Pointers on Presentations
James S. Plank
EECS Department University of Tennessee
CS494/594 August 22, 2019
A fantastic resource.
●
●
Read Dr. Vander Zanden’s advice on giving effective research presentations:
http://web.eecs.utk.edu/~bvz/presentation.html
BVZ’s Advice.
The Four Golden Questions of a research talk:
Where are we now? Where are we going? How did we get there? What is left for the future?
●
– – – –
Paraphrase from Dr. Plank: Keep the audience apprised
of the bigger picture: Why should they care?
BVZ’s Advice.
Average adult attention span is 20 minutes.
Check out what your audience is doing in 20 minutes. Give them a mental break.
●
– –
●
If you change topics, provide a summary and a break.
●
Introduce concrete examples before formalism
– Introduce concrete examples before formalism – Introduce concrete examples before formalism – Introduce concrete examples before formalism
With software, do a demo before explaining the details.
BVZ’s Advice.
●
BVZ & Fonts.
●
● ●
Dr. Vander Zanden says that “sans-serif” fonts are better than not.
I don’t agree. I’m fne with Times-Roman.
However, some fonts realy suck. Stck wit te normal ones like Times-Roman, Helvetca or Arial. Don’t use tis font.
And changing colors for no good reason is distracting.
●
BVZ’s Advice.
Limit your bullets per slide to 5 or 6. Make your bullets short and snappy.
People read all of your slides before they start to listen to you.
● ●
–
● ● ●
Do not use complete sentences unless they are pithy. Do not clutter your background.
Put a header on every slide with a title.
BVZ’s Advice.
● ● ●
A picture is worth 1000 words.
Screen snapshots and code listings are useless. Special effects are typically distracting.
BVZ’s Advice: During the Presentation
You are your own worst critic. Do not read your slides.
Shockingly, it’s harder to read your slides when your bullets are pithy.
Bring a glass of water.
Keep the presentation moving
Don’t get bogged down by questions. Control the talk.
● ●
–
● ●
– –
●
Face your audience, not the screen or your notes.
●
●
●
●
●
My additions to BVZ’s Advice
Talks and papers are two different beasts Papers need to be complete and correct.
Talks are there to get your audience interested in your work. And then to read your paper.
Corollary: Talks need to be neither complete nor correct.
But they need to sustain the audience’s interest in your work!
My additions to BVZ’s Advice
Allow the audience to navigate where you are.
Outline in the beginning
(with timings for long talks)
Tell them where you are.
Remember to summarize between sections.
●
– – – –
Know your slide style and go with it. Know your talking style and go with it.
(First few words of every slide)
If things get too dry, give the audience a break.
● ●
–
●
●
– – – –
My additions to BVZ’s Advice
Prepare and iterate.
Pictures, pictures, pictures.
Slides filled with text are lazy (including these) Unreadable graphs and graphics are lazy.
What’s good in a paper is often not good in the talk.
If you can, know your venue & your equipment.
●
–
Neither powerpoint nor openoffice/libreoffice are really portable.
If someone wants you to put your talk on a jump drive for their computer, do due diligence.
–
● ● ●
My additions to BVZ’s Advice
SPEND TIME ON YOUR GRAPHS!!!!
What’s good in a paper is often not good in a talk.
Strive for clarity, simplicity, cleanliness.
My additions to BVZ’s Advice
Don’t be afraid to annotate/highlight things that are important in your graphs.
Put the same graph over multiple pages and highlight different things.
Here’s an example from a talk I gave at USENIX FAST in 2013.
●
–
–
Performance
● ● ● ●
3.4 GHz Intel Core i7-3770
256 KB L2 Cache, 8 MB L3 Cache
Performing buffer-constant on various buffer sizes Lots of comparisons.
Performance
● ● ● ●
3.4 GHz Intel Core i7-3770
256 KB L2 Cache, 8 MB L3 Cache
Performing buffer-constant on various buffer sizes Lots of comparisons.
Performance
● ● ● ●
3.4 GHz Intel Core i7-3770
256 KB L2 Cache, 8 MB L3 Cache
Performing buffer-constant on various buffer sizes Lots of comparisons.
Performance
● ● ● ●
Memcpy & XOR are as you’d think.
“Anvin*2” is a technique for multiplying 128 bits by two in any
Galois Field with just a few SSE3 instructions. (Linux Kernel RAID-6).
3.4 GHz Intel Core i7-3770
256 KB L2 Cache, 8 MB L3 Cache
Performing buffer-constant on various buffer sizes Lots of comparisons.
Performance
If you’re fast enough, you can see effects of saturating the L2 and L3 caches.
● ● ● ●
3.4 GHz Intel Core i7-3770
256 KB L2 Cache, 8 MB L3 Cache
Performing buffer-constant on various buffer sizes Lots of comparisons.
Too big
Performance
Traditional techniques (Rizzo, Jerasure, Onion Networks) don’t get close to cache line speeds.
(BTW, both axes are log axes)
Performance
Non-traditional techniques do better, but require amortization for w=8 and w=16.
Performance
●
Our techniques perform identically to “Anvin*2” for w = 4, 8 and 16.
Cache limited.
● ●
Alternate mapping makes a significant difference. w=16 and w=32 show some amortization effects.
–
My additions to BVZ’s Advice
●
●
●
●
● ●
When people ask for your slides, give them PDF and not PPT / ODP.
Make sure that the PDF has citation information on page one.
Go through your slides, and make sure that the PDF looks good (check your animations).
You can give them 1000 pages – you don’t care about their paper and ink costs.
They will be lazy and will steal your slides. So don’t let them make you look bad!
● ●
Don’t be hungover. Mind the onion loaf.
My additions to BVZ’s Advice
Go over your presentation before you give it.
Even if you have given it before.
If you are inexperienced, go over it “live”. If parts are really hard, then script them.
●
– – –
My Biggest Disaster – DEC SRC, 1990
The senior people present:
https://en.wikipedia.org/wiki/Edward_D._Lazowska https://en.wikipedia.org/wiki/John_Guttag
https://en.wikipedia.org/wiki/Kai_Li
Kai Li Princeton. (my advisor)
Founder of Data Domain (sold to EMC
in 2009
for 2.1 billion)
https://en.wikipedia.org/wiki/David_Cheriton
https://en.wikipedia.org/wiki/Butler_Lampson
Butler Lampson
SRC. Founder of Xerox PARC. Turing award winner.
Ed Lazowska Washington. Multiple advisory
boards to congress &
the president.
David Cheriton
Stanford. “Professor billionaire” 580th wealthy person in the world (Forbes)
John Guttag MIT.
EECS dept head
1999-2004.
My Biggest Disaster – DEC SRC, 1990
The students present (that I remember):
What did I do?
●
●
●
●
I had given the talk three months before to 300 people, so I didn’t even give it a browse.
I went to Gordon Biersch the night before and had about 6 beers.
I didn’t mind the onion loaf.
And I got to watch faculty and students alike view me with disdain and disappointment.
(Go over the hall of shame)