How to Write a Paper Ralph Martin Tuesday, September 24, 2013

How to Write a Paper
Ralph Martin
Tuesday, September 24, 2013
Two Things are
Necessary
Content
Style
Without good content, style is irrelevant
Without good style, content is not enough
Papers can be rejected for poor writing
Tuesday, September 24, 2013
Content
This is your great new idea
The whole problem might be new
Or you may have a new way of doing a step
Or you may put existing ideas together to
solve a problem in a new way (this is weaker)
If the problem is not new, your method must
be an improvement
Tuesday, September 24, 2013
Abstract and Paper
The abstract is a self-contained document
The reader may not have the paper, just the
abstract
The abstract should tell the reader whether
to read the paper or not
It should summarise the key contributions of
the paper, and other important content
Tuesday, September 24, 2013
Paper Structure
Introduction
Related work
Theory, Algorithms
Experimental results
Conclusions
Tuesday, September 24, 2013
Introduction - 1
Don’t be afraid to partly repeat the abstract
The abstract is a separate document
State what problem you are solving in a
straightforward non-technical way
Who are the potential users of your idea?
Why is the problem of current interest?
Tuesday, September 24, 2013
Introduction - 2
Explain what your contribution is to the topic
Explain clearly how what you have done is
different to previous work
Summarise the content of the rest of the
paper, making clear which parts are novel
Tuesday, September 24, 2013
Related Work - 1
Include all major similar work
Do not summarise each piece, but analyse it
Show why each reference did not solve your
problem:
solved a related problem
gave an inefficient, limited or inappropriate
solution
Tuesday, September 24, 2013
Related Work - 2
Either you must be the first person to work
on a new problem
Or, otherwise, it is not good enough to be
different:
Your work must be better in some way
Compare the references to your work,
showing how your ideas are better
Tuesday, September 24, 2013
Related Work - 3
Some references will be to work you use
E.g. an algorithm, a set of test data, etc.
You should describe such work and how you
use it
Make sure all due acknowledgements are
made
Tuesday, September 24, 2013
Overview - 1
It is often helpful to give an overview of
your new method, theory, etc
This can be in the Introduction, or in a
separate section
It gives the reader a map to follow the rest
of the paper
Tuesday, September 24, 2013
Overview - 2
Define technical terms
Explain which parts of your method are
automatic, and which require user input
State any assumptions about the kind of
input, the form needed for the output, etc
justify they are realistic
explain how they simplify the problem
Tuesday, September 24, 2013
Detail Sections
For each main Section and Subsection
summarise what it is about, what its
purpose is, how it fits in overall
give the details
summarise what has been shown,
demonstrated, proved etc.
Tuesday, September 24, 2013
Theory - 1
For each new theoretical result, make sure
there is:
a statement of how it fits into the overall
solution to the problem
a clear sub-problem goal that is
addressed by the theoretical result
a clear statement of the result
Tuesday, September 24, 2013
Theory - 2
Where appropriate, give simple examples to
show how the theory works in practice
Tuesday, September 24, 2013
Algorithms - 1
State clearly for each algorithm
How does it fit into the solution to the
overall problem
What are the inputs?
What are the outputs?
Give a pseudocode listing
State all steps in correct logical order
Tuesday, September 24, 2013
Algorithms - 2
If there are tuning parameters
what values did you use?
how did you choose them?
how stable is the algorithm if they are
changed?
Tuesday, September 24, 2013
Algorithms - 3
Make clear
where user input is required
what form the user input takes
how the user chooses suitable input
Tuesday, September 24, 2013
Experimental Results - 1
Each experiment should have a purpose:
state clearly what experiments were done
what settings were used
state why each experiment was done
give the results: outputs, timings, etc
draw conclusions from the results
Tuesday, September 24, 2013
Experimental Results - 2
Give results to show behaviour
with different types of input data (typical,
extreme)
different settings of tuning parameters
Give comparisons with existing methods or
alternative methods where feasible
Tuesday, September 24, 2013
Experimental Results - 3
Make explicit what you infer from the
results, e.g.
how method scales with size of input
how method is affected by tuning
parameters
how results compare to other methods
Tuesday, September 24, 2013
Experimental Results - 4
Give objective results if possible, e.g.
Graph showing timing as a function of size
of input, best fit to graph
Tests on the same data using different
methods
Clear and meaningful metrics by which
you can decide which method is best
Tuesday, September 24, 2013
Experimental results - 5
Be honest:
what was successful and worked well
what did not work so well, and why not
what the limitations of your method are,
and how you might overcome them
Tuesday, September 24, 2013
Conclusions
Summarise the novel contributions of the
paper
Explain why your results are significant and
useful
Summarise the experimental evidence, or
theoretical proofs, which demonstrate that
your new method is better
Tuesday, September 24, 2013
References
Try to follow the journal’s required style!
Most important are
consistency
completeness (publisher, page numbers,
date,...)
Tuesday, September 24, 2013
Figures - 1
Each figure should have a definite purpose:
to help clarify the text
or demonstrate a particular experimental
result
Give each figure a short caption
Refer explicitly to the figure in the text of
the paper, saying what the figure shows
Tuesday, September 24, 2013
Figures - 2
Don’t make figures too small
Make figure labels consistent with the text
Place figures as near as possible to where
they are first mentioned in the text
Make sure figures are not misleading
Tuesday, September 24, 2013
Other issues - 1
Do not just state how your method works
Say what alternatives were possible
Justify the choices you made
Tuesday, September 24, 2013
Other issues - 2
Make sure all mathematical symbols and
abbreviations are defined
Use standard terminology and notation
Make sure all steps are explained, at least
by a reference
Put just one good idea in each paper!
Tuesday, September 24, 2013
Other issues - 3
Use consistent terminology:
keep the same name for the same thing
Use consistent mathematical symbols:
use meaningful symbols like P for a plane
use related symbols for related quantities
use each symbol to mean only one idea
Tuesday, September 24, 2013
Other issues - 4
Remember that
stating something
explaining something
are different!
Do not just say what you do, but why, how it
works, what the aims are, etc
Tuesday, September 24, 2013
Other issues - 5
Do not assume the reader shares your
understanding of the problem
understanding of its background
overview of the approach
Things obvious to you may not be obvious to
the reader – what does the reader know?
Tuesday, September 24, 2013
Other issues - 6
Avoid long captions, footnotes, parenthetical
remarks – keep the text linear
Avoid buzzwords like paradigm
Use technical words in their correct
technical sense e.g. vital = necessary for life
Tuesday, September 24, 2013
Checklist - 1
Is your paper ready?
(And are you sending it to the right place?)
The best way to tell if it is ready is to get a
refereeing form for a journal or conference,
and score your own paper!
Typical points are as follows...
Tuesday, September 24, 2013
Checklist - 2
Does the paper address a topic of interest to
the journal or conference?
Is it written in a style appropriate to that
readership?
Does it address an important topic?
Is it clear what the novel contribution is?
Is it technically sound?
Tuesday, September 24, 2013
Checklist - 3
Is the writing clear?
Are the Figures clear and useful?
Is the paper too short or too long?
Could someone else reproduce your work?
Are the references adequate?
Is the Title suitable?
Tuesday, September 24, 2013
Homework - 1
Short Homework: Read
How to Get Your SIGGRAPH Paper
Rejected
Jim Kajiya
http://www.siggraph.org/sites/default/
files/kajiya.pdf
Tuesday, September 24, 2013
Homework - 2
Longer Homework: Read this excellent 100
page book!
How to Run a Paper Mill
John Woodwark
http://www.johnwoodwark.com/inge/docs/
Pmill.pdf
Tuesday, September 24, 2013