how-to-read-a-technical-paper

Read it once, painstakingly taking down every word you don't understand into a glossary associated with the paper. It should be written in your own language, with references to the other terms associated with the paper.

Specifically call out areas you don't think you fully understand with TODOs.

Once that's completely done, re-read it again with the glossary in-hand for reference.

Next, talk about the paper with someone. Pay attention to how you talk about it. Are you able to reference the terms of art? Do you say "that thingy" often?

Then, set a time interval, after which you will re-read the paper again. I don't have a suggestion for how long this time interval is, yet, but it ought to be long enough that some of the material will be a little fuzzy... just at the edge of you completely forgetting it.

The idea is to use active-retrieval technique, not simply for a single fact (like a flash card of a vocabulary word) but for a much larger concept (the concept in the paper).

Next, set a longer time interval, after which you will re-read the paper again. Each time you read the paper it will become less of a burden. The first time might take days to wade through the material. The second and third times will each be faster.

podcast-giant-robots-173-hiten-shah#repeatedly-do

You only get better at what we repeatedly do

 
quote-professionals-practice-till-they-cannot-get-it-wrong#quote

"An amateur practices until he can do a thing right, a professional until he can't do it wrong."

 
blog-post-deep-learning-is-easy#read-and-understand

You should try to read (and understand) the original deep belief network paper.

 

Interesting that he calls out understand separately. What would you do otherwise? Scan the words?

Referring Pages

re-read-things