1 of 31

Share our know-how and refine each other’s research skills.

​

Currently planned seminars:

  • Paper reading and organization (this one)
  • Picking and choosing conferences (Martina)
  • Coding practices for research (Mattia)
  • The developer toolbox (?)

The “Awesome Research” seminars

QR to this series

presentations

2 of 31

Reading fast and slow

How to search, read, and search (again) papers at scale

​

​

​

​

​

​

Mattia Setzu

Research fellow @ UniPI

27 times and running reader of the month

Me after Salvatore fires me

( I can finally read in peace)

3 of 31

  • Reading is useful
  • Cataloging is even more useful
  • You can read in different and well-defined ways
  • You can combine said ways in different situations
  • With the right tools, you can “read” better and more effectively

I will* convince you that

​

*probably

4 of 31

  • Get an introduction to a topic/field
  • Update our knowledge of the literature
  • Get new ideas
  • Review papers
  • Submit papers

Why do we read here in academia?

5 of 31

Reading slow�The archeologist

​

  • Some idea of what to look for
  • Obsessed w/ detail
  • Constantly on the field
  • Categorizing ad libitum

Reading fast�The librarian

​

  • Doesn’t even read all books
  • Categorizes online
  • Master at searching

6 of 31

Reading slow�The archeologist

​

  • Some idea of what to look for
  • Obsessed w/ detail
  • Constantly on the field
  • Categorizing ad libitum

Reading fast�The librarian

​

  • Doesn’t even read much
  • Categorizes online
  • Master at searching

Reading?�The pharaoh*

​

  • Reads sparingly
  • Only brief texts
  • Yet knows what’s going on

​

*The pharaoh’s guards were Sardinian.

7 of 31

​

Surveys, technical reports, papers

​

Surveys, papers

​

Reports, digests,

tweets

Reads…

8 of 31

​

Surveys, technical reports, papers

Week or less

​

Surveys, papers

Two weeks/Month

​

Reports, digests,

tweets

Weeks/Months

every…

Reads…

9 of 31

​

Surveys, technical reports, papers

Week or less

Annotates

​

Surveys, papers

Two weeks/Month

Annotates/�Indexes

​

Reports, digests,

tweets

Weeks/Months

Plans

Reads…

every…

and…

10 of 31

Reading slow�The archeologist

​

  • Costly
  • Effective

Reading fast�The librarian

​

  • Fast
  • Frail

Reading?�The pharaoh

​

  • Fast
  • Generic

learns through

catalogues for

knows through

searches for

11 of 31

Thinks in papers/tags

Thinks in tags

Thinks in QA

Reading slow�The archeologist

​

Reading fast�The librarian

​

Reading?�The pharaoh

​

learns through

catalogues for

knows through

searches for

12 of 31

  • Get an introduction to a topic/field [A]
  • Update our knowledge of the literature [A/L]
  • Get new ideas [P]
  • Review papers [A]
  • Submit papers [A]

Why do we read here in academia?

13 of 31

  • Google Scholar
    • Limited number of authors
    • Try to avoid authors of the same group (overlap)
    • Sparingly use the “articles similar to” feature
    • DON’T read emails directly filter them out*
  • Arxiv (very sparingly, arxiv sanity is better yet)
  • Conference proceedings (filter by ranking)
  • Github (awesome repos, researcher groups)
  • Twitter (kinda difficult)
  • Usually no need to catalogue everything ASAP

Tools of the slow reader: excavators

​

​

​

*https://github.com/msetzu/google-scholar-pls-stop

Dug A. Boring

14 of 31

Try to answer, in the least amount of time/text:

    • What problem is the paper tackling?
      • Title
      • Abstract
      • First paragraphs of the introduction
      • Figure 1

​

(Quick/Slow)read

Dug A. Boring

Dug A. Boring

15 of 31

Try to answer, in the least amount of time/text:

    • Where is the novelty?
      • Title
      • Final paragraphs of introduction
      • Abstract

​

(Quick/Slow)read

Dug A. Boring

Dug A. Boring

16 of 31

Try to answer, in the least amount of time/text:

    • What approach is applied?
      • Figure 1
      • Background
      • Introduction

(Quick/Slow)read

Dug A. Boring

Dug A. Boring

17 of 31

Try to answer, in the least amount of time/text:

    • What are the main contributions?
      • Last paragraphs of introduction
      • Conclusions

(Quick/Slow)read

Dug A. Boring

Dug A. Boring

18 of 31

Try to answer, in the least amount of time/text:

    • Are they convincing?
      • Experiments (figures first)

(Quick/Slow)read

Dug A. Boring

Dug A. Boring

19 of 31

  1. Apply the above quickread techniques
  2. If any of them fail, BIN the paper

Quickread

Dug A. Boring

Dug A. Boring

20 of 31

paper info�(tags included)

papers

all tags

collections

Tools of the slow reader: catalog

21 of 31

Paper info

Automatic

​

Collections

Depends on your application

​

Tags

See next

Tools of the slow reader: catalog

22 of 31

tags

color-coded

markers

notes

Tools of the slow reader: catalog

23 of 31

Tags*

  • Task
  • Data
  • Algorithm
  • Model

​

Color code

  • Intro
  • Method
  • Setup
  • Results

​

Notes

  • Paper summary (review)
  • Unclear points (student)

​

​

​

*https://msetzu.github.io/about/research#public-library

Tools of the slow reader: catalog

24 of 31

  • Never read when tired or in the evening
  • Never split a read
  • Most of the time, only really need to understand:
    • task
    • approach�hence, quickread

​

Tips from the good archeologist

Dug A. Boring

Dug A. Boring

25 of 31

  • You Only Read Once (YORO)
  • Score To Ignore (STI)
  • No Shame In Binning (NSIB)
  • No Need for Copies (NNC)
  • You Only Catalogue Once (YOCO)
  • Tag Well, Search Well (TWSW)
  • Highlight Well, Ignore Later (HWIL)

Proverbs from the good archeologist

Dug A. Boring

Dug A. Boring

26 of 31

  • Emulate tag hierarchy, e.g., lat:disentanglement, lat:representation, etc.
  • Save searches for fast indexing
  • Search everything (notes too!)

Tools of the fast searcher: tags

Theresa Lotte O’Docks

27 of 31

  • Always think high-level (no papers, no tags)
  • Aggregate search results w/ tags, e.g., __project:globelief
  • Ask questions, write one-sentence answers
  • Summarize in small docs
    • Notion
    • Dropbox Paper
    • Obsidian (coolest note-taking app ever)

Tools of the good pharaoh: docs

Sapir A. Middut

28 of 31

Tools of the good pharaoh: docs

Sapir A. Middut

29 of 31

How do I apply this? New field

​

Asks high-level questions

Find the (un)known unknowns

​

Set up high-level tags

Informing the archeologist

​

Gather surveys/papers

Quick reads

Initial reports to the librarian

​

Sets up tags

Reports questions to the pharaoh

​

Decides what to answer

Sets up document

Informs the librarian

​

Sets up searches

Informs the archeologist

​

Deep reads

Tags

​

30 of 31

​

Quick read

Tag

Report new tags to librarian

​

Updates tag registry

May report back to pharaoh

​

Updates doc/tags

May set up a new question

​

How do I apply this? Known field

  • Try to stay in this mode as often as possible
  • Always quick read as an archeologist unless the pharaoh�needs better answers
  • Can add 90% of the final tags when reading the abstract

31 of 31

Reading fast and slow

How to search, read, and search (again) papers at scale

​

​

  • Read WITH PURPOSE
  • Read MINIMALLY
  • Think MAXIMALLY
  • Write SIMPLY

QR to this series

presentations