How to efficiently read the paper

How to efficiently read the paper

Andrew Ng

Paper needs to understand how much time?

For people who are new to machine learning, understanding a relatively simple thesis may take an hour, it is not uncommon. However, sometimes you may occasionally find that you need three hours or even longer to really understand the papers.

The most important thing is to keep on learning, getting better means more stable learning, rather than focusing on a period of time to read a lot of papers. Its rote in a short time, it is better to read the two papers a week starting next year.

General approach

Step 1. prepare a paper list

Try to create a list of research papers, including any text you have (such as micro-channel public number, etc.) or learning resources.

Step 2. go over the list

Basically, you should be in a parallel way to read research papers, meaning a deal papers. Specifically, try to quickly browse and understand every article, rather than read them all, maybe you read each article of 10-20%, maybe this is enough to make your article at hand there is a high level of understanding . After that, you may decide to remove some of the papers of them, or just browse one or two papers, read them again.

If you read:

5-20 papers (in selected areas, such as speech recognition), which may be enough knowledge, you can achieve a voice recognition system, but may not be at the forefront of research or make you.

50-100 papers, you might use this area (voice recognition) have a good understanding.

Step 3. depth understanding of math section

Try again from the beginning of the process to derive a mathematical text. Although it takes some time, but this is a good practice.

Step 4. Run the code

Download open source (if you can find it) and run it.

Step 5. reimplement

Re-implemented from scratch: If you can do that, then this is a strong signal that you have to really understand the algorithm at hand.

Specific methods

Do not read from cover. Instead, they need multiple passes through the paper.

Step 1. Read the article title, abstract and drawings

By reading the article title, abstract, key network architecture, and perhaps experimental section, you will be able to concept papers have a general understanding. In depth study, there are many research papers will summarize the entire paper into one or two graphics without having to laboriously read the text.

Step 2. Reading Introduction conclusion + + + Figure skip other

Introduction, conclusion and summary is the author trying to carefully sum up their own place of work, in order to clarify the reviewer why their thesis should be accepted for publication. Also, skip the "related work", the purpose of this section is to highlight the work done by other people, and to some extent the work of the relevant work. Therefore, it might be useful to read, but if you're not familiar with this topic, it is sometimes difficult to understand.

Step 3. read the text, but skip the math section

Read the text, but it does not make sense to skip the part, excellent research means that we publish something that is on the boundary of our knowledge and understanding

When you read the paper (even the most influential paper), you may also find that some parts of no use, or no meaning. So, if you read the paper, some of which does not make sense (which is not uncommon), you can skim. Unless you want to master it, it would take more time.

When we read the papers how systems thinking

When you read a paper, try to answer the following questions:

  • What the author attempts to complete (solve the problem)

  • The key element of this approach is what (boundary method)

  • What you can do (what the text of the method is insufficient)

  • Do you want something other reference materials ()

If you can answer these questions, we hopefully can reflect you have a good understanding of the paper.

Yang Feng

First of all, it is to read a paper, that's for sure.

Secondly, in the course of reading the paper, you can not just look at each other to do something, but to be good to refine the paper to solve any problem ,

Finally, what kind of paper to use the method.

Topics legislation in question in the process, we first need to define what you want to solve the problem is. Secondly, what is your solution Yes.

In this process, you can choose to address the issue considered important or very interested in the issue. You can go to find other researchers did not take into account, there is a problem a lot of room for improvement, you can choose from two angles.

Liu Zhiyuan

Each paper does not need to read all from beginning to end. An academic paper usually includes the following structure, we use numbers to mark the suggested
reading order:

  1. topic
  2. Summary
  3. introduction
  4. experiment
  5. This paper work (usually the third)
  6. references
  7. in conclusion
  8. appendix

According to this order, the basic after reading the title and abstract, roughly determine the relevance of this paper and their own research, and then you can decide whether to intensive introduction to academic and experimental results to determine the value, if you read this article to know the details of working methods. In addition, if you want to understand the relevant work and future work, it can be targeted to read the "work-related" and "Conclusions" and other parts.

Raise academic grid

  • title
  • Summary
  • in conclusion
  • chart

If the article is very important to me, to continue to the next step

  • Introduction (for areas they are familiar with, this step can be omitted, but to make it clear what the author wants to solve the problem)
  • result
  • discuss
  • method

Guess you like

Origin www.cnblogs.com/fengyubo/p/11410417.html