Speed Survey
This article is for AI researchers.
Inspired by the survey method of Ochiai-sensei, I would love to encourage the “Speed Survey” method as a basement of research activities.
Speed focus
In the speed survey, it’s all about the fight against the time. Therefore, you need to use a stopwatch to know how long do you take to read one article. (conference paper, journal paper)
Similarly to Rubik’s cube, or Full Marathon, You cannot do on day 1. You need exercises to do it, but on average, one to two months are all enough to reach a 5-minute level.
Objective: Grasp only the overview
Of course, it’s impossible to understand every single detail within the five minutes. It is only to grasp the main ideas of many papers instead of a deep comprehension of a single paper.
After reading 20–30 papers, we can reach out to the point where you can find some common underlying insights. For example, many methodologies have the same approaches, or the problem statement tend to change over a couple of years, and so on.
Six Questions
In order to do it, just try to answer the six questions and ignore any other details.
You can come back to the in-depth survey later, but you should first focus on the speed survey.
URL Here:
https://www.dropbox.com/s/gsbt0zovvnmrjco/Survey_Template.pptx?dl=0
Key Figure is useful in order for you to memorize the concept visually. For this purpose, I would suggest curating the information in a powerpoint manner.
The goal is to make this slide within 10 minutes.
Methodology
My suggestion is that, instead of reading through all the contents, you should come back and forth to the question list and write summaries.
First Read (30sec)
Comprehend the abstract carefully.
First answer (30sec)
Several questions can be answered already. Understand and memorize what is unknown yet. (not write down at this moment. just check if you can answer)
Second Read (1.5min)
Read through the introduction and conclusion. Also, see several key figures and try to understand it by taking a look at the system/proposal part. Don’t take time. You may have a bunch of unknown words but ignore it for now. If you feel some words are the core concept of the paper, you may Google it. If Google doesn’t let you know the exact idea, maybe this is not a core concept (or the paper is not a quality one.)
The second answer (30sec)
Most of the questions can be answered already.
(not write down at this moment. just check if you can answer)
Third Read (2min)
Read the proposal/systems and experiments and check whether the answers are right or not. If there's a misunderstanding, make a fix. Finally, come back to Abstract again and conclusion. If there’re questions incomplete, fill in.
After the third read, then you can take another 3–5min to fill in the form to complete the survey.
Pacing
Spend 1 hour per day. No more than 1 hour. Record the time for one by one.
Before the survey, you should prepare what papers to read within the day.
- Week 1: Try to complete 1 survey forms. Get feedback from senior researchers.
- Week 2: Try to complete 2 survey forms. A later part of the week would be good with 3 forms. Get feedback from senior researchers.
- Week 3: Try to complete 4 survey forms.
- Week 4: Try to complete 6 survey forms.
