Skip to content

papers

Jump to bottom
Eugene Wu edited this page Nov 4, 2016 · 2 revisions

Setup

Managing your papers

  • I like to keep one repo per project, but makes releasing code more of a pain:

      SUPERPROJECT/
    docs/
      sigmod_16/
      vldb_16/
      README.md            -- all project notes
    src/
      allmycode/

  • See /docs/example16 for an example folder structure and latex template

  • David Blei's lab manages their publications in a single repository: https://github.com/blei-lab/publications

Writing papers

See the notes of questions to answer before picking a project. These are the minimum questions to address in your paper.

It also helps to see how other papers evolve. I've built a tool called latexsnapshots that will generate snapshots of a paper as it evolves.

Responding to reviewer comments

Collaborative Writing

I find the following writing process useful when working writing with others. This also assumes that the paper is not written last minute :)

  • For the first several drafts, use google docs/quip/paper or any other collaborative writing tool. During this process, the goal is to agree on the sequence of arguments and desired figures to have in the paper. I don't care as much about technical notation.
  • Once the structure is agreed upon, ship the text into latex and edit there.
  • Some collaborators like using ShareLatex or Overleaf. I don't have a personal preference, so I go with whatever my collaborators want.

Reviewing Papers

    tldr; think like an engineer and look for things to take

Fill in the following form, with an emphasis on the positive aspects of the paper:

# Positives 

## What is the key idea in this paper?  Why is it a good idea?

## Are there other reasons why this could be a good idea not mentioned in the text?  Suggest them.

## Are there other applications where you could use these ideas?  Describe them

## Which of the experiments helped you understand the paper?

## What else did you like about this paper?

# Improvements

## What writing improvements could have made it easier to understand the paper?

## What techniques would make the ideas _even better_?  In what ways

## What experiments/plots would help you better understand how well the ideas work?  
## Are they _possible_? (remember: getting private data is hard)

What other people say about reviewing:

Clone this wiki locally