Continuous learning is a critical skill. The internet houses a panoply of learning material for almost any topic under the sun. YouTube is my go-to for learning anything and everything - for it is easy to digest and entertaining. I’d regard YouTube to be on the lower side of information density. Contrastingly, reading acadmic papers is something I’ve begun recently. YouTube and academic papers are on the opposite extremes of information density - which is why I like reading papers.
Through this post, I share my inclinations to read papers, some frameworks to understand them and how communities like Papers We Love are essential to catalyze essential discourse.
My Experience
I had my first stint with reading research papers for college. We had a final year project which involved preparing literature surveys and finally producing a paper of our own. This introduced me to looking up and understanding papers. Finally, writing one was a good experience on producing structured technical literature.
One of my friends introduced me to 6.824, MIT Distributed Systems course by MIT. This course is taught through dissecting research papers - from the early Map Reduce to Raft and finally large scale distributed systems like Blockchains. One of my biggest takeaways from this course is the exposure to the fascinating world of understanding computer science from a perspecive of academic research - the key aspect being able to grasp the evolutionary nature of scientific development.
Essentially, science isn’t particularly definite. It is an evolutionary body of knowledge and experimentation lies at the heart of it. Each experiment produces a new block of knowledge, thus expanding our sphere of understanding. Similarly, future experiments will build on the knolwedge from past experiments. If we see each paper as an experiment or observation, we will be able to see the chronological development of this body of knowledge. In subtle sense, the paper is a part of a story being told of science’s evolution - a paper explains why such an experiment was carried out, how it was carried out, what the observations were, etc.
Critical Thinking
Consequently, while we go over a paper, we may sometimes pause to ponder over the process of experimentation carried out in it, the style of implementation, etc. This opens up an arena for discussions regarding the multitude of factors considered while developing that paper. As a result of this, we subconsciously imbibe the fact that everything brings with it a tradeoff - some subtle and some apparent. This kind of mindset, I believe, is crucial for deepening your understanding of a subject and might possibly open door for innovation.
Information Dense in a Hyper-marketed World
Critical thinking demands comprehensive knowledge in the subject matter and unfortunately, the currrent world is bombarded with hyper marketed content - to milk out every last cent of profit and to exploit market segments. This is especially apparent in industries like blockchain and finance. In situations like this, a research paper, as it is verified by academically certified panelists, is a boon for gathering information on a topic. And it is upto you to zoom in or out on the details - and this provides you the flexibility to easily understand the topic, verify the information and decide whether to continue in this direction.
Summary and Implementation
After reading a paper, I try to summarize the learnings. Sometiems I take a step further and try to implement aspects of it or in its entirety. This is time demanding but makes the understanding concrete.
Community
Previously I had mentioned critical thinking which involves understanding the tradeoffs, rationalizing the choices taken by the author and thiking of ways to improve the paper. While all of this involves only the effort of the self, it is infinitely more pleasureable to discuss this paper with a like minded group of individuals. Pinging off ideas off of each other, trying to understand the subject matter through a different perspective from another perpective, etc is invaluable. This is where communities like PapersWeLove come into the picture. I urge your to explore these.
Frameworks for Reading Papers
Academic papers are structured and almost every paper has a similar pattern. Throughout my journey, I’ve discovered a few frameworks for reading papers. These frameworks guide you to easily understanding a paper - whether you are looking for surface level information to decide if you’d want to continue down this path or if you want to deep dive into the domain. The Three Step Pass is one of most popular ones I’ve found. Below is an outline. The full method is available as a paper here (kinda meta right?)
The Three-Pass Approach
Follow the three passes, where each pass will conclude with you gaining deeper and deeper knowledge of the subject.
First Pass
This is a quick scan which will help you decide whether to do any more passes.
- Read the section headings to get a bird’s eye view of the paper.
- You can glance over the references, mentally ticking off the ones you’ve already read.
At the end of the first pass, you should be able to anser the 5 C’s:
- Category of paper (measurement, analysis, description, review, etc)
- Context
- Correctness
- Main contributions
- Clarity: Ss the paper well written
Second Pass
This step is to understand the key points, so ignore details such as proofs.
- Look carefully at illustrations and correctness - this helps to separate shoddy work.
- Remember to mark unread references and unknown words - revisit them back to learn about the background of the paper.
After this pass, you should be able to summarize the main premise of the paper, with supporting evidence, to someone else.
Third Pass
This is an attempt to virutally re-implement the paper. This stage will help you identify not only the paper’s innovations, but also its hidden findings and assumptions.
- Identify and challenge every assumtption. And think about how you yourself would present a particular idea.
- Jot down ideas for future work.
After this pass, you should be able to reconstruct the entire paper from memory - pinpoint implicit assumptions, missing citations to relevant work, potential issues with experimental or analytical techniques.
Conclusion
In conclusion, while reading papers is a pleasure, it is time consuming. A lot of information is available in more concise forms - some in enjoyable video formats. Reading papers isn’t a necessary skill but an experience on its own.
If you want to go down this path, here is a convincing article on reading academic papers by the founding members of the PapersWeLove team: You should be reading academic computer science papers