Paper Writing Checklist

General

• Don’t leave writing the paper until the last minute. Aim for a complete draft a week before the deadline.
• Sections should not start or end abruptly. The last line of a section should serve as a natural connector to the next section.
• It should be easy to read the paper in a big hurry and still learn the main points.
• Make certain that the reader is in nodoubt what the idea is. Be 100% explicit: “The main idea of this paper is…”, “In this section we present the main contributions of the paper.”

Structure

• Order of paper: At the end of your paper, the order of sections should be: Acknowledgments, Appendices, then References.
• Acknowledgments should be a unnumbered section. You can use the command for this.
• Appendix section number should be a letter (Should look like Appendix A. Title).

Abstract

• Include: contributions, approach, results, advantages.

Introduction

• The introduction has several parts:
  • 1. Background and motivation (1 paragraph)
  • 2. Overview of the paper and contributions (1-2 paragraphs)
  • 3. More details and summary of the approach
  • 4. Summary of the results and conclusions
• New trend: specifically list your contributions as bullet points
• DO NOT write this (or something similar): ““This paper is organized as follows. In Section 2, we introduce our algorithm. In Section 3, we test our algorithm. In Section 4, we run some experiments. In Section 5, we cut and paste text from the Introduction.”.
• Sell your work, but do not lie. All claims should be supported with experiments, proofs, etc.
• Avoid grandiosity (“we propose a framework…”)

Related work

• Avoid phrasal templates of the form ‘Smith et al. do X. Miller et al. do Y’.
• Just describing what another paper is doing is not enough. We need to compare and contrast.

Background

This section is all about laying down some notation as well as an explanation of most of the concepts. Ideally, even if I am not an expert in the precise subject matter of your paper, I should at least get a rough idea of what you are doing later on, what your methods are relying on, and—potentially—what your theorems might be. Think of this section as the foundation upon which your method is built. There is no ideal length here, but I would aim for something that is at least shorter than your ‘Methods’ section.

Methods

• Show simple toy examples to let people get the main idea
• All jargon and technical terms that you use should be defined in the ‘Background’ section.

Experiments

• A comparison with other methods (typically, if something is mentioned in the ‘Related Work’ section, you should compare to it).
• A set of ablation studies where you analyse the performance of individual parts of your model.
• Have one full page as an absolute minimum for most machine learning papers

Figures

• “Teaser” Figure One on Page One/Two
• Any text inside a figure, such as in a legend, must have a readable font size (almost matching but not exceeding the size of the other text).

Math / Equations

General rules

• Formulas and equations need to be contained in complete sentences with proper punctuation.
• Don’t start a sentence with a formula.
• Make sure that you introduce each variable that you use. Describe each variable as precisely as possible.
• Avoid using equations/references as nouns (e.g., don’t write “(12) suggests” or “[12] claimed”)
• Instead, use explicit references like “Definition/condition (12) suggests” or “Maryam [12] claimed”
• Avoid informal implication phrases like “This implies…” or “So, …”
• Be more explicit, e.g., “Invoking the constraint (4), we see that inequality (8) implies…”
• Prefer natural language for quantifiers (use “for all” and “some” instead of ∀ and ∃)
• Use formal verbs for results (replace “we get” with “we derive/conclude/infer/…”)
• Distinguish between a function and its value (function f vs. value f(x))
• Distinguish between random variables and their instantiations (uppercase Y vs. lowercase y)

Formatting

• Use human-readable notations for equations.
• Augment math notation in text with human interpretations.
• Before displayed equations, punctuation (e.g., “:”) is not necessary unless you would use the same punctuation if the equation were not there.

Table

• Format and group table columns in a human understandable way, specially when using related metrics/datasets.

Grammar

• Avoid passive voice as much as possible. Active sentences are clearer and more direct; passive are more indirect
• Be consistent with tense. Avoid switching at all costs, also avoid using future tense if it can be avoided (Eg. “In Section 3 we will show…”)
• Avoid filler words at all costs (can, could, would, may, in order to, etc.).
• When using pronouns, make sure the reference is clear: eg “The dog was barking, so John put dogfood in his bowl.”

Misc

• Be extremely clear on contributions. Never blur the lines between what had been done before and what you did.
• Once you have written the initial text, try to delete around ⅓ of the words. That’s typically how much “fluff” there is (aka cutting the clutter).
• Introduce only symbols and acronyms that you use in the paper.
• Footnotes should be after “.” and “,”
• Avoid line breaks / paragraphs that result in lines containing a single (or few) words. It looks lazy.
• When referencing figures, sections, etc. (“Figure 1”, “Section 2.1”), please make sure the words “Figure”, “Section”, etc., are uppercase
  
• Fill the page limit and avoid poor formatting / other whitespace. It looks lazy.

Formating

• Please use correct quotation marks in Latex ``correct quotation'' (copy-paste this if unclear).
• Mathematical equations follow standard rules of orthography. Do not forget periods and commas after equations and avoid unnecessary colons. eg (✔) The relationship between the radius and area of a circle is A = r^2 . (❌) The relationship between the radius and area of a circle is: A = r^2 .

Tips

• Use \usepackage{hyperref} and \usepackage{hyperref} instead of simply \usepackage{hyperref} to make it easier for a reader to jump to the references and back.
• Acronym + citation: “proximal policy optimisation ..”.
• Use the cleveref package for intelligent cross-referencing of figures, tables etc.

Citations

• Cite the correct version of published papers. Google scholar often defaults to the arxiv version, rather than the conference paper. PaperLib is incredibly useful for this. It automatically scrapes the correct version of published papers from Google Scholar.
• Do not use bracketed citations as nouns unless mandated by an editor. For instance, ‘This method is great for graphs [1].’ is acceptable, while ‘See [1] for more details’ is not.

References

• Highly Opinionated Advice on How to Write ML Papers by Neel Nanda.
• How to ML papers - A brief guide by Jakob Foerster (Assoc Prof in ML at UniofOxford).
• Paper Writing Best Practices by ICLR
• Checklist for effectively writing papers in stat-ML by Aaditya Ramdas. Associate Professor at CMU.
• How to write a good paper by William T. Freeman at MIT and Google Research.
• Writing Research Papers by Aaron Hertzmann at University of Toronto.
• Paper Writing Workshop by Karanveer Singh, Moayad Elamin, Shreyas Piplani.
• Writing (Computer Vision & Machine Learning) Papers from the Reviewer’s Perspective by Matias Valdenegro.
• How to Write a Scientific Paper by Bastian Grossenbacher Rieck.
• Checklist of common JMLR formatting errors by Journal of Machine Learning Research.
• How to write a great research paper by Simon Peyton Jones at Microsoft Research Cambridge.
• Guide to Writing Mathematics by Hong Kong University of Science and Technology.
• A Primer of Mathematical Writing by Steven G. Krantz.
• Deep Paper Gestalt by Jia-Bin Huang