Write The Manuscript On LaTeX/Overleaf

I recommend to use Overleaf to save time for LaTeX environment setup, and to keep track of up-to-date progress among collaborators.

Handy Packages

• zebra-goodies for inline note-taking macros (\todo, \note) to explicitly highlight writing tasks.
• changes for manually markup changes of text, such as additions, deletions, or replacements. This will be helpful for journal submission that needs the “marked changes” revision.

Best Practices

• Use sections/ folder to organize the manuscript. A CS/AI manuscript typically has the following sections:
  1. Introduction
  2. Related Work
  3. (Optional) Study Design
  4. Methods
  5. Experiments
  6. Conclusion
• The Study Design section is optional, but highly recommended when a novel dataset or research task is proposed (see my BIBM’25 paper EpicAge). It typically contains:
  1. A brief introduction to data source and cohort (samples used in the study)
  2. A formal mathmatical problem definition
• Use macro \modelname to define the proposed model’s name \newcommand{\modelname}[0]{\textsc{EpiCAge}\xspace}. We often find that the \modelname can be changed time-to-time.
• Use figures/ folder to store all figures in the manuscript. Prefer .pdf format for vector graphics to ensure quality in publication.

How to Write A Lot

• Begin with the outlines of the manuscript (e.g., a list of subsections for each section), and then fill in the details.
• Do not worry about the quality of writing at the beginning. The most important thing is to get the main ideas down on paper, and then refine the writing later.
• Begin with the Study Design, Methods and Experiments sections, which are typically more straightforward to write. Then move on to the Introduction and Conclusion, which often require more creativity and polish.
• Write each paragraph’s punch line for the Introduction to convey the main message. This will be helpful when discussing with co-authors.

Sharing Data and Code

• Use https://anonymous.4open.science/ to host the codebase for anonymous sharing during the peer review process.
• Use Figshare or Zenodo to host the data for anonymous sharing during the peer review process. After acceptance, you can update the dataset with the final version and make it public.

Draw Figures

• I use MS-PPT to draw Figure1, motivaition examples, and flowcharts.
• I use LaTeX’s native tikz and pgfplots package to draw model architecture and results figures (e.g., bar plots, line plots, heatmaps). It is a hard learning curve at beginning. But it can keep the raw experimental results and the manuscript in the same LaTeX file, which is very convenient for revision and journal submission. It also ensures the consistency of fonts and styles between figures and text.
  • Recently I also used Python’s matplotlib and seaborn to draw some figures.