by Chris J. Maddison (adapted and expanded from How To ML by Jakob Foerster)
This guide lays out a formula for writing machine learning papers. It might seem absurd to follow such a formulaic approach when writing up the conclusions of your very unique research, but this formula is both flexible enough to communicate your contributions and structured enough to make your ideas accessible to a wide audience. The format will vary slightly between papers, but there are some rather rigid general rules. This guide includes one instantiation of those rules.
As a junior researcher, it’s probably a good idea to stick to this format. Formulas like this help the community understand what you did and where it fits in the broader context. They also help readers that skim find the key bits of information. In other words, this format is like an API that allows readers to interact flexibly with your research.
Over time, as you develop as a researcher and writer, you will learn to bend these rules. But, at the outset, you should only beak a rule that you understand and are good at executing. If that entreaty doesn’t convince you, reviewers largely expect this format. If you break significantly with this approach, you run the risk of confusing reviewers and harming your paper’s chance at publication.
<aside> 💡
There are no hard and fast rules in scientific writing, but, if you’re new to it, it can help to follow this structure very closely.
</aside>
Each bullet corresponds to approximately one sentence that answers the question.
Each bullet corresponds to approximately one paragraph that answers the question.