The Paper Writing Guide

Without publication, science is dead!
Gerard Piel [5]

The Role of Publications

• While many papers may be badly written, it is key for communicating our ideas to write well. [1]
• Good writing is a skill you can learn! [1]
• It’s a skill that is worth learning: You will get more brownie points (more papers accepted etc)! Your ideas will have more impact! You will have better ideas (most important)! [1]

The idea-to-paper model: Classically, people have thought of science as a process where the paper writing just becomes the final stage of the process: First, we have an idea. Then we do research for a couple of years. Finally, we sit down and concisely write up what we have been doing all these years. Years later, the research appears in a journal and will probably be read by the people who notice the paper a few years later. This model is still common in many sciences, even in Statistics (see here).

(Figure from [1])

The evolutionary model: Today, most successful groups follow the evolutionary model where an idea is directly put into a paper and serves as a key ingredient of our research idea development process. We even try to write the paper as if we had done the research before trying to get the results with place-holders for the results in order to get feedback by tossing ideas around among us at IAS and with international colleagues.

(Figure from [1])

The evolutionary model has the crucial advantages that:

• It forces us to be clear and focused. [1]
• Crystallizes what we don’t understand. [1]
• Opens the way to dialogue with others: reality check, critique, and collaboration. [1]

In computer science, we even write papers in an evolutionary sequence. Most journals in CS will be ok with authors having published a first preliminary version of the idea in one or two conferences and subsequently publish the finalized paper in a journal. IEEE Transactions on Robotics explicitly encourages this policy and may reject (weaker) papers if they haven't been tested as conference papers first. More on that later. So, let me summarize:

1. Writing isn't what you do after you have an idea. It's how you develop an inkling into an insight.
2. Turning thoughts into words sharpens reasoning. What's fuzzy in your head is clear on the page.
3. "I'm not a writer" shouldn't stop you from writing. Writing is a tool for thinking.

Generally, it is good to be ambitious. You clearly want to make a lasting impact on your field with each of your papers and with their combination into the thesis, both of which should be lasting resources to your peers [4]. According to Lemire, there are three strategies to write an ambitious paper:

1. Pick a new problem. Define the problem. Be the first to propose a solution. The problem should be simple and concrete. This way often the best to get highly cited and become famous.
2. Try to explain something significant nobody has managed to explain.
3. Improve by a wide margin what others have done. Can you reduce the error rate by half? Can you double the speed? Don’t waste our time with incremental gains (e.g., 10% faster) nor on large comparative papers (e.g., comparing 100 off-the-shelf methods).

If you scream out now: I don't have an amazing idea!, you have fallen for a common misconception among Ph.D. students early in their degree. You do not need to have a groundbreaking amazing idea with results fully fleshed out before starting to write. You may probably feel that everyone else does have such an idea. Reality is: ideas grow! Start with any idea, no matter how weedy and insignificant it may seem to you at the start of the paper [1], just make sure it is simple and concrete but not yet in the literature (if you aim at Strategy 1), not yet understood (if you aim at Strategy 2) or something known but simple and concrete where you can fool around at improving the existing algorithm (if you aim at Strategy 3). Writing the paper is how you develop the idea in the first place. It usually turns out to be more interesting and challenging that it seemed at first [1] and it will grow in your hands into solid research and often an unexpectedly good paper.

Not persuaded on the publication model? Well, don't kid yourself and look at the "side effects". You may have gotten into this game out of your love for robots and machine learning, your curiosity about intelligence, and your drive to know the truth, but you won't be able to get a job and stay in it unless you publish. You need to publish substantial articles in internationally recognized, refereed journals. Without them, you can forget a career in science. This description sounds brutal, but there are good reasons for it, and it can be a joyful challenge and fulfillment. Science is shared knowledge. Until the results are effectively communicated, they in effect do not exist. Publishing is part of the job, and until it is done, the work is not complete. You must master the skill of writing clear, concise, well-organized scientific papers [6].

Writing a paper is about communicating an idea. To do so, and to do it clearly, (1) you must have something to say, (2) have someone to say it to, (3) organize what you want to say, (4) arrange it in the order you had it said in, (5) write it, rewrite it, and re-rewrite it several times, (6) be willing to think hard about it and (7) work hard on mechanical details such as diction, notation, and punctuation. That's all there is to it [7].

The following lessons were extremely helpful for me and they may be for you as well? Use them as guideline for writing your next publication. The guide is parted into several highly different skills, i.e., (1) the process of creating a document that successfully carries a scientific point home (i.e., a paper, ideally a journal article or at least an ERA A+ conference paper), (2) the key technical skills for good writing, good typesetting and good English, and (3) checklists on possible rights and wrongs.

Developing Scientific Ideas into Papers while Doing Research

The purpose of your paper is to convey your idea from your head into understandable results in your reader’s head!!! Everything serves this single goal. You do not want to flood the reader with all details you ever encountered on the topic but convey an idea that can be understood and potentially employed by the reader [1]. Before you start, you may want to think about these points (all from [4,blog]:

• What is your central message? E.g., what point are you making? Most papers should make a single point.
• Why is this message important? I.e., why should the reader take his precious time to read your paper?
• How are you going to make your point? I.e., What experiments can you run? What theorems can you prove?
• Has this point been made before? I.e., How is your contribution different from what has been said a thousand times before?

As pointed out by Daniel Lemire, the evolutionary process leads to very productive way of writing Ph.D. theses consisting of the following steps [4,slides]:

• Come up with a [general] topic.
• Read everything about it.
• Write about what you learn.
• Ask new questions. Write them up.
• Seek answers in the literature. Ask your peers.
• Eventually, you will answer new questions: keep writing it up.
• Have different projects, at various stages: emergent, half done, almost done, submitted, in revision, in press.
• Start writing the papers before the research is completed. Take your time. Revise your writing continuously.

At the end of your Ph.D., you will realize that this way is [in today's world] probably the optimal research process. In fact, you will notice that you cannot get "Writers Block" if you stick with this way of keeping research and writing in sync! Nevertheless, we will need several key steps outlined in the following sections.

Step 1: Towards the Internal Draft (Emergent Paper)

At the center of the paper writing process is how an idea can be developed most efficiently into a desired central message that is refined as long as we keep working on this one topic. This central message is next developed into a title, an outline, and a first draft. [3]

Central Message: Even before starting to write your paper, you need to Develop your Central Message which the paper (or Bachelor's or Master's thesis or book) is intended to carry home to the greater community. [2]

Effective Titles: Your title will be read by many more people than the rest of your manuscript. Indexing services will use the title to categorize your paper. Authors who cite your paper will include the title in their list of references, which, in turn, will be read by thousands of readers. A good title should lure the casual browser to read further. No one will bother to even read your abstract if your title is boring or lacks relevance. Hence it is crucial to Develop an Effective Title [2].

Effective Outlines: The best strategy to start writing is with an outline. This first version will not be an outline that you are used to, with Roman numerals for each section and neat parallel listing of topic sentences and supporting points. This outline will be similar to a template for your paper. Initially, the outline will form a structure for your paper; it will help generate ideas and structure the process of obtaining the needed results. Always include all your available visuals (figures, tables, formulas, equations, and algorithms), and list your findings so far [3].

You may notice that the content of your paper will have become fairly clear at this stage to you, your co-authors and your advisor. The next stage deals with making it equally clear to a wider audience than buddies and friends. At this stage, your research papers needs to be easy to skim by using meaningful section headers, bullet points and simple figures. Have you achieved this goal?

Step 2: What to do with the Paper? (Towards Half Done)

The completion of the outline is the key moment when we plan what to do with the paper [3], i.e.,

• we need to decide for which venue to aim while completing the paper [3], and,
• for Ph.D. students, we need to see how it related to their Ph.D.

These considerations result are in sync with your goal: Your long-term goal is always to write a journal paper which also serves as thesis chapter! How you write depends on the journal/type of reader you are addressing [8]. The completion of the outline is the best moment to decide to which venue you will submit the paper [3]. Many people come up with three choices and discuss them with their mentors and colleagues [3]. Having a list of conference and journal priorities can help you quickly resubmit your paper if your paper is rejected [3]. In computer science, we usually write papers in an evolutionary sequence. Most journals in CS will be ok with authors having published a first preliminary version of the idea in one or two conferences and subsequently publish the finalized paper in a journal. However, do note: conference papers are not publications - not even top venues such as NIPS, AAAI, R:SS or ICML!

Choosing to which venue to send your manuscript requires careful thought. If you are entering a new field or are new to the publishing game, please always make sure to consider the following aspects (based on [2]):

1. Does the venue currently publish papers on subjects such as yours? If you were looking for papers like the one which you, in which venues would you look?
2. How is its standing in the community? Which venues have the best reputation for publishing in your field? Ask colleagues which venues they respect. Look at recent articles and judge their importance. Is the Editorial Board or Program Committee composed of leaders in their fields?
3. Is the venue competitive? What is the conference rejection rate? How high is its impact factor in comparison to the alternatives?
4. Which venues are most likely to be cited by others in your field?
5. Is the journal published by a society? Society journals are usually the most prestigious and have the largest circulation (Note: this does not apply to conferences). Be wary of new journals (in print or on the internet), especially those not sponsored by a society or by a spam publisher (e.g., Hindawi, In-Tech, MDPI, ...).
6. Is the venue open-access?

Once you have chosen a venue, start writing for that venue, obtain its LaTeX format and its author guidelines. The more certain you are about which venue you are targeting, the easier it is to write the final version of the manuscript.

A conference paper is just an advertisement of an upcoming journal publication. How effective this advertisement is depends on whether the paper fits the venue and how competitive the venue is (more competitive venues tend to be better for publicity, see here for journal quality and here for conference rankings). If a highly competitive venue keeps rejecting your paper, it probably means that it is not the right venue for it but not (necessarily) that you had a bad idea. However, if you have a failed paper that is simply not publishable (e.g., boring incremental results as well as negative results are sadly often not considered worth accepting to reasonably good venues in computer science), then this paper should not end up in the trash but in http://arxiv.org (which is the "citable" version of a paper trash bin where sometimes ideas are rescued from obscurity).

Generally, it makes a lot of sense to aim high early in your academic stage: if you have a great idea and get rejected, you still have loads of time to improve it, try again, and, if everything else fails, finally send it to a worse venue. For conferences, aiming for a better venue has the benefit that you learn quickly what the knee jerk reaction of the cool kids is who dominate the field. Having been area chair and associate editor for most major venues in our field, I can tell you that their reviews are frequently quite short-sighted, e.g., saying how your paper relates to what is currently hot. On the other hand, the lower quality conferences often have a longer memory of past work and hence will give you other advice. However, if your paper is so good that you can dare to submit it to R:SS, then IROS, Humanoids or ICRA should take it for sure. Early in your Ph.D., early in your post-doc, early on your tenure-track clock or past tenure, it always makes sense to aim high!

There is little difference in the turnaround of conferences and, hence, only the acceptance rate will alter your publication cycle. However, the choice of the conference will only make a difference to your speed of getting a Ph.D. in two additional ways:

1. Papers in very good conferences will sometimes received preferred treatments by editors (e.g., Jan's 2006 R:SS paper had an easy time getting into IJRR) but not always (e.g., Jens' NIPS paper received a lot of unrelated criticism by Machine Learning).
2. Some conferences has special issues (e.g., R:SS in IJRR and AURO, ICANN in Neural Networks, ESANN in Neurocomputing, …) or special journal tracks (e.g., ECML in Machine Learning…) both of which speed up the publication of your journal paper.

Generically, special issues of journals are often offering a fast access to some of the best journals in computer science. As a Ph.D. student, these are a perfect choice for getting your papers out fast. However, as a post-doc or faculty, be aware that these can be heavily discounted by your tenure committee as they are considered much easier to get (which often is true after all).

Choosing to which journal to send your manuscript requires careful thought. In the moment, where you feel that you have done all conference publications which you wanted to do on your topic (usually, just 1 or 2 related conference or workshop papers). Now, choose a journal which you intend to target. Obtain the LaTeX format and the author guidelines. The more certain you are about which journal you are targeting, the easier it is to write the final version of the manuscript. Then:

1. Copy and paste your contribution together in the format of that one journal and make sure that this journal paper is a milestone in your field, hopefully not to be topped any time soon.
2. Refine it, refine it, refine it! It needs to feel sufficiently polished that the editor is willing to make an effort, e.g., compile the reviewer's comments into homework instead of choosing the easier option of a rejection. That will save you a year!
3. Don't procrastinate too long! Having the conference paper in the journal review loop is crucial and if (2) applies, the remain bugs will be solved in the revision.

For journals, we can clearly say that better venues frequently have a faster turnaround. A reason for this is that they can establish high standard by simply killing all old papers, e.g., the IEEE Transactions on Robotics will only allow for one round of revisions and the Journal of Machine Learning Research (JMLR) decides right away between a rejection, an acceptance with minor changes and an accept as is. However, these journals will not help you to get your work out but only expedite already excellent work. If your paper is likely to need work, you may rather want to go to a good middle class journal and have the editor push you through many cycles of revise-and-resubmit.

Generally, there is a rough rule: For interesting original research aim only for journals with an impact factor of 1 or higher! These journals' articles are on average cited at least once a year within the two years after they got published. Journals without impact factor cannot be considered scientific publications!

Step 3: Fleshing out the Details (Between Half Done and Almost Done)

Also, keep in mind some role models -- people you know or famous papers. The general aim is to be attractive to non-experts as much as can be expected, while interesting and not offensive to experts. [8]

Developing an Effective First Draft of your Manuscript: After you have prepared your ideas and written an outline, you are ready to start writing your first draft. Note the word "first". Thinking that once we have written a draft, the paper is done is a common mistake. Instead, the first draft is only one part of the whole writing process that leads to a finished, presentable, and hopefully excellent paper. The idea of the first draft is to get the ideas out, to flesh out your answer, and to give you some content to shape and change into a finished paper. You may need to write a number of drafts in order to get to that final paper. Or, possibly, if you have really thought through your ideas and planned well, you may only need to write one draft that you then keep editing. It depends on what happens in the first draft - and it is best not to expect too much or too little from it.

Writing an Effective Abstract will improve the chances of your manuscript being accepted, encourage people to read it, and increase its impact. An abstract is a condensed version of the manuscript, which highlights the major points covered, concisely describes its content and scope, and highlights its core points in abbreviated form. It is usually the first section read and sets the tone of the paper for the reviewer. It is not a summary of the paper but a stimulating first invitation to actually start reading it.

Read Attach:hows_to_write_your_first_research_paper_2011.pdf

To be continued next week…