1 of 15

DEMYSTIFYING RESEARCH PAPERS

BY – ATHARVA ANAND JOSHI

2 of 15

THE WHY

​

  • Crucial to further one’s understanding of the field.
  • Keep up with the pace of research in this field.
  • Stand on the shoulder of giants.

3 of 15

WHY IS IT DIFFICULT?

  • Authors tend to assume significant background knowledge from readers
  • Academic syntax is dense and thus difficult for readers to parse
  • Mathematical expressions are typically condensed and equations reordered for concision, often skipping steps in derivations.
  • Substantial knowledge gaps are filled if a reader has read cited papers.
  • Not all conclusions drawn are correct. Some papers stretch their claims.

4 of 15

HOW A PAPER IS GENERALLY ORGANISED

​

  • Title
  • Abstract
  • Introduction
  • Methods

​

​

  • Results
  • Discussions
  • References
  • Appendix

​

​

5 of 15

THREE-PASS APPROACH

​

  • First Pass: Bird’s Eye View of the Paper
  • Second pass: Understand Contents but not details
  • Third Pass: Understand Paper in depth

6 of 15

FIRST PASS

The first pass would take you around ten minutes:

  • Title, Abstract and Introduction
  • Sections and Sub-sections
  • Mathematical Foundations
  • Conclusions
  • Take a look at References
  • Identify the “Big Question”

​

​

7 of 15

FIRST PASS

After this pass you can answer the following questions:

  • What type of paper is this? (Measurement / Analysis / Research Prototype)
  • What is the domain of work?
  • Do the assumptions seem genuine?
  • What are the key contributions of the paper
  • Is the paper well-written?

​

​

8 of 15

SECOND PASS

The second pass would take you around an hour.

  • Read in greater detail, but avoid proofs.
  • Make a note of questions you would like to ask the author.
  • Look at the figures and diagrams, whether they are properly labelled.
  • Mark unread references for future reading.

​

9 of 15

THIRD PASS

The third pass would take you around a day.

  • Virtually re-implement the paper: Make the same assumptions as the author and re-create the results.
  • Challenge every assumptions made by the author.
  • Besides the contributions, focus on the limitations of the paper. You might get new ideas for future work.
  • Form an educated opinion on the paper and discuss it with your peers.

​

10 of 15

https://blizzard.cs.uwaterloo.ca/keshav/home/Papers/data/07/paper-reading.pdf

​

11 of 15

IS THIS THE ONLY WAY?

12 of 15

NEWSLETTERS AND JOURNALS

​

  • Import AI by John Clark
  • Machine Learnings by Sam Debrule
  • Nathan.ai by Nathan Benaich
  • Distill.pub

​

13 of 15

FINDING PAPERS

  • arXiv
  • Google Scholar
  • Social Media
  • Friends, Professors, Colleagues
  • Libraries (For Paid Articles)

​

14 of 15

AI PEOPLE ON TWITTER

​

  • Andrej Karpathy
  • Yann LeCun
  • Michael Nielsen
  • Jeff Dean

​

​

  • Ian Goodfellow
  • Chris Olah
  • Jack Clark
  • Francois Chollet

​

15 of 15

REFERENCES

​

​