COMPUTATIONAL

INTELLIGENCE
(INTRODUCTION TO MACHINE LEARNING)

SS 18

2 VO 442.070 + 1 UE 708.070

Institute for Theoretical Computer Science (IGI)

TU Graz, Inffeldgasse 16b / first floor
www.igi.tugraz.at

Institute for Signal Processing and Speech Communication (SPSC)

TU Graz, Inffeldgasse 16c / ground floor
www.spsc.tugraz.at
Organization
• Lecture / VO:
• Tuesday, 11:00, HS i13
• Anand Subramoney and Guillaume Bellec (IGI) Part I
• Assoc. Prof. Dr. Franz Pernkopf (SPSC)
Part II
• Practical / UE:
• First practical on Friday 9th of March, HS i11
• 12:30-13:30 – if your last name starts with A-L
• 14:00-15:00 – if your last name starts with M-Z

• Anand Subramoney and Guillaume Bellec (IGI) Part I

• Dipl.-Ing. Christian Knoll (SPSC) Part II
• Homework in teams of up to 3 (use newsgroup to form teams)

• Website: http://www.spsc.tugraz.at/courses/computational-intelligence
• Newsgroup: tu-graz.lv.ci
Organization
• Lecture / VO:
• Class cancelled on the 13th of May

• Practical / UE:
• Class cancelled on the 16th of May
Organization
• Office hours:
• Both Anand and Guillaume:
• Time: Every Tuesday 14:00 – 15:00
• Place: Our offices at Inffeldgasse 16b/1

• Exam:
• Written exam for this year’s course:
• July onwards
• Exam has two parts:
• IGI (first half of semester) + SPSC (second half)
• Language: English
• Positive grade if positive on both parts!
Materials (for IGI part)
• No textbook required
• Lecture slides and further reading on tech center

• Materials for further study:

• Online Machine Learning course
coursera www.coursera.org/course/ml
udacity de.udacity.com/course/intro-to-machine-learning--ud120

• Book by C. Bishop, Pattern Recognition and

Machine Learning, Springer 2007.

• For SPSC part (second half):

• Announced by Franz Pernkopf
Acknowledgments
• IGI Slides based on material from Stefan Häusler (IGI), Zeno Jonke
(IGI), David Sontag (NYU), Andrew Ng (Stanford), Xiaoli Fern
(Oregon State)
INTRODUCTION +
MOTIVATION
Machine Learning
• Grew out of Artificial Intelligence
What is Artificial Intelligence?

Source -- Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig
But what really is AI?

Turing test
Turing test

AI – “You’ll know it when you see it”

Components of AI
• Natural language processing
• Knowledge representation
• Automated reasoning
• Machine learning
• Computer vision
• Robotics

-- Russel and Norvig

Machine Learning
• Grew out of Artificial Intelligence

• The ability to “adapt to new circumstances and to detect

and extrapolate patterns” – Russel and Norvig

• Arthur Samuel (1959). “Field of study that gives

computers the ability to learn without being explicitly
programmed.”
When do we need computers to learn?
• When human expert knowledge is missing
• For example, predicting whether some new substance could be an effective
treatment for a disease

• When humans can only do it “intuitively”

• Flying a helicopter
• Recognize visual objects
• Natural language processing

• When we need to learn about something that changes frequently

• Stock market analysis
• Weather forecasting
• Computer network routing

• Customized learning
• Spam filters, movie/product recommendations
Applications of Machine learning
• Machine learning is used in a wide range of fields including:
• Bio-informatics
• Brain-Machine interfaces
• Computational finance
• Game playing
• Information Retrieval
• Internet fraud detection
• Medical diagnosis
• Natural language processing
• Online advertising
• Recommender systems
• Robot locomotion
• Search engines
• Sentiment analysis
• Software engineering
• Speech and handwriting recognition
• Stock market analysis
• Economics and Finance
• Credit card fraud detection
• …….
Autonomous car
• Waymo/Alphabet
• https://www.youtube.com/watch?v=TsaES--OTzM

+ UK, France,
Switzerland,
Singapore
 Bipedal robot
• ATLAS (Boston Dynamics/Alphabet)
• https://www.youtube.com/watch?v=fRj34o4hN4I (three month ago)
• https://www.youtube.com/watch?v=aFuA50H9uek (last week)

http://spectrum.ieee.org/automaton/robotics/humanoids/boston-dynamics-marc-raibert-on-nextgen-atlas
 https://blog.openai.com/openai-baselines-ppo/
AI for robotics
• OpenAI (2016)
robots cannot only
learn from accelerated
simulated
environments
Web search
Image search
• Google image search
https://images.google.com
Face recognition
• Facebook
• http://www.youtube.com/watch?v=l4Rn38_vrLQ
• iPhoto
• Cameras, etc.

• Microsoft cognitive services

• From face, can recognize age,
gender, emotions!
• https://www.microsoft.com/cognitiv
e-services/
Scene and text recognition
• Microsoft Seeing AI project
• https://www.youtube.com/watch?v=R2mC-NUAmMk
Machine Translation
• Skype and PowerPoint real-time translation (Microsoft)
https://www.youtube.com/watch?v=rek3jjbYRLo
https://www.youtube.com/watch?v=u4cJoX-DoiY
Learning to reason

• Human level performance at video games from ATARI

2600 (Google Deep mind 2015)
• Beating world champion of GO (G. Deepmind 2016)
• Beating champion chest program (G. Deep mind 2017)
Brain Computer Interface
• “Neural Dust” tiny neural implants from Berkley (2016)
• (not much AI in BCI for now but it’s coming)
• https://www.youtube.com/watch?v=oO0zy30n_jQ
CLASSICAL PROBLEMS
AND APPLICATIONS
Recommender systems
Spam filtering
• "Spam in email started to become a problem when the Internet was
opened up to the general public in the mid-1990s. It grew
exponentially over the following years, and today composes some 80
to 85% of all the email in the world, by a "conservative estimate".
• Source: http://en.wikipedia.org/wiki/Spamming

data prediction

Spam
vs.
Not Spam
Data visualization (Embedding images)

• Images have
thousands or
millions of pixels.

• Can we give each

image a coordinate,
such that similar
images are near
each other?

[Saul & Roweis ‘03]

Clustering
• Clustering data into similar groups

http://scikit-learn.org/stable/auto_examples/applications/plot_stock_market.html
Clustering images

Set of images
Growth of Machine Learning
• Preferred approach to
• Speech recognition, Natural language processing
• Computer vision
• Robot control
• Computational biology
• ….
• Accelerating trend
• Big data (data mining)
• Improved algorithms
• Faster computers
• Availability of good open-source software and datasets
Some of the future challenges
• The scientific challenges
• Learning from fewer data (1-shot learning)
• Generalization
• Energy efficient hardware and algorithms
• Understanding animal intelligence

• Ethical issues of AI
• Privacy
• Intelligent weapons
• Replacing artisans with robots
COURSE CONTENT
What we will cover
• IGI part: • SPSC part:
• Introduction • Parametric & non-parametric
• Linear regression density estimation
• Non-linear basis functions • Bayes classifier
• Logistic regression • Gaussian mixture model
• Under- and over-fitting • K-means
• Model selection • Markov model & Hidden Markov
• k-NN
model
• Graphical models
• Cross-validation
• PCA
• Regularization
• LDA
• Neural networks
• SVM
• Kernel methods
• Multiclass classification
INTRODUCTION:
TYPES OF ML ALGORITHMS
Types of Machine Learning algorithms
• Supervised learning
• Given: Training examples with target values
• Goal: Predict target values for new examples
• Examples: optical character recognition, speech recognition, etc.

Learning from
• Unsupervised learning examples (data)
• Given: Training examples without target values
• Goal: Detect and extract structure from data
• Examples: clustering, segmentation, embedding (visualization), compression,
automatic speaker separation

• Reinforcement learning (not in this course)

• Given: Feedback (reward/cost) during
trial-and-error episodes Learning by doing
• Goal: Maximize Reward/minimize cost (trial and error)
• Examples: learning to control a robot/car/helicopter etc.
• see Master‘s course “Autonomously Learning Systems”
Supervised Learning: Example
• Learn to predict output from input (learning from examples)
• Target values (output) can be continuous (regression) or discrete
(classification)
• E.g. predict the risk level (high vs. low) of a loan applicant based on
income and savings

Applications:

• Spam filters
• Character recognition
• Speech recognition
• Collaborative filtering (predicting
if a customer will be interested in
an advertisement …)
• Medical diagnosis
• …
Unsupervised Learning: Example
• 90% of collected data is unlabeled
• Ex. Find patterns and structure in data

Clustering art
Unsupervised Learning: Applications
• Market partition: divide a market into distinct subsets of
customers
• Find clusters of similar customers, where each cluster may
conceivably be selected as a market target to be reached with a
distinct marketing strategy
• Data representation: Image, document, web clustering
• Automatic organization of pictures
• Generate a categorized view of a collection of documents
• For organizing search results etc.
• Bioinformatics
• Clustering the genes based on their expression profile
• Find clusters of similarly regulated genes – functional groups
INTRODUCTION:
SUPERVISED LEARNING
Regression and classification
Simple regression example
Top 50 movies (top first weekends)
Avengers
700
The Dark Knight
Million USD

600
Total income

500

400

300

200

100 X-Men Origins:

Wolverine
0
0 50 100 150 200 250
Million USD

First weekend income Data source: http://www.boxofficemojo.com

„The Hunger Games: Catching Fire“: 158 Mio. USD on opening weekend.
How much in total?
Predicted: ~418 Mio., Actual: 424 Mio.
Simple regression example (cont’d)
700
600
• Data set: Input , Output 500
400

y
300
First Total 200
weekend
i 100
0
0 100 200 300
x
Avengers 1 207 623

Iron Man 3 2 174 409

Harry Potter
and the 3 169 381
Deathly… m data points
The Dark Knight
Rises 4 161 449 (data samples)

The Dark Knight 5 158 533

… …
 Simple regression example (cont’d)
700
600
• Data set: … 500
400

y
300
200
Training set 100
0
0 100 200 300
x

Hypothesis
Learning algorithm

Parameters
Test input
„Hypothesis“
x h
Prediction
Non-linear regression
10
training data
y 0 regression

-10

-20

-30

-40

-50
0 20 40 60 80 100

x x
Training set
Non-linear hypothesis, for example
Learning
algorithm

Test input
„Hypothesis“
x h
Prediction
 Regression with multiple inputs
2
2
1.5
1.5 1
y 1 y 0.5

0.5 0

0 -0.5

-0.5 -1

-1.5
-1
10
-1.5
10 0

6 8
2 4
0 -10 -2 0
-6 -4

6 8
2 4
-10 -2 0
-6 -4

linear hypothesis non-linear hypothesis

 Multiple inputs continued
2

y 1.5

0.5

-0.5

-1

-1.5
10
Training set
0

6 8
2 4
-2 0
-10 -4
-6

Learning algorithm
i 1 2
Test input
1 5.3 -2.1 2.31

2 0.4 3.5 -1.3 „Hypothesis“ h Prediction

3 1.2 0.9 1.9

4 -0.3 0.1 -0.7

5 … … …
Simple classification example
„labeled data“
decision boundary
Tumor size Malignant?
(mm) benign
i malignant
x y

1 2.3 0 (N)
Tumor size (x)
2 5.1 1 (Y)

1
0
3 1.4 0 (N)

4 6.3 1 (Y)

5 5.3 1 (Y)
Example hypothesis:
… …
1 if x >
labels
 Classification with multiple inputs
linear decision boundary
Tumor Age Malign
size ant?
i (mm)

x1 x2 y

1 2.3 25 0 (N)

Age (x2)
2 5.1 62 1 (Y)

3 1.4 47 0 (N)

4 6.3 39 1 (Y)

5 5.3 72 1 (Y)
Tumor size (x1)
… … benign
malignant
 Non-linear classification
linear decision boundary non-linear decision boundary
Age (x2)

Age (x2)

Tumor size (x1) Tumor size (x1)

Both hypotheses fit the data quite well.
Which one fits the training data better?
Which one would you trust more for prediction?
 Supervised learning (Regr., Class.)
• Discrete vs. continuous outputs
(classification vs. regression) Training set

In the next few classes we‘ll cover:

Learning algorithm
• Learning algorithms for
regression and classification
Test input
(linear regression, neural nets, „Hypothesis“ h Prediction
SVMs, etc.)

• Supervised learning in practice

(overfitting, etc.)
How to extend to images or sound ?
• Find the best way to represent the data as vectors
(i.e. tables of numbers)

• Light intensity of each pixel for images, time-wise amplitude of

air pressure for sounds

• Knowing the data structure helps to design better

representations. When the data is compressed into a lower
dimensions recognition is made easier.
What is next?
• Linear regression

• Gradient descent

• Non-linear basis functions

Supervised, unsupervised or
Reinforcement Learning ?
Regression or classification ?