Machine Learning
What it is and why you should care
Dr. Christian Wiele
@christian_wiele
©2017 gezeitenraum
What is Machine Learning?
2
©2017 gezeitenraum
Machine Learning is the
predominant area of
artificial Intelligence
(often AI and ML are used interchangeably)
3
©2017 gezeitenraum
Machine Learning
=
autonomous pattern
identification and detection
4
©2017 gezeitenraum
Will we soon be dominated by
super-human machines?
5
©2017 gezeitenraum
There are absolutely no
indications of a nearby
singularity!
6
©2017 gezeitenraum
Why should you care about
Machine Learning?
7
Is it more than a hype?
©2017 gezeitenraum 8
ML gets the Value out of Data
©2017 gezeitenraum
Computer vision
➡e.g. autonomous driving, cancer detection
Speech recognition
➡e.g. Siri
Natural language processing (NLP)
➡machine translation, chat bots
General patterns
➡click stream analysis, fraud detection, security, recommender
systems
9
AI might be over in EU before it starts
©2017 gezeitenraum 10
http://www.techzone360.com/topics/techzone/articles/2017/01/25/429101-eus-right-explanation-harmful-restriction-artificial-intelligence.htm
Skill Levels of Machine Learning
©2017 gezeitenraum 11
apply ML as
black box understand research
©2017 gezeitenraum
So, What is Machine learning
about?
12
Core Challenge of Machine learning
©2017 gezeitenraum 13
What
humans are
good at
What
Machines
are good at
(73492.232 +
2049.3827) *
883792.45
Learning approaches
©2017 gezeitenraum 14
supervised learning unsupervised learning
BananA
Grape
Banana
X
X
X
X
X
X
X
X
X
X
X
X X
X
X
X X
X
X
X
X X
XX
X
X
X
AIpoly
©2017 gezeitenraum 15
©2017 gezeitenraum
How it works
16
Classical programming paradigm
©2017 gezeitenraum 17
If (pixel1 == … And Pixel2 == …)
Then …
Else IF …
Goal: minimize the error
©2017 gezeitenraum 18
error
{Parameter 1
Parameter 2
The fruit fly of ML: MNIST
©2017 gezeitenraum 19
55.000 handwritten digits 28x28 pixels
Neural Net: digit recognition
©2017 gezeitenraum 20
Input
1
2
3
4
5
6
7
8
9
0
output neurons
2
3
4
5
6
7
8
9
10
1
12
2
3
17
18
1
25
24
………
…
…
hidden neurons
output
Artificial Neuron: Internal view
©2017 gezeitenraum 21
logic:
result = one number
Output:
1 = activate
0 = not activate
Activation
©2017 gezeitenraum 22
hidden neuron 12:
red dominantes —> fire
input mask / weights weighted input
hidden neuron 13:
blue dominantes —> don’t fire
Neural Net: digit recognition
©2017 gezeitenraum 23
Input
Output-Neuronen
1
2
3
4
5
6
7
8
9
0
2
3
4
5
6
7
8
9
10
1
…
12
2
3
17
18
1
25
24
………
…
Innere-Neuronen
Output
What are relevant pixel structures?
©2017 gezeitenraum 24
neuron 12 neuron 18 neuron 25
©2017 gezeitenraum
So, how does learning work?
25
learning process
©2017 gezeitenraum 26
Neural Net: Learning Process
©2017 gezeitenraum 27
Input
Output-Neuronen
2
3
4
5
6
7
8
9
10
1
12
2
3
17
18
1
25
24
………
…
…
Innere-Neuronen
Output
1
2
3
4
5
6
7
8
9
0
error
error
error
slightly
change
parameters
©2017 gezeitenraum
How to get stared with
Machine Learning?
28
Skill Levels of Machine Learning
©2017 gezeitenraum 29
apply ML as
black box understand research
©2017 gezeitenraum 30
https://www.coursera.org/learn/machine-learning/
fast.ai
©2017 gezeitenraum 31
http://fast.ai
Toolset
©2017 gezeitenraum
Python
➡Theano, Tensorflow, Keras, NumPy, …
➡Jupyter notebook
Linear algebra
Access to GPU
➡AWS, Google, or your own …
Time and motivation for experimenting
32
The machine learning expert
©2017 gezeitenraum 33
Geschafft
©2016 gezeitenraum 34
Vielen Dank!
Dr. Christian Wiele
christian@gezeitenraum.com
@christian_wiele
@gezeitenraum

Machine Learning - What it is and why you should care