[nighthawk454]

A nice trick for most papers is to skip the middle (at first). Just don’t read most of the lines of math. Focus on the inputs and outputs.

If the premise and conclusion don’t make sense on fundamentals the math isn’t likely to fix it. Most lines are literally equals signs - just walking you through some equivalencies as proof. A large statement saying “If ABC, then … (and then … and then … and then …) and finally XYZ”

The middle ‘and then’s aren’t really that important if the conclusion XYZ isn’t interesting. Or much more commonly, the ABC premise is false anyway so who cares.

Most readers I’d wager are not sitting here deciphering opaque gradient derivations every single paper. Just skip it unless it proves worthy

[robotresearcher]

Good advice. I ran an academic lab for a long time, read and reviewed a lot of papers, and trained students to read them. My process is as follows.

In order, and quickly, I look at the abstract, the picture at the top of page 2 (unless it was on page 1 like vision and graphics papers tend to do), the references, the conclusion, the lit review, if I’m still interested I try to scan the whole thing to decide what the main point of the paper is. Then if I care to, I start again and read it linearly start to finish.

If I’d agreed to review a paper, I don’t get to abort the process. Otherwise I can bail at any step.

[throw_pm23]

It's sad that papers in this area are so standardized in format though. Some of the classic papers break every convention and are written in very idiosyncratic ways. I want to read papers that can be enjoyed slowly and that change my life, not papers where I get trained to read them quickly in a certain way and that have prescribed sections and predictable figures. Life is too short for this rote work.

[idiotsecant]

The paper is a tool for conveying understanding. Standardized tools are a good thing.

[peppery]

Agreed, idiosyncratic voice is so life- and mind- affirming in papers. (Do you mind sharing examples of three papers that you did enjoy slowly and change your conceptual life?)

[nighthawk454]

Nice! Yeah that workflow feels pretty relatable. My other rule of thumb is if it's an architecture paper (system or neural net) without a diagram, I immediately bail haha