Chapter 1: Rethinking Scientific Paper Reading

As a Ph.D. student, I often find myself skimming through numerous research papers, constantly on the lookout for fresh ideas. However, I've discovered that many of these seemingly innovative works often prove to be of little value to my own research. This isn't always due to inaccuracies; rather, it's usually because the context of these papers differs significantly from that of my field.

Reading outside one's discipline isn't inherently negative. In fact, if you can adapt an intriguing idea to your own area, it can lead to a novel contribution. However, I feel that most of the easily applicable ideas have already been explored. If a concept hasn't gained traction in your field, it often indicates that attempts have been made before without success. Consequently, many researchers opt for a safer route, focusing on established work and making only minor advancements. This tendency results in a plethora of papers that are incremental and, frankly, rather dull.

After investing time in numerous papers that ultimately didn’t contribute to my field, I’ve set a higher threshold for what I choose to read. Occasionally, a captivating title catches my eye. Yet, before I get too enthusiastic about potential applications, I now investigate the authors, their affiliations, the journal of publication, and the broader context of the work. This approach allows me to dismiss most papers, even if their titles seem appealing.

The reason for this caution is that exploring a vastly different context requires delving into numerous additional papers to fully grasp the specific work’s background. Do I have the luxury of time for that? Absolutely not. Moreover, do I find those additional readings as engaging as my primary interests? Rarely. I also feel a sense of unease when diverting my attention towards findings from another discipline instead of deepening my expertise in my own. Thus, I’ve adopted a more conservative approach to papers that appear innovative. My motto has become: “Not so fast!”

Chapter 2: A Focused Approach to Machine Learning Papers

In the past, I was often eager to explore machine learning (ML) papers, but after reviewing numerous studies, I've grown skeptical of their claims. My current strategy involves a few key criteria: If the authors do not provide their code openly (for instance, on GitHub), I tend to disregard the paper. As Linus Torvalds famously stated, “Show me the code; talk is cheap.”

I also look at the number of citations a paper has received. Since I am not an ML specialist, I consider citation counts as a proxy for the paper's credibility. This approach taps into the wisdom of the crowd; a frequently cited paper is less likely to be full of inaccuracies. Additionally, I don't rush to get excited about new publications from the past couple of years. It’s wise to let papers mature over time—often, the best ideas are the ones that endure.

When I come across a highly cited paper from a decade ago, I prioritize it over a more recent paper with minimal citations. Established works often get discussed in blog posts, which is valuable because it suggests that someone might have elucidated the ideas more effectively and possibly even provided code.

These are the guidelines I now follow when reading papers. If you have thoughts or suggestions on this approach, I’d be eager to hear them!

Chapter 3: Embracing Minimalism in Research

It's essential to return to the fundamental principles of design and research.