Rule 1: Under whatever scenarios, no plagiarism is allowed in our group. Not from others' work, not from your own previous work, not from any GenAIs. Check all the numbers you write in the paper and make sure they can be reproduced. You are responsible for what you've put into the paper/report.
- How to be a successful PhD student?
- So long, and thanks for the Ph.D.
- Collection of advice for prospective and current PhD students
- Awesome Tips
We use Slack for most communications. Ask in #random channel if you don't know things
Rule No.1: Respect each other’s time. Don’t waste time & being unproductive. We’ve reserved the time for each other, so please use these slots wisely. You should start your experiments at least the next day after each meeting. Analyzing results & finding maybe bugs & thinking take time. If you just start the experiment one day before the meeting, please just cancel the meeting.
-
Write weekly report: Ask senior students what this is.
-
Prepare meeting slides/notes for individual meeting. This post is what I read during my PhD. I recommend everyone read it. Especially on "You and Your Advisor" part.
- Indicate what is the project goal/current stage goal. I can probably already forget the context. So take 1-2 mins to quickly refresh us to avoid going back and forth to ask clarification questions.
- What you’ve done last week
- What results you got —> what’s the conclusion. No need to read numbers. We can read pretty quickly. Takeaway messages matter more.
- What problems you have encountered —> what is your plan next week to solve these?
-
Read papers and participate in #paper-reading channel
-
Attend and contribute to weekly group meeting. Sign up for presentations, your work or interesting paper sharing.
- Set up a private github repository for each project under lime group when starting the project. Ask PhDs to add you if you're not a LIME Github member.
- Clean up code after submission and release the code/data when the paper is accepted.
- Use brach/tag wisely for logging experiment results
- You can use wiki and issue to track your progress and experiments
- Push your code frequently! Our servers don’t have disk backup.
- Create a slack channel for each project and invite your collaborators (guest if they are not from USC). Channel name should be "proj-xxx".
- Keep logs for your projects. Pin the link to the Slack channel.
- Write down your experiments settings and keep your main results
- Write down random ideas and remember to go back and check them
- Unfortunately as a faculty, we have much more to deal with. For your submissions (paper or any applications), if you want detailed comments from me (and multi-round), I need your draft 2 weeks earlier than the official ddl. If you give me the draft one week in advance, I will help with the abstract and introduction and skim through the paper. Otherwise, I probably won’t be able to go really deep. In addition, if your draft was not ready one week before the ddl, you probably shouldn’t submit your paper.
- ps: draft means the storyline has been made; at least half of the contents were there; there can be some tables/related work to be filled in. But you have gone through the paper and do some basic check like we said in Paper Writing
-
Have you tried to answer these questions?
Add \newcommand{\jz}[1]{{\footnotesize \textcolor{blue}{[[Jieyu: #1]]}}} to the preamble of your LaTeX document. Ensure that the xcolor package is included by adding \usepackage{xcolor} in the preamble as well.
- LaTex Style (by Jordan Boyd-Graber): https://users.umiacs.umd.edu/~jbg/static/style.html (especially the table part)
- Write the Paper First (by Jason Eisner): https://www.cs.jhu.edu/~jason/advice/write-the-paper-first.html
- How to not write a paper? https://people.csail.mit.edu/alinush/communication/writing.pdf
- Shitty First Drafts: https://wrd.as.uky.edu/sites/default/files/1-Shitty First Drafts.pdf
- Everyone can write better
- Crafting Papers on Machine Learning: https://icml.cc/Conferences/2002/craft.html
-
Let’s respect each other’s time. Before sending your paper to others for review, try your best to reduce small/naive/careless/obvious errors. I will wait to continue reading your paper if there are too many small/careless errors. So make sure you have done some reading/grammar check (using GPT or Grammarly) before asking me to do it. The devil is in the details. You may have put great efforts in your paper writing, but such errors will make the reading very unpleasant and hurt your reputation (and all co-authors').
-
Do not underestimate the writing. A brilliant idea can be rejected because of bad writing. A normal idea can get high scores with great writing. Since most of us are not first language in English, it will require more efforts in the writing. So start writing early. Personally, I will start writing at least one month before the ddl.
-
Plan your time wisely. This does not only apply to the writing. It should apply to your whole research and PhD life. Do not do things last minute.
-
Do not assume readers will know all the background knowledge. I think NLP paper is more enjoyable to read is because the community encourages the “explain to me like I’m 5” style.
-
Write the paper in top down style. List the section title; with 1-2 sentences about what you’re going to write in each section/subsection, and then fill in the details.
-
Avoid unnecessary bolding/capitalization. You don’t have the bold the caption. But you need to add “.” after the caption
-
Start to reaching out to others for paper clinic early. And please help each other to improve their papers. They will help you later when you need it.
-
Make sure you use the correct model name format. e.g., GPT-4 not “GPT4”. LLaMA not “LLaMa”.
-
Use the right citations. If the work has been accepted to a conference, use the bibfile of the accepted paper instead of the arxiv one.
-
Make sure you've removed your name in your code, especially in file path. If you want to have an anonymous github link in your paper, that is OK. But make sure you remove the file path tha has your name in it. Or you can just say you will release your code and data after paper acceptance.
-
Try to learn from my comments. In my comments, I tried to explain in detail why I want to make such edits. Later when you get senior (and our group gets larger), I won’t be able to provide such detailed comments.
-
High-level:
- Logic flow: Do you go with the flow as: what is the research question + why it is important —> what have been done + why existing work is not enough —> What have you done and why it works & how does it work (performance wise)?
-
Low-level:
- Long sentences check: do you have sentences over 3 lines? (double line format) If so, rewrite them.
- Is every sentence adding new information? If no, delete it.
- Is this new ‘terminology’ has been explained previously? If no, add the definition.
- Word choice: did you use the same word in very short distance? Choose a synonym.
- Check Yue Zhao's collection: https://github.com/yzhao062/cs-paper-checklist
- Good and Bad Procrastination
- Some Tools: e.g., https://www.rescuetime.com/
Before making public talks (conference or your oral), sign up for reheasal talk in the group. Create a google doc and ask people to provide you suggestions (if you want suggestions).