Tips for students when writing AI research papers

When I mentor university and high school students in creating exciting research projects in AI and machine learning, an important stage of our work is the creation of papers that help their ideas to be shared, validated, and reproduced.

From my experience as a reader and writer of these types of papers, I wanted to compile some of my suggestions to make an excellent paper. This list is non-exhaustive and non-rigid (in fact, it is only partial, and I will be adding to it as I continue finding and discussing cases with my students!), but I would anyone writing similar papers to consider these ideas when preparing their manuscripts.

(1) Do your best to share any dataset you create for your research. This encourages others to continue your research, and invites derivative works and positive impact on the field.

(2) When you introduce your dataset, include figures showing samples of each class of data (with respective annotations). Sometimes this in the form of a class label or a bounding box on an image, other times it may be a spectral representation of audio data, other times it may be a table showing the frequency of each class within the dataset when the data itself is not possible to visualize. Be creative, and seek to make your data accessible for your reader to understand.

(3) Share your code as often as possible on a GitHub repository. This not only makes your ideas easier to share, but also helps remind you to code in a way that promotes readability, reproducibility, and collaboration.

(4) Try to remove subjective implementation details from your writing. Annotation is a tedious task, but your reader does not need to know how much time it took you to annotate your dataset, or which folder you saved your annotations to on your hard drive. If the statement doesn’t help a reader to reproduce your paper regardless of their specific implementation method, or does not directly support your research question or hypothesis, it is best to leave it out (or move it to an appendix section concerning implementation details — or even better, a GitHub README with your code!).

(5) If you are making a claim that isn’t well-known to your reader base and isn’t something you show in your work, be sure to cite the source for your claim. Having a research mentor look over your paper is a great way to receive feedback on which claims require citation.

(6) When dealing with machine learning models, readers (usually) want to know your test performance, not your training performance. If you think your project falls outside of this evaluation rule, consult with your research mentor.

(7) Authorship and Acknowledgements are two very different things, and including someone or some organization in either section can carry significant professional, social, and emotional weight. When in doubt, discuss who should be included and how they should be included in your paper with your research mentor.

Happy writing!

Did I miss any tips you would include? Let me know in the comments!

Looking for a mentor for your own research project? Reach out to me at researchwithross@gmail.com.

Related posts:

Penerangan adalah proses atau hasil dari pencahayaan suatu ruang atau objek oleh sumber cahaya. Sumber cahaya sendiri adalah objek atau perangkat yang menghasilkan cahaya, seperti lampu, lilin…

A new program in Suffolk County has been launched that will cause almost 48,000 vehicles to be booted or towed. Any vehicle whose owner has not paid three parking tickets, three traffic violations or…

A while ago I bought some new clothes online, but this is what happened.. “Clothes shopping” is published by Julienne Bay.