On reading multiple papers

It is always better to have a list of possible papers to read. This list shouldn't be static. You should add new interesting papers and remove those that are not useful/low quality etc. It is not necessary to read all papers from the list and often it is better to read multiple papers at the same time because they can complement each other.

On reading one paper

Don't read papers from the first to the last word. Always strive for understanding but also be effective with your time. Andrew Ng recommends doing multiple passes. Like in the following fashion:

• Title, abstract, figures.
• Intro, conclusion, figures again and skimming through the rest.
• Read but skim math.
• Read the whole thing and skip parts that aren't relevant.

Questions you should be able to answer

After reading the paper (and during) one should think about at least some of the following questions:

• What are the authors trying to accomplish?
• What are the key elements?
• What can you use yourself?
• What other resources do you want to follow?

General advice

• Try rederiving math from scratch.
• When it has open-source code, try it.
• If you want a really deep understanding, code it.
• Be consistent. It's not about cramming but spacing it and reading for years.
• Be wise with your time.
• Keep a paper to read around all the time.