Notes On Research Skills
Building machine learning systems often requires a lot of creativity. However, it is best to based your innovations on the results of others that comes before you instead of re-inventing the wheel everytime you build something new.
The problem is finding examples of similar ML systems can sometimes be an onerous task. In this short article, I note down some of my experience on how to do literature review, especially when you are in an industry setting like me.
Below are some quick bullet points:
- Avoid reading articles, especially when a lot of them are introductions to a commercial solutions. You do this by focusing on the research paper only. One way to do that is search the keyword you are thinking about on scholar.google.com instead of google.com
- Read the first few paper carefully, even when there are many things you don’t understand. Note down things you don’t understand.
- Take some time to consolidate your knowledge after a few reads. This step would help you create a map of knowledge, what you know and what you don’t know about the subject.
- Leverage the references and related works to explore more. In case you are exploring a new domain, usually you will stumble a complicated paper with terminologies you don’t understand. By following the references, often you can find an earlier papers that describe the basic concepts in more details.
- Find open-source implementation of the mentioned algorithms. By using open-source implementation, you can quickly try out those algorithms to understand it better before diving deeper or create a custom version if necessary.
This article will be updated regularly whenever I have new ideas or learn something.
