For those who are a graduate student attempting to publish, check this out!

The goal of this post is to share some guidelines with graduate students on publishing in journals and conferences. It relies on my past experience as a PhD student. Considered one of my AI (deep learning) papers [1] recently crossed 1000 citations per Google Scholar [2]. In keeping with Web of Science™ database [3], only ~ 0.026% papers have over 1k citations. While it’s great to see the post-publication impact the paper has had, the purpose as much as the paper getting accepted was lots of trial and error. This paper was rejected multiple times, took a few years, and underwent multiple iterations before it was accepted. I actually have tried to distill my learnings here right into a guided process. Hopefully, the rules below will show you how to in your publishing journey.

Publishing is difficult, especially in case your school requires submission to high impact journals/conferences. Most PhD programs have publishing requirements, and it might probably even be the rationale for holding up your graduation.

Goals

1. Reduce time to publication
2. Amplify impact of paper

Listed below are a few of my learnings that you simply might find useful to scale back stress that comes with publishing. To attract an analogy, I believe starting a startup company and the PhD publication process have a couple of things in common. Each have an uncertain future initially, but each could end in a novel contribution of their respective areas in the long run. Similar to a startup, your publication process needs a survey, vision, strategy, iterations, and scaling. My thoughts are organized below in the identical sequence.

Survey

Read up on current state-of-the-art in the realm you are attempting to publish. Remember you are attempting so as to add to what’s already on the market. Many publications are open-access, so finding papers on the newest research mustn’t be a problem. Plus, your university must also have the option to give you obligatory access. Google Scholar and ResearchGate are great open sources. Provided that technology and scientific research are advancing so fast, I’d also follow top researchers and corporations in your area of research in LinkedIn, Twitter, and similar sites for the newest updates. If applicable, I would also explore statistics and data repositories for exploratory evaluation. On the pace of the survey, really useful numbers vary from reading 1 to 7 papers per week in preparation for literature review and having 30 to 200 papers included/referenced in your survey.

Vision

After the survey, develop a vision of what topic you would like to publish in. A startup starts with an issue it intends to resolve. The founders develop a vision that aligns with them and the issue they intend to resolve. To assist discover a match for yourself, start with the next questions:

1. What are a number of the unresolved problems in your area of study/major? E.g., this [4] is a superb reference list for unresolved problems in various disciplines.
2. What were a number of the energetic research areas discussed within the survey papers?
3. Which course[s] excited you essentially the most during your pre-qualifying exam coursework?
4. What are the areas of experience of your advisor?

Try to choose the subject from a region that has at the least three overlaps within the Venn diagram above. Write an abstract and get feedback out of your advisor. Ensure it includes your primary goal and proposed outline of the paper. The more critical the feedback, the more adjustments that shall be required at this stage.

Strategy

Throughout the initial stages of a startup company, its primary goal is to experiment, maximize learning, and land on constructing something novel the market actually needs, i.e., finding the elusive product-market fit. In your case, that might be landing on a publication, i.e., paper-publication fit. Each require strategy.

Before you give you a technique, consider the next:

1. How much time and energy are you able to commit to publication effort? How are you going to balance other commitments?
2. How much time does your advisor should guide you? Attempt to see should you can arrange a recurring touchbase along with your advisor.
3. What’s your goal time to get obligatory publications to graduate? Ensure that is realistic.
4. How are you managing stress? Note that ~ 50% PhD candidates in North America drop out before they get their degree [5]. You’ll need some sort of life-hack to get through PhD.

Once your figure these questions out and do some retrospective, give you a publication strategy that features:

1. Paper topic/problem
2. Paper type, outline and goal length
3. Goal list of publications to undergo:
   – Rank by acceptance-difficulty/impact-factor
   – Include their turn around (review cycle) times
   – Make sure the publication meets all criteria set by your institution
4. Anticipated goal date to realize publication success
   – account for n iterations and time to revise paper n times

Iteration and Pivot

When you and your advisor are aligned on the strategy/focus, start your research and check out to get to some extent where you’ve got something to indicate for (aka an MVP or minimal viable product in startup terms). Once you’ve got an MVP paper, undergo journals/conferences starting at the upper difficulty range but have known shorter review cycles. Shorter review cycles allow you to iterate and incorporate feedback quicker and pivot strategy sooner, if needed.

Don’t get disheartened by rejection and as an alternative use the critical (or harsh) feedback to make daring changes to your paper and/or strategy. Be certain that you’re evaluating, responding to, and incorporating the feedback/gaps provided by the reviewers. Rejection with feedback is a blessing and a significant a part of the method. Consider it as training an AI model. The weights of the factitious neural networks are adjusted based on feedback from labeled/training data in the course of the training process. As shown in figure 3 below, these adjustments are initially higher and progressively tapers off whenever you catch up with to paper-publication fit (or global minima in AI/ML).

Pick journals/conferences that align along with your papers research and browse the submission guidelines rigorously. There may be nothing worse than waiting for months for feedback, only to learn the paper you submitted cannot be considered since it isn’t aligned with the journal/conference’s theme or you missed an essential submission step. If applicable, it would help to research past papers from journal/conference you’re submitting to and cite relevant work from those papers. Also, in case your co-authors (e.g., advisor) have had successful papers in certain journals, try them as well.

Scale

You’ll progressively get the hang of the method that leads you closer to publication after a couple of iterations, because the rejection reasons get less critical. It could take a very very long time to get there. Note that your first publication could most certainly be essentially the most difficult and take the longest time.

In machine learning, there’s a method called transfer learning [6], where you possibly can apply knowledge gained from solving one task to a different related task, with decreasing amount of learning effort. Similar to that, all of the validated learning out of your first publication will are available very handy in speeding up your 2nd, third and subsequent publications.

Faster feedback cycles and iterations are the important thing. Use that to your advantage to get to the obligatory variety of publications. Persist with what works for you, and proceed to learn and tweak the method for more impactful publications. Good luck!