Robotics 2
How to Write a Paper
Giorgio Grisetti, Cyrill Stachniss,
Kai Arras, Wolfram Burgard
1
Why Writing a Paper?
General:
Documentation of scientific results and
findings
Individual:
Document your scientific results and
findings
Communicate with colleagues
2
Potential Impact of a Published
Paper
Scientific importance
Improved evaluations
Better job opportunities
Better chances for getting funding
Reputation
However, a bad paper can have a very
negative impact on your reputation.
3
When Should I Write a
Paper?
Is my scientific result – at least to the
best of my knowledge – novel?
Did I consider sufficient related work
to give a positive answer?
Do I have experimental or analytical
results that justify this?
Sources for the General
Description
http://www.daniel-lemire.com/blog/rules-to-write-a-good-research-paper/
Further reading:
http://www.findaphd.com/students/life2.asp
Fun:
http://members.verizon.net/~vze3fs8i/air/airpaper.html
5
How to Start?
Sexy start: tell the reader early why he should read
your paper.
Don’t summarize, sell!
A good abstract answers the question
“why should I read this paper?”,
it does not summarize the paper.
Convince us early that your paper is important.
Recipe for a good 4-sentence abstract is:
1. state the problem
2. say why it is interesting
3. say what your solution achieves
4. say what follows from your solution.
6
What else Should be in the
Paper?
You should clearly say what your
contribution is.
Reviewers are lazy, they do not want to
have to figure out what your message
is.
Spend some time telling the reader
exactly what your contribution is.
Spell it out, do not assume reviewers will
read the paper carefully.
7
What else Should be in the
Paper?
A review of related work in the
introduction: you can relate your own contribution
to all of the related work.
A large reference section: people like to be
cited, so make sure you cite every paper that
might have some relevance.
Experimental evidence: you need to confront
your idea with the real-world and report on how
well it fares. Compare explicitly your results with
the best results elsewhere.
Acknowledgement of the limitations of your
work.
8
What else Should be in the
Paper?
Relevant and non-obvious theoretical results: it
is easier for people to build on your work if there
is some theory and it helps give people confidence
in your work.
Pictures! Really, even if you feel silly doing it or
that you think you can’t draw. A picture can help
tremendously in communicating difficult ideas.
Original examples over original data sets.
A conclusion telling us about future work and
summarizing (again) the strong points of the
paper.
9
Pedagogy and Style?
Use strong verbs (replace “we made use of categorization” by “we
categorized”).
Always give the example first, and the result next.
Use as few parenthesis, footnotes and bold characters as you can.
Use a spell checker. Just do it.
Use a tool such as style-check.rb to check for verbose phrases and
other common mistakes.
Learn about and use unbreakable spaces.
Do not use negations…
Avoid UA (useless acronyms).
DUAT: Do not use acronyms in titles.
Your writing will be in an active voice… (hint: avoid the verb “to
be”) (Every time a student uses passive voice, God kills a kitten).
Employ uncomplicated terms.
Learn to use the em-dash—it is a good friend.
Short sentences—no more than 15 words—are better.
Make your research papers easy to skim by using meaningful
section headers, bullet points and simple figure.
10
Words You Can do without
Temporal words such as “now”, “next” are
either useless or a sign of a bad structure.
Avoid the future tense (the word “will” in
English) to refer to something coming up
next in the document.
Most adverbs such as “very” are useless in
a research paper.
Keep your emotions in check: the reader
may not care for your surprise, pleasure
and sadness.
11
Things to Check
Are section headers consistent with respect to case? (”Our
Methodology” versus “Our algorithm”)
Do the figures look nice? Are the fonts large enough for easy
browsing? Are they readable once printed out in black-and-white?
Can we see any compression artifacts?
If the page limit is x pages, do you have an x pages long paper?
Do you have at least one figure?
Is the layout of each page elegant?
Do you have widows or orphans?
Did you spell check?
Do you have a step-by-step toy example for every new algorithm
being introduced? Present your examples early.
Are all equations arithmetically correct?
Can you replace some mathematical notation by plain English?
Are all terms defined?
Is the mathematical notation consistent? (If you use t for time in
the first section, do you use t to note the term in the second
section)
12
Things to Check
Are the title and the abstract geared toward making the paper
attractive?
Do you summarize your contribution in the introduction?
Is the bibliography consistent? (If you abbreviate first names once,
do it all the way through. If you have page numbers once, have
page numbers throughout.)
Is the spelling of all proper names correct? You would hate to get
your paper reviewed by someone who would find his name misspelt
in your paper.
Are the captions correct? Do you put the table caption before or
after the table? Do you put the figure caption before or after the
figure? Do you center captions or not?
Do you refer to a figure as “Fig. 1″ or as “Figure 1″? Which one is
correct?
Are all internal references correct? If you refer to Fig. 10, does
Figure 10 exists? (Some LaTeX package can mess this up, so
always check!) Are all tables and figures referenced in the text?
If this is a recurring conference or a journal, have you compared
your paper with ten or so other articles to make sure that yours is
consistent with how these other papers look and feel?
Do you use the right fonts? Be watchful: sometimes the font for the
section header can differ from the font used in the main text.
13
The Reviewing Process
Reviewers need to figure out whether the paper is
an advance over the state of the art or not.
This includes to check whether the paper is
theoretically and experimentally sound.
Take the comments of the reviewers seriously and
modify your paper according to their
recommendations.
Look for the most critical statements.
Even if it should get rejected:
What should be changed to improve it?
Which related work should be considered?
To which approach should it be compared?
Can the writing be improved?
And simply accept the recommendations.
“Fun:” http://www.youtube.com/watch?v=-VRBWLpYCPY
14
Editing Recommendations
Use LateX!
There is a class file for almost every
desirable feature. But there is no need
to use all features.
Use a versioning system such as CVS
or SVN, especially when you
collaborate with colleagues.
Be consistent: Fonts, equations,
citations, …
15
Figures
Colors are great, but can one
distinguish them in a b/w printout?
Avoid screen-shots of non-iconic
pictures.
Are the fonts large enough?
Are the lines thick enough?
Is the resolution high enough?
16
Examples (Color)
new new
old old
17
Examples (Font)
18
Examples (LaTeX)
Text in equations
vs.
Hard spaces
19
Examples (Images)
Initialization
a>0
20
Structure of a Paper
Abstract
Introduction
Related Work
“The Approach”
Experimental Results
Conclusions
Bibliography
21
Example Paper
Probabilistic Navigation in Partially Observable
Environments, R. Simmons and S. Koenig,
IJCAI '95, Montreal Canada, July 1995.
http://www.cs.cmu.edu/~reids/papers/probNav.ps.gz
Why this paper?
representative of a wide class of good papers
interesting robotics paper
22
The Abstract
Write one statement
about the general
problem
Tell what this paper is
about
Describe how it solves
the problem
Emphasize what is new
or better
Mention the evidence
indicating the
advantages of the
proposed approach
23
The Introduction
1. Start with a motivation
2. Tell what this paper is about
3. Explain what makes this work
relevant
4. Maybe add a section about the
structure of the paper
24
Potential Section on the
Structure of the Paper
This paper is structured as follows. After discussing
related work in the following section, we will present
our <<name or property>> approach to <<the
problem>> Section III. In Section IV we then will
present experimental results demonstrating the
advantages of our (algorithm / formalism /
representation).
Questions:
Such a paragraph can also be left out as it is relatively
generic
Does it make sense to write “The conclusions will
conclude the paper”?
25
Related Work
Put your paper into
the scientific
context.
What is the work
previously done by
others?
Describe for every
other paper, how
your work differs.
Summarize in which
way your paper
goes beyond the
state of the art.
26
Citations
cite scientific papers rather
than text-books
cite original work rather than
overview articles
cite novel work
cite relevant contributions
(outstanding conferences and
journals)
don’t forget the old stuff
talk to others (advisor,
colleagues) about what
relevant papers are
limit self-citations to an
appropriate number
10-20 citations, depending
on the amount of related
work
reduce information in
citations to the relevant
amount
27
The Technical Part
1. Describe the work you have done in a way
that other( student)s are able to re-
implement it.
2. Describe the foundations, if necessary.
3. Give sufficient technical details.
4. Include the underlying equations!!!
5. Add figures to make your description more
easily understandable.
6. Mention the advantages of the approach.
7. Describe the complexity.
28
2
29
5
30
5
31
5
32
5
33
The Experimental Results
1. Explain why you make the individual experiments
2. Motivate simulation and real-robot experiments
3. Give a detailed explanation of the individual
experiments
4. Eventually, use graphs and tables to summarize
your experiment.
5. Compare your approach to alternative ones
6. Perform statistical tests indicating that your
approach is “significantly better than alternative
techniques”
34
1
35
5
4
7
2
36
5 5
37
The Conclusions & Outlook
1. Again describe the
approach presented in
this paper
2. Again mention the
advantages and what is
novel compared to
previous approaches
3. Mention the
implementation and the
successful outcome of the
experiments
4. Potentially discuss options
for future work
Don’t be too critical on
your own work
Don’t be too enthusiastic
about what else could
and maybe should have
been done.
38
Finally…
Papers are written to advance the
state of the art.
It is better to focus on the content
rather than on the appearance
Still, the appearance is also important
(show your perfectionism)
Talk to other people (and your
advisor)
There maybe are alternative ways of
writing papers.
39
