do literature search; try textbooks, conference papers, journals
it's often best to read the primary sources
skim less relevant papers/parts
(skimming efficiently and effectively is an art!)
implement your method
devise thorough tests to check that it's working
compare against simpler methods, if possible
You'll have only 20 minutes to talk, plus 5 minutes of questions,
so you'll need to be brief.
Tell us just the most important information, skip the rest.
A reasonable allocation of time (ballpark figures):
6 min. background
6 min. method
6 min. results / comparison
2 min. conclusions
Check that your audience understands variational basketweaving before
you dive into variational basketweaving with wavelets.
Don't prepare too many slides!
For me, about 2 minutes per slide.
Choose them carefully.
Diagrams & pictures are often better than text.
Keep text on each slide to 10 lines or less, say.
Slides need not say every word that you plan to utter.
Respect Murphy's Law (jammed printers, computer crashes...).
In case of calamity, just do the best you can (the show must go on...).
8-12 pages in length (if 11 point font).
This includes everything.
Introduction: Why is problem interesting?
How is it traditionally solved?
Why is there a need for new methods?
Summarize previous work (briefly).
Where did you get it? (cite papers / your contribution).
Pseudocode is good.
You can assume the reader has done the reading in 859E, but
don't assume they've read all of the papers you cite,
so summarize the relevant conclusions of those papers, briefly,
if they're important to your project.
If you found things lacking in the papers you were working from,
explain what you had to fill in.
Pictures and graphs good.
Select just the best.
What hardware, what parameter values?
Give speed and accuracy statistics.
Lengthy tables of numbers or code listings not necessary.
Does your method work as you hoped? (be unbiased, skeptical).
How does it compare to other methods (can you tell?).
Ideas for future work (what would you do next if you had time?).
At least three, probably no more than ten.
A good printer in the CS realm is "spectrum", Wean 4212.
Optional: make a web version.
Good for publicity (for course and for your talents).
Deadline: end of semester.
Put in students/yourname/project, and send me email.
Condensed is good, e.g. one paragraph summary and one or two
nice looking results images / graphs.
You can also, or alternatively, simply turn over an electronic
version of your report.
JPEG good for pictures. HTML or PS or PDF good for text.
The program latex2html may be handy.
15-859E, Hierarchical Methods for Simulation
17 Nov 1998