"Patience You Must Learn My Young Padawan!" - Revenge of The Sith

Good Citizen of CVPR 2018 | Notes about 'How to write a Good Research Paper' by Bill Freeman

• Start by stating which problem you are addressing. Audience must care about it, tell they why they should care about it.
• What other solutions exist, and why they aren't satisfactory.
• Explain your own solution, compare with other solutons, say why it's better.
• Talk about related work where similar techniques and experiments used have been used, but applied to a different problem.

Example paper organization

• Introduction
• Related work
• Main idea (optinal) - show using a simple toy examples to let people get the main idea
• Model (optional)
• Algorithm -- alg1, alg2, alg3...
• Experiments -- exp1, exp2, exp3... - show results , usage of graphs, bar charts etc.
• Discussion - Conclusion - you are talking the reader after he read the whole paper. how can it change? what this work open up? how we approach these problems? where the work may lead? Future work is not that useful, not write it.

Write a dynamite Intro

• Make paper easy to read for everyone.
• what your paper is about,
• what problem it solves,
• why it is interesting,
• what is really new in it,
• why it is so neat.

General writing tips

• Keep the reader in mind. What does the reader know so far? Motivate the reader for what follows? What does the reader expect next and why? Anticipate their needs: treat reader as you would a guest in your house? Would you like something to drink?
• Omit needless words. No unnecessary words, sentences - The same reason that a drawing should have no unnecessary lines. Clean writing. -- Easier to get main idea, --you can add more stuff.
• The figures and captions can help tell the story. The figure captions should be self-contained and the caption should tell the reader what to notice about the figure.
• Equations - your sentences should flow smoothly when all but the simplest formulas replaced by "blah". You can name your equation.
• Tone : kind and gracious.Written form a position of security, not competition. Ex: Elegant paper by Hertzmann etc. Your paper should not indicate that you are great but the rest are idiot.
• Don't succumb. Develop a reputation for being clear and reliable. No pressure for over-sell. Convey the right impression of performance.
• Title : Shiftable Multiscale Transforms. Should have been : what's wrong with wavelets?

Quick and easy reasons to reject a paper

• Do the authors not deliver what they promise?
• Are important references missing?
• Are the results too incremental?
• Are the results believable?
• Is the paper poorly written?
• Are there mistakes or incorrect statements?

---

Good Citizen of CVPR 2018 | Notes about 'How to Write a Good Paper' by Jitendra Malik
The title is important

• It should capture what is special about the paper.
• from the title you should have a guess of the content of the paper, recalling the title should help recall the paper.
• if every other paper in the field could use the same title, it is a bad title.
• Spend some time inventing terms-memorable and spefic are best.

Opening lines:

• Too many papers begin with "object recognition is an important problem in vision"
• Trying think about story, contribution
• WE NEED TO FIND A CATCHY OPENING LINE. Happy families are all happy, unhappy family is unhappy in its own way. Beauty of this opening line is it capture something about the novel, they are memorable, they are intimately connected to the story.
• For each part find out questions for; intro, abstract, related work, methods, experiments, conclusion

Introduction

• Reviewer - mostly decide after introduction.
• After intro, the rest of the paper is to back up the claims made in the introduction,
• Introduction : what the whole story is about, what is the problem, what is your contribution, was it worth to tacklng etc.. - otherwise you lost the audience,
• It is good to make introduction interesting and crisp,
• Histroical style - trace complexities of the problem -- create that figure one what you did it and figure 2 how you did it. -- good writing comes from good reading. READ PAPERS FOR STYLE.
• For your approach you need to be clear, experiments; decide on baselines and study people who do this very well.

Figures and Tables; one the papers and written figures should explain much such as take out figures and put in a slide and you are done. The best way to write a paper is to first give a talk on it.If you have three big ideas write three papers, not one.

---

Principles to Thrive in the Research Community by Derek Hoiem.
Part of team;

• Don't crowd the ball - don't do what most people do.
• Avoid popular topics.
• Aim ahead of your target - where are the current directions the field taking?
• In paper writing and reviewing, focus on potential to teach
• Aİm to teach and inspire - that's what people value.
• what do we teach in this paper? WRITE IT IN YOUR PAPER -- good idea and demonstrate it.
• grow and adapt in your career.
• Keep branching out and explore new topics and new roles. Just as flower explores and bounds the plant.

---

How to Do Good Research and Evaluation by Cordelia Schmid

• Planning for incremental research
• think about long-term goals / hard problems
• Find out good and necessity parts
• Do not miss baselines

How to do good evaluation : Journal Papers: Extend description of thee method, in depth evaluation, constructive feedback from the reviewers
Research in Context by Adriana Kovashka
Fact : We do research to understand images automatically

• Context : Where does the average person see images most frequently?
• Context : Who cares about this research
• Context : Who does this research?

Is this a good research topic?
Understanding persuasion

• Think about why our research matters
• Think about its impact on society
• Share datasets and involve community in your work
• reach out to groups outside the community whose involvement and perspective we could benefit from
• help foster graduate and undergraduate students
• involve undergraduates in research and get them excited

---

Doing (Good) Research by Vladlen Koltun
A collective adventure
Ways to contribute:

• Solve a problem people care about
• introduce useful tools
• build an inspiring system, demonstrate compelling app
• propose a new problem
• service and infrastructure(opencv, caffe)
• exposition(jordan's tutorials)

Identifying A Specific Technical Problem:

• formulate a larger goal - larger goal meaningful to you(there is a deaf people in my life)
• analyze bottlenecks
• understand state of the art

Making a Contribution:

• Read the papers
• look for unwarranted assumptions
• what are the limitations? When will this break? How could this be done better?

Reimplement a state-of-the-art technique:

• Reproduce results
• then bombard it with controlled experiments, look for surprises, cracks that lead to deeper realizations

Diversions

• Be on the lookout for interesting contributions
• many important results are not what the researchers set out to find -- farkli seye bakarken bambaska bisey kesfedebilirsin.

Publication:

• write your results
• quality over quantity( not many paper but one paper - think about community)
• do not compromise on methodology or ethics
• be willing to bury drafts and move on
• Publication is not the goal contribution is the goal

Make sure non of your work are sloppy and make your work significant, some interesting work arises while working on something else
High standards: Bury weak, boring, and sloppy results
Research over time

• Research baguettes research - get started follow some of the recipies, start working on a problem that will lead to more ideas more work
• keep track of favorite problems, revisit occasionally
• go back to the larger goals
• READ. A LOT.
• WRITE DOWN IDEAS. TALK TO PEOPLE

QUIET TIME FOR READING, WRITING, THINKING. Research needs a lot of work In science as in the lottery, luck favors those who wager the most -- that is, by another analogy, those who are tiling constantly the ground in their garden.

---

Back to Home