Skip to main content

Paper Writing and Venues

danger

DO NOT use LLM for writing, even drafts. It undermines your learning progress. It's also an academic misconduct in my view.

You don't want do be listed below. It's very bad for your present and especially future.

Paper-writing rules in our group​

  1. Research Lifecycle:

    1. Read papers & textbooks, have discussions, etc. => have (vague) ideas.
    2. Write a proposal. At least 4 pages excluding reference list. Title, abstract, intro/motivation, background, related work, proposed methodology.
    3. Discuss and iterate with Agustinus at least 3 months before submission deadline (see below).
    4. If go ahead: do experiments, writing etc. If not: back to (i).
    5. Have complete draft (i.e. 8 pages, unpolished and only toy experiments totally fine) as early as possible.

      • The more you get feedback and iterate, the higher the quality of the paper, the higher the probability of acceptance.

        important

        Follow the breadth-first approach when writing.

  2. Internal Deadline: One week before the real submission deadline.

    • If you don't have a ~95% polished draft yet by then, Agustinus won't allow you to submit.
      • ~95% means 8-9 pages, figures are all there, proofs are all done, look-and-feel roughly there, etc.
      • Missing some additional (not the main ones!) experiments is fine.
      • Unpolished grammars are fine at this stage.
    • We strive to only publish quality works. You don't want to be remembered as someone who publish slops.
    • If you missed this internal deadline, this would mean you have to wait for the next venue, and you lose time. Consequences are: miss out on fellowships, delayed internships, delayed citations, etc. Bad for your future! The alternative---submitting/publishing slops---will also be even worse for your future.
    • So, manage your time well. Don't procrastinate. Write even before doing experiments.

  3. Follow Writing Advice faithfully.

Venues​

Conferences​

All of them are great! But: ICML, NeurIPS, and ICLR are more competitive.

  • ICML
    • Submission: end of January
    • Conference: July
  • UAI
    • Submission deadline: February
    • Conference: July
  • NeurIPS
    • Submission: end of May
    • Conference: December
  • ICLR
    • Submission: end of September
    • Conference: April/May
  • AISTATS
    • Submission deadline: October
    • Conference: April/May

Journals​

  • Nature Computational Science
    • Scientific application focused
  • TMLR
    • ML fundamental focused
  • JMLR
    • Avoid, review process too long