How to give good talks
CC-BY
Fabian M. Suchanek
62
When do you give talks?
2
As a person in industry:
• when presenting an idea for a start-up to investors
• when summarizing a project to your boss
• when invited to a conference, event, or university
As a student:
• to defend your master’s thesis
• to present a student project
• when applying for a job
• (in the Softskills seminar)
As a PhD student or scientist:
• to present an idea to colleagues
• to present a scientific contribution at a conference
• to present your work at an institution
This lecture: academic talks
(although much of it applies
elsewhere, too)
Why should you care?
3
• boring talks are a key annoyance in the scientific culture and elsewhere
• good talks will keep the audience receptive, and thus
make them understand your point better
• good talks will inspire feedback and questions, which is helpful for you
• good talks will add to your positive reputation, and
open the way for citations, collaborations, and invitations
[picture by Ville Säävuori]
Disclaimer
4
A talk that does not follow these recommendations is not necessarily “bad”.
A talk that does follow these recommendations is not necessarily good.
(The recommendations are still evaluation criteria for the Softskills Seminar.)
The following are my own recommendations for what I consider “good talks”.
Other people may have other opinions.
How to give good talks
5
•
Structure
•
Style
•
Slides
•
Speech & Body Language
•
Preparing the talk
•
Giving the talk
•
Being in the audience
Common Structures for Talks
6
Scientific talks:
1) Motivation
2) Optional: Related work
3) Approach
4) Experiments
5) Conclusion and future work
There are many different good ways to structure talks!
Here are common ones (but there are others)!
This is what we will focus on
Application talks:
1)
Curriculum vitæ
2)
Main scientific results/projects
3)
Target‐specific achievements
(teaching, organization, etc.)
4)
Integration into the target-team
(Why I’m your best choice)
Lecture:
1) Repetition of the big picture
2) Motivation for the specific topic
3) Prerequisites and basics
4) Different approaches
5) Summary
Presenting a scientific paper
7
A scientific publication (also: article, paper) is a document of usually 4-30 pages
that explains a novel solution to a problem.
Presenting a scientific paper: Context
8
A scientific publication (also: article, paper) is a document of usually 4-30 pages
that explains a novel solution to a problem.
Consider giving a bit of context:
- who are the authors of the paper?
- when and where was it published?
- what type of conference is it?
- what is the impact and significance of the paper?
1) Motivation
9
The motivational part of the talk should make clear
• what is the background
• what is the problem
• why the problem is important
• why the problem is hard
Introduce the domain of interest and the key concepts
What is the given input?
What is the desired output?
This is a key issue!
If the audience does not understand this,
they will not understand anything!
What would be possible
if the problem were solved?
What is the complicating factor?
2) Related work
10
A discussion of related work shows that you know about the domain.
However, the related work is not the focus of your work.
Better focus on the following key questions:
• Don’t we already have an existing solution for this problem?
• Why does a trivial solution not work?
Abstractly name a few methods, say why they do not work.
How would you solve the problem naïvely?
Why does it not work?
Many people in the audience will follow your talk only until here.
So make sure they understood at least the problem and the difficulty.
3) Approach
11
Explain the main approach
•
first the main idea
•
then the approach
Examples:
Anis Harfourche
,
Issa Memari
If you can, show a demo!
The unsurpassed way of presenting an approach
is to simulate it with an example!
The safest way to make sure you understood
what you say is to use your own example!
3) Approach: Technical details
12
Definitions:
• Introduce them only if they are strictly necessary
• Introduce them right before they’re used
• Illustrate them with an example wherever possible
Venn Diagrams are great
for illustrating sets
[Jonathan Lajus]
3) Approach: Technical details
13
Formulas and theorems:
•
Introduce them only if they are strictly necessary
or a main contribution
•
Introduce them only if you can (and will) explain them
•
Give the intuition of the formula/theorem
algorithm
output
ground truth
output that is in ground truth
as proportion of output
4) Experiments
14
The experiments are always the same: some diagram where
your curve is the highest. Be precise, but don’t bore the audience
with details.
Precision
size of training dataset
Their system
Our system
5) Conclusion
15
Also mention the weaknesses
of your approach,
potentially combined with an
outlook about future work,
and end on a positive note:
your achievements.
This will prevent the most awkward questions!
But avoid giving the impression that the
current work is incomplete!
The talk should have made clear two things:
1) What is the problem?
2) What is the solution?
How to give good talks
16
•
Structure
•
Style
•
Slides
•
Speech & Body Language
•
Preparing the talk
•
Giving the talk
•
Being in the audience
A good talk is like a chocolate bar
17
Picture by
Bodo
. No link with the Lion brand. Idea inspired by
Rainer Gemülla
.
The first thing you notice is the packaging.
It is designed to arouse your curiosity.
You need something similar for a talk:
something that arouses curiosity.
Capturing the audience
18
[Thomas Rebele]
In scientific talks, the “packaging” can be
•
an intriguing story (even if fictional)
•
a fun fact
•
a crazy claim
•
a link to the audience
•
a joke (watch out for culture‐dependent sensibilities)
Fast internet access is a legal right in Finland!
We do not need databases actually.
Remember how difficult it was to find an appartment in Paris?
A running example
19
In the ideal case, the capturing item gives rise to a running example:
an example that illustrates the motivation, the difficulties, and the solution to the problem.
Example: the search for appartments can serve as a running example for a talk about
collaborative filtering, Web search, databases, or instance matching.
...or a chocolate bar can
serve as a running example
to discuss scientific talks.
The talk is like the chocolate itself
20
•
keep it joyful and interesting
•
tell a story from the beginning to the end,
let the audience know what to expect,
and then fulfill the expectations
You want the audience to enjoy the talk!
[stevepb]
...but there are nuts in the chocolate
21
Make sure there is also one in‐depth part in your talk!
•
to give the details to those who understand
•
to show that you understood
•
to show that the problem is non‐trivial
A talk without a “nut” will be considered shallow!
[stevepb]
Make sure that your talk has
•
an algorithm
•
an insight
•
facts
•
an argument
Non‐technical papers
22
Avoid the empty package
also for non‐technical papers
such as
surveys, opinion pieces, or papers from the social sciences.
• convey interesting key messages, provide your own synthesis
• juxtapose different approaches, or bring material from other sources
• add your own opinion