Contents

Reading Machine Learning Research Papers

Context for this article

Literature review is the part of the research process that allows you to understand the context of your research, and design your experiments, and form arguments based on the results. As such, literature review is unavoidable.

As important as literature review is, I find that this process takes up a significant amount of time. On top of that, when you are doing preliminary literature review, not all research papers are relevant or important. As a result, being able to filter out non-relevant research papers quickly saves a lot of time and being able to grasp the key ideas from papers that are relevant is even better.

With this in mind, I took some time to identify the series of steps that work the best for me in terms of the amount of time I invest into performing literature review. The process that I adopted for this spanned a semester: for every research paper that I read, I reflected upon how much time it took for me to (i) know if the paper that I had read was relevant to my current research focus, (ii) understand the general idea that a relevant paper was trying to convey. I read different sections of the papers in different order and kept track of the efficacy and the efficiency of the reading process.

Steps to follow

Here is the order that I have found to be the best for quickly identifying relevancy of papers and understanding the ideas presented in them.

The first three steps are best for identifying if the paper that you are reading is relevant to the research you are trying to do. The latter steps are only to be taken if you have established that the paper is in-fact relevant.

  1. Read the Abstract. Abstracts are typically short and provide a surfacial information about the paper.

  2. Read the Conclusion. The conclusion section wraps the paper up, providing information about the quality of the results.

  3. Peruse the Tables and Figures. The idea here is to quickly understand the processes (or methods) and/or results that are being illustrated with the use of tables and figures. There is not need to try to get a detailed understanding of the tables and the figures. Go through all the tables and figures presented in the paper (including the appendix if any).

Pause and Reflect
By this point you should know if the paper is actually relevent to the research you are engaged in. You can continue reading the paper at hand or abandon it with a high degree of confidence in your perception of its relevance.
  1. Read the Results and Discussions Sections. In most cases, you are only interested in the results of the paper you are reading as you can use the results to form your arguments. If that’s all you need, then you can stop reading after this point. However, if the paper that you are reading is one of the central papers that your research is based upon, then you should continue reading the next sections.

  2. Read the Introduction and Related Work. These sections actually provide a full context to the results section at the time of the publication of the paper. Further, it also sets you up to be able to understand the methodology section if you have to read it.

  3. Read the Methodology Section. Usually, this is the longest section to read and also the most challanging to understand. As a result, reading this section only makes sense if you really need to understand the paper thoroughly.

  4. Read the Appendix. Most papers don’t have an Appendix attached to it. In case there are any, you can read them last. In most cases, reading appendix earlier might not make sense, unless you are going through tables and figures.

Final Notes

If you got this far into this article, the you might find the following papers quite interesting:

  • Lennox, R., Hepburn, K., Leaman, E., & van Houten, N. (2020). ‘I’m probably just gonna skim’: an assessment of undergraduate students’ primary scientific literature reading approaches. International Journal of Science Education, 1-21.

  • Hubbard, K. E., & Dunbar, S. D. (2017). Perceptions of scientific research literature and strategies for reading papers depend on academic career stage. PloS one, 12(12), e0189753.