Ask HN: How do you conduct literature survey for your research? | Hacker News Reader

Ask HN: How do you conduct literature survey for your research?

I am getting started with graduate research in Machine Learning and have a problem at hand. Other than consulting your research advisor, what methodologies do you follow to get the most information to formulate a new (and hopefully better) approach to the problem.

I am particularly interested in your discovery and organizational methodologies for the materials you read during literature survey.

View on HN

You have to read a lot of papers. Begin with survey papers for your field (ML is still a very broad topic, so see if you can narrow it down). Use Google Scholar to find papers by searching for various keywords pertinent to your subject. Then, you start looking at the citations (in both directions). Look at the papers that are cited by that paper, as well as the papers which cite the paper that you are looking at. Do this over and over (perhaps weeks and/or months), and you will have a good handle on the state of the art for that particular field.

Books will not give you the state of the art information, because they are outdated the moment that they are published. Wikipedia is often too light in its coverage of the latest research. Reading the latest research papers is truly the only way to know what the latest research is.

> Begin with survey papers for your field

What exactly do you mean by survey papers?

I do have a narrow set of keywords that I should be looking for but only from common-sense. I have been compiling a list of generic approaches that might be helpful down the line.

> Reading the latest research papers is truly the only way to know what the latest research is.

I've observed that often research papers get light on details of implementation just to not deviate from the core idea. While understanding the core idea helps, various new black-boxed techniques emerge in every paper (techniques which are cited in another paper itself). I've tried to prevent myself from being overwhelmed by keeping them as black-boxes for the sake of understanding the core idea, but when it comes to implementation, I feel the urge to understand those techniques intermediate techniques in detail which often put me on a longer path. Is this expected or should I adapt faster?

Reply on news.ycombinator.com