Reading research papers is sometimes difficult and very time-consuming. So the most important thing you need to guarantee is that you are always reading the papers you are really interested in. You must firstly list the key points you are working on and find relevant research papers to read.
Generally, we should read about 10-20 papers in a week. But that does not mean that we need to understand every detail of a paper we just read. Based on the level of our interests, we should have different strategies towards those papers. Here are 3 steps which could help you succeed in reading machine learning papers.
In the first step, our goal of reading this paper is to get a high level understanding of it. The following issues are waiting to be solved:
- The main contribution of this paper
- What problem does it want to solve
- Am I interested in this paper? Which Part?
Then more specifically, how to read one paper following above steps? First, we could read the title:
So if you are interested in word embeddings applications in recommendation system, just go ahead and read the abstract, which could help you get a brief idea of what this paper is doing:
Then you could go through the paper and read the title of each sub sections and some core sentences decribing the main ideas. At last, the conclusions and future work should also be given enough attention:
If you finish this step, try to answer the above questions and if is not clear, go to read the paper again. Moreover, a warm tip is that I would also read the reference and see if I read the related papers.
After that, a great idea is to compare your opinions about this paper with other people. Here a very interesting website called machine learning reddit let you know the most recent research directions in machine learning and compare your opinions towards one paper with others and you will find some interesitng views from other researchers.
Ok. If you still have strong interests of this paper, go to the next step.
In the second step, you could start to read every English word of a paper and try to understand the details of how the author solves the problem listed in the paper. Some nice papers also have github links so that you could get a deeper understanding of, for instance, the algorithm by reading the source code in the github. In this part, you should have a lot of questions. Then do not hesitate to ask questions in the following links:
There are also many other resources in website which could help you understand the paper. Resources such as tutorials, workshops are very benificial to your improvement in the relevant topics. Hence just google it, you are bound to find many useful things. However, I won't go through details of math in this step.
In this step, I like to take notes on Jupyter Notebook and record my steps in reading this paper. It's a good habit to note your improvement in machine learning!
Math is the core part in machine learning papers. In this part, I would get deep understanding of one paper by work out all the mathematical inference. It is always the most difficult step but once you finished it, you would have enough confidence to talk with other researchers about the ideas of this paper.
Here I give some interesting machine learning links which could help you find interesting papers: