The formal aspects of writing a research paper
Authors, Institutions, Acknowledgements:- Spell your institution correctly, this is important. Either "University of Tübingen" or "Max Planck Institute for Intelligent Systems, Tübingen". The word "Tübingen" needs to be there, with exactly this spelling and the umlaut.
- Use your official university or max planck email address, not a private one.
- In a paper with myself as author, always ask me which of my affiliations comes first and second, and which email address to use.
- If you are a PhD student, please always tick the "I am a student" box (no matter whether you are enrolled or not). It makes you elegible for the student paper awards.
- In a final paper, always ask me what the acknowledgements need to be.
Help readers to get the main message:
- Use meaningful section and chapter headings. Rather than the section "Our main result" write in the heading what the main result is: "Our main result: spectral clustering is consistent". You might also use bold paragraph beginnings or bold font sentences to stress something. Ideally, a reader who just scans the paper and reads titles, bold statements, and the theorem titles should know what the paper is about. As a nice example, consider Lucas Netgan paper (link soon).
- Figures A plot always has axis labels in a
readable font and a title (describing parameter settings, for
example). A plot always has a caption, and this caption needs to
contain a concise summary about what the plot shows. The plot should
be understandable without flipping back and forth between text and
plot. If the plot shows experimental results, the caption should
summarize the setup of the experiment (e.g., parameter choices), so
that one would be able to reproduce the plot. The
interpretation/discussion of the results in the plot typically end up
in the main text, not in the caption.
Plots need to be black-and-white readable (if you print it on a black and white printer, one still needs to be able to distinguish different lines). Use different symbols, not only differnt colors. When you use colors, use a color scheme that is friendly to color-blind people (not red-green, rather blue-orange, etc). All lines should be thick enough, if in doubt increase the line width in your plot. All fonts in the figures need to be large enough. They should appear roughly as large as the font in the text. This might mean that you need to set the font size in python to 60 or so. Use meaningful captions.
Lots of formalities for writing:
- Citation style: please use the natbib style, never use a number-based citation style until you really have to (it is very unfriendly to readers). The latex commands are:
\usepackage[round,comma]{natbib}
\bibliographystyle{plainnat}).
In the text: Please make sure you understand the differences between the commands \citet and citep and \citealp and use them accordingly. Example:
As has been shown in \citet{Maier} -> as has been shown in Mayer (2020)
It has been proved \citep{maier} that ... -> It has been proved (Mayer, 2020) that ...
It has been proved (\citealp{maier}, Chapter 3) that -> It has been proved (Mayer, 2020, Chapter 3) that ... Form of the reference list: bring it into a good shape, this does require work and is really important (you might cite your reviewer, so make sure he/she is not upset!).q In particular, you always need to edit bibtex entries that you downloaded from places like google scholar. Please look at some of our latest papers for examples how all of this works.- For papers at well-known conferences, use a short version, such as: authors, title, ICML, 2020. If the conference is not well-known, put the full name. You do not need to specify editors or page numbers.
- For journal papers, use the standard format.
- For books, give a chapter. Nobody will find what you mean by flipping though a 500 page book.
- Only cite arxiv papers if the paper has not yet been published (you need to check for each of your references, go to the author's homepage).
- Be consistent: all authors with full first name, or all first names abbreviated (which is easier).
- Do not include Isbn, url, etc.
- Make sure the captalization works correctly (for words like DNN, GAN, Bayes, ... )
- All propositions, theorems, lemmas, ... should have a "descriptive" heading in bold, so the reader knows what the intuitive content of the statement is. As an example, consider this paper.
- Capital letters: Whenever a theorem, lemma,
definition, figure, ... gets a number, you need to use a capital
letter. If there is no number, then use lower case letter. Examples:
As we show in Lemma 5 ...
The following theorem is really nice.
In Figure 4 we can see ...
The third equation is wrong.
In Equation (17) we state ...
- Latex references: there is a difference between \ref and \eqref:
In Equation \ref we prove -> In Equation 17 we prove ...
In Equation \eqref we prove -> In Equation (17) we prove ... .
It also helps to use unbreakable spaces: In Figure~3 we show that ... Then the word "Figure" and the "3" cannot end up on different lines. - For all the math, use one counter throughout the whole paper, not a separate counter in chapters and not separate counters for theorems, propositions, etc. Definition 1, Proposition 2, Theorem 4, Proposition 5, Conjecture 6, ... Exception: Figures have their own counter.
- Never use the abbreviations "i.e." or "e.g." in the text, always write it in words. In general, if an abbreviation is used not very often, say less than 5 times or so, simply don't introduce it at all and use the full words instead. Only use abbreviations for things that you use all the time.
- Punctuation of formulas: When you write a
formula in an equation environment, still use full stops or commas
after the equation if necessary. Each sentence needs to end with a
full stop. Example:
This leads to Equation $$X^2 = 1.$$ (full stop in the equation)
Because we have shown $$X^2=1,$$ which is really beautiful, we can conclude ... (comma in the equation). - Try to avoid footnotes. They distract the reader, and draw much more attention to something that you feel like hiding. Simply write it in the text, potentially in parantheses. (Few exceptions exist).
