unconsciously) if they will accept the paper or not (if no technical issues)
   2. Self contained: do not need to read the rest of the paper.
   3. Be clear on what the problem is, why our solution is new and how it’s better.
   4. Don’t do…
      1. Too long.
      2. spend too much space on the background and related work.
      3. Don’t boost the contribution.
   5. Structure
      1. First paragraph: a general overview of the research field, warm up the reader and to prepare for the problem statement, 2-3 sentences.
      2. 1-2 paras, state the problem, and existing solutions
      3. limitations of existing solutions that motivate this paper.
      4. Proposed solution: high level idea
   6. What’s new
      1. why is the proposed solution different from and better than existing solutions?
      2. 1-2 most impressive highlights.
      3. Make the originality of the paper crystal clear and stand out
   7. Hey, if they can really deliver this, that will be very exciting. I will read the rest of the paper.
   8. Include a table.
   9. Use examples: how main idea works in a special case.
   10. For DL paper
       1. 1 para: why is the problem is important
          1. Don’t discuss related work too much here
       2. 2 para: why is the problem difficult
       3. 3 para: what is your key idea (you can point to the teaser figure)
          1. Give the intuition of your idea, get the reader hooked.
       4. 4 para: give important technical details.
       5. 5 para: How do you evaluate. What are most impressive results.
       6. 6 para: a summary of your contribution (using bullets)
5. Flow of ideas
   1. One paragraph contains one complete idea.
   2. Method section:
      1. start with an overview of the section
      2. then, give a general description of the method
      3. end with the technical details
      4. always from the more general to the more detailed explanation
   3. Related work
      1. not a mere description of the state-of-the-art
      2. show you know the state of the art
      3. show your method solves a new problem, or the problem that were not solved perfectly.
6. General rules of thumb for the length
   1. abstr 200 words
   2. intro 1 page
   3. related work half a page

Details

1. Git: track all histories, and LATEX CODE IS CODE.
2. Eliminate grammar and typos
3. Keep a clear structure.
4. Use the active form.
5. Transitional words,
   1. Two half sentences, as since, or else
   2. Two sentences.
      1. However, In addition, Further, Nevertheless, Fortunately, Unfortunately, Surprisingly.
      2. To make matters worse, to further exacerbate the problem the bad news is
      3. The implications are two-fold, It turns out that, As an example, to take … a step further
   3. Sections and paras: It only remains to see …, The simple anser to this question is, the only challenge that remains now is, to address this challenge, we first present, next, we evaluate, we are now ready to.
6. Don’t use long sentences
7. Don’t be too colloquial, too formal, or too emotional
8. Counts and uncounts
   1. A few, a number of, fewer
   2. less, lower
9. Oxford Dictionary of English, Thesaurus
10. Typeset your paper correctly and beautifully
11. draw.io

Write good papers FAST

$\max Q(p) \text{ s.t. } t(p) \le T$

Maximize the quality of your paper, subject to deadline constraint

Know your deadline

1. Work towards your deadline
2. Set your dealine realistically, 1-3 months.

Allocate your time well

1. Think seriously why you miss your mileatone deadlines.
2. Don’t procrastinate and delay all the work to the days before deadline.

Don’t follow, be the lead

1. Start your team at the begining of your project
2. Only invite trustworthy collaborators when building your team
3. Get them to commit their time
4. Warn them the though work ahead along the way
5. Promise them an exciting outcome
6. Remind them about their milestone deadlines often
7. Persist if they forget about your requests
8. Cancel thier membership if they consistently fail
9. Have backup plans

Choose a problem with the right size

1. It must be feasible to be completed before deadline
2. Goal: make solid improvments over one paper, and one paper only
3. Choose a paper you enjoy reading; But if it may take a long time, give up and choose choose another paper
4. Know your problem and pick ahead of time the conference where your paper’s scope fits into