Jia-Bin Huang Profile picture
Sep 13, 2021 β€’ 15 tweets β€’ 6 min read β€’ Read on X
How to write a paper that looks like a good one?

You worked super hard and did great research, but somehow the reviewer 2 just doesn't buy it. Why? πŸ€”

It's probably because your paper does not look like a good paper *visually*. πŸ™„

How? πŸ‘‡πŸ‘‡πŸ‘‡ #AcademicTwitter
*Figure 1: WHAT did you do?*

Nothing is more frustrating than not being able to figure out what the paper is about until page 5. 😠Show a TEASER figure on the very first page highlighting the inputs/outputs/key findings.
*Figure 2: WHY did you do it?*

Motivate and justify the key insights/ideas of your work. It is often helpful to illustrate this more clearly by
1) SIMPLIFYING with a toy example and
2) CONTEXTUALIZING with prior work.
*Figure 3: HOW did you do it?*

Show an *overview* figure on how your method works. Label everything so that it provides a clear roadmap of the entire paper.
*Move figures/tables to the top*

Add "[!t]" parameter to your figure/table so that LaTeX will try placing them on the top of the page. Why?

Figures/tables are much easier to understand than reading plain texts. Moving them to the top helps readers quickly understand your work.
*Self-contained figure/table caption*

Whatever you want to say for the figure/table, say them in the caption. It's annoying to find and match the corresponding texts describing the figure/table in your paper.

More on avoiding mental correspondence:
*Concise notations*

Use SINGLE letters for your math notations. Examples:
β€’ Color_j -> C_j
β€’ Net -> F(\cdot)

All the other descriptions should be within
\mathrm
*Short titles*

Add titles (e.g., using \paragraph) to your figure/table captions and the main texts. They make your paper more structured and organized and help your readers navigate the paper with ease.
*Clean table*

Follow simple design principles for making clean a table:
β€’ no \line, use \toprule, \midrule, \bottomrule
β€’ no vertical lines
β€’ left align text
β€’ center align numbers
β€’ group and remove repetition with multirow/multicol
*Avoid empty spaces*

Fill the paper into full page limit. It gives your readers a sense of a POLISHED and not RUSHED paper.

More on Deep Paper Gestalt: arxiv.org/abs/1812.08775
Hope this helps! If your paper looks like a good paper, reads like a good paper, then it's probably a good paper.

Happy writing!

Any additional tips on making your papers look awesome?
Opps... I didn't mean "clean a table"...
Example "What did you do?" figure

The teaser should specify the input/output and applications that your work could enable.

Source: pose-with-style.github.io
Example "Why did you do it?" figure

Here they use a TOY example to highlight the key issue from PRIOR WORK (wavelet transforms), i.e., the coefficients of original signal and the shifted version change dramatically.

Source: Shiftable Multiscale Transforms '92
Example "How did you do it?" figure.

This figure provides an overview of how the method work by connecting inputs and outputs and all the intermediate steps.

Source: robust-cvd.github.io

β€’ β€’ β€’

Missing some Tweet in this thread? You can try to force a refresh
γ€€

Keep Current with Jia-Bin Huang

Jia-Bin Huang Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @jbhuang0604

Oct 26, 2023
How to get unstuck?

"It doesn't work." πŸ€·β€β™€οΈπŸ€·β€β™‚οΈ

In most research projects, ~99% of the time your experiments DO NOT work. What should we do to get ourselves unstuck? πŸ€”

Sharing some tips I found useful. 🧡
*Help your mentors help you*

Tell them:
1⃣ WHY did you do the experiments?
2⃣ HOW did you do the experiments?
3⃣ WHAT did you see from the results?

Your mentors have more experience than you and are willing to help.

But you have to help them help you.
*Make your experiments easy to understand*

If you only verbally describe why/how/what you did, it's difficult for others to fully understand the process.

πŸ‘‰ Visualize your workflow
πŸ‘‰ Document the details

Your mentors are more likely to catch some issues here and there.
Read 11 tweets
Jun 5, 2023
How to draw an overview figure?

Creating a clear and informative overview figure is crucial for visualizing HOW your method works.

But how? πŸ€” Let's deep dive with 🐒 Image
*Choose the right level of abstraction*

Simplifying complex procedures helps improve clarity.

Ask yourself what the key message you want to convey. Don't overwhelm your readers with unnecessary details. Image
*Think in terms of computational graph*

Most methods process some INPUT with some COMPUTATION to produce some OUTPUT.

Visualize the flow with a "computational graph".

β€’ Nodes: Computation
β€’ Arrows: Dependency Image
Read 14 tweets
May 30, 2023
How to write a rebuttal for a conference?

Writing an effective rebuttal helps answer questions, address reviewers' concerns, clarify misunderstandings, and help the AC make an informed decision.

But it takes work to write a good one. 😟

Sharing some tips I found useful. 🧡
*Start positive*

Start with summarizing all the strengths noted by the reviewers and adding quotes to provide evidence.

Remind the reviewers and AC of
"Why should this paper be accepted?"
*Neutralize negative comments*

AC and other reviewers may only see all the NEGATIVE comments you responded to.

Neutralize them.

limited novelty πŸ‘‰ technical novelty
insufficient experiments πŸ‘‰ experimental validation
missing baseline πŸ‘‰ baseline comparison Image
Read 9 tweets
May 3, 2023
How to meet with your advisors/mentors?

If you are a grad student, having effective regular meetings with advisors or mentors is absolutely crucial for your success!

Here are some tips on how to make the most of it! 🧡 Image
*Present results*

❌ Collect and present the results you got in the last week? Terrible idea! 😱

Your advisor sees your results for the first time in the meeting? It means they don't have time to understand and think about them.

βœ… Share and discuss results async FREQUENTLY.
*Make an agenda*

❌ Dive into technical details too quickly.

βœ… Make an agenda. Manage the meeting to ensure you cover all the topics you want to discuss.
Read 8 tweets
Apr 15, 2023
How to do experiments?

Junior students often feel stressed before the weekly meeting with their advisors because their experiments do not go well. 😩😰😱

Some tips on why, what, and how to do experiments. 🧡 Image
*Why? πŸ€”*

❌ Do an experiment to get improved performance.
βœ… Do an experiment to test a hypothesis.

Many students trying to show improved results with experiments are missing the point.

Your goal of experiments should be to validate/test your research questions.
*What? πŸ€”*

What experiments should we do?

This involves three main steps:
1⃣ identify key research questions
2⃣break them down into baby steps
3⃣design experiments that best answer those questions
Read 11 tweets
Apr 5, 2023
How to multitask?

Feeling overwhelmed with multiple tasks on your to-do list? 😩

How do we effectively manage and juggle between tasks? 🀹 πŸ€Ήβ€β™‚οΈ πŸ€Ήβ€β™€οΈ Some tips below.πŸ‘‡ Image
*Calm down*

Don't panic! With good prioritization and strategy, I am sure you will nail this!
*Prioritization*

Multitasking is just an illusion of rapid context switches.

To do well, identify each task's importance and urgency and develop a strategy.
Read 10 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(