Machine Learning

ABDELA AHMED, PhD

© University of Gondar, 2022
[Link]@[Link]
Course outline for ML
Course Description

 This course is about an introduction to ML

 It is a set of topics that will help anyone master the

most important algorithms and concepts in
machine learning – including not only deep learning
but also a lot of other things- and to build effective
learning systems.

3
Objective
 To build computer systems that learn from
experience and that are capable to adapt to their
environments

 Upon completing this course, you should be able

to:
 Explain and differentiate between classical and modern
machine learning techniques
 Identify potential application areas where machine
learning techniques can be useful
 Implement the solution, and evaluate the results

4
Contents: has six main chapters

1. Overview of Machine Learning

 Introduction to AI and ML
 Machine Learning vs Datamining
 Machine Learning vs Statistical learning
 Applications of Machine learning
 Challenges in Machine learning
 Basic Mathematical Concepts for AI and ML

5
Contents…

2. Classic Machine Learning

 Classification and regression
 Basic steps of classification
 Logistic and linear regressions
 K-nearest neighbor (KNN)
 Decision Tree
 Naïve Bias
 Support Vector Machine (SVM)
 Data preprocessing and representations
 Evaluation methods

6
Contents…
3. Deep Learning
 Artificial Neural Network
 Convolutional Neural Network
 Deep Boltzmann Machine
 Deep Belief Network
 Autoencoders
 Recurrent Neural Network
 Long Short Term Memory

7
Contents…
4. Clustering
 Overview of unsupervised learning
 Partition based clustering
 k-means
 K-Medoid
 Hierarchical clustering
 AGNES-agglomerative
 Density-based clustering
 Grid-based clustering
 Model-based clustering
 Evaluation of cluster quality

8
Contents…
5. Reinforcement learning
 Introduction to reinforcement learning
 Bandit problems and Online learning
 Overview of Some popular reinforcement learning
algorithms (Q-learning, Deep Q-learning, Policy gradients,
Actor Critic, and PPO )
 Application of Reinforcement learning
 Challenges of Reinforcement learning

9
Contents
6. Advanced topics
 Semi-supervised learning (S3VMs)
 Generative Adversarial Networks
 Dimensionality réduction techniques (PCA, LDA, FA)
 Handling Imbalanced Datasets
 Handling multimodal Inputs
 Ensemble Learning
 Transfer Learning
 Interpretability of Deep Learning Models

10
 Course Prerequisites
 In order to be successful in this course, you will need a working
knowledge of the following:
 Fundamental understanding of
 Calculus (partial derivatives),
 Linear Algebra (vector/matrix manipulations, properties),
 Basic statistics (probability, common distributions, Bayes Rule, mean,
median, mode and maximum likelihood)
 Familiarity with programming on a Python development
environment and other ML frameworks

11
please install these tools
 Course Prerequisites
 In order to be successful in this course, you will need a working
knowledge of the following:
 Fundamental understanding of
 Calculus (partial derivatives),
 Linear Algebra (vector/matrix manipulations, properties),
 Basic statistics (probability, common distributions; Bayes Rule, mean,
median, mode and maximum likelihood)
 Familiarity with programming on a Python development
environment and other ML frameworks
 Run deep learning models on your local computer and remote
machines such as Google Colab
Numpy
Pandas
Matplotlib
Scikit-Learn

12
please install these tools
 Resources: Books

 Tom M. Mitchell, Machine Learning, McGraw Hill

 Bishop, Pattern Recognition and Machine Learning,
Springer, 2006
 Ethem Alpaydin, Introduction to Machine Learning, The
MIT Press, 2nd Edn., 2010
 Bengio, Y., LeCun, Y., and Hinton, G. Deep Learning.
Nature 521: 436-44, 2015.
 Marc Toussaint Maths for Intelligent Systems, 2017.
 ...

13
Resources: Journals
 Journal of Machine Learning Research [Link]
 Machine Learning
 Neural Computation
 Neural Networks
 IEEE Transactions on Neural Networks
 IEEE Transactions on Pattern Analysis and Machine
Intelligence
 Annals of Statistics
 Journal of the American Statistical Association
 ...
14
Resources: Conferences
 International Conference on Machine Learning (ICML)
 [Link]
 European Conference on Machine Learning (ECML)
 [Link]
 Neural Information Processing Systems (NeurIPS)
 [Link]
 International Conference on Learning Representations (ICLR)
 [Link]
 Conference on Computer Vision and Pattern Recognition (CVPR)
 [Link]
 Uncertainty in Artificial Intelligence (UAI)
 [Link]
 Computational Learning Theory (COLT)
 [Link]
 International Joint Conference on Artificial Intelligence (IJCAI)
 [Link]
 International Conference on Neural Networks (Europe)
 [Link]
 ... 15
Resources: Datasets
 UCI Repository:
[Link]
 UCI KDD Archive:
[Link]
 Statlib: [Link]
 Delve: [Link]

16
Resources: Top ML researchers

17
CHAPTER 1:
Overview of ML
In this overview, we will explain
 artificial Intelligence (AI), Machine Learning (ML)
and Deep Learning (DL)
 how DL helps solve classical ML limitations
 types of Machine learning algorithms
 work flows of Machine Learning
 application areas of Machine Learning
 challenges of Machine Learning

19
What do you know about Machine learning?
 How can we solve a specific problem?
 As computer scientists, we write a program that encodes
a set of rules that are useful to solve the problem

Figure: How can we make a robot cook? 20

 AI breakthroughs: what do you know about ML?

 How can we solve a specific problem?

 As computer scientists, we write a program that encodes
a set of rules that are useful to solve the problem
 In many cases, it is very difficult to specify those rules.
e.g., given a picture determine whether there is a cat in the
image

As of 2015, computers can be trained to

perform on this tasks than humans using the
latest ML techniques

As of 2016, computers have achieved near-

human performance for machine translation21
 AI breakthroughs: what do you know about ML?

 How can we solve a specific problem?

 As computer scientists, we write a program that encodes
a set of rules that are useful to solve the problem
 In many cases, it is very difficult to specify those rules.
e.g., given a picture determine whether there is a cat in the
image

As of 2015, computers can be trained to

perform
“About on this
100 year ago,tasks than humans
electricity usingevery
transformed the
major [Link]
AI hasML techniques
advanced to the point where it
has the power to transform… every major sector in
coming years “ -Andrew NG, Stanford University
As of 2016, computers have achieved near-
[Link]
human performance for machine translation22
Definitions
 How can we solve a specific problem?
 As computer scientists, we write a program that encodes a set
of rules that are useful to solve the problem
 In many cases it is very difficult to specify those rules, e.g.,
given a picture determine whether there is a cat in the image
 The real question is what is learning?
 Using past experience to improve future performance
 For a machine, experiences come in the form of data
 Learning simply means incorporating information from
the training examples into the system
 Learning systems are not directly programmed to solve
a problem, instead develop own program based on
 Examples of how they should behave
 From trial-and-error experience trying to solve the problem

23
What do you know about Machine Learning
 Machine learning is the study and construction of
programs that are not explicitly programmed, but learn
patterns as they are exposed to more data over time.

 Instead of writing a program by hand, we collect lots of

examples that specify the correct output for the given
input.

 A machine learning algorithm then takes these

examples and produces a program that does the job.
 The program produced by the learning algorithm may look
very different from a typical hand-written program. It may
contain millions of numbers.
 If we do it right, the program works for new cases as well as
the one we trained it on.
24
Definitions
 AI : the branch of computer science dealing
with the simulation of intelligent behavior
in computers.
 Any program that can sense, reason, act, and
adapt.
 A rule-based system where it doesn’t
learn as more data comes in.

 ML : the study of programs that are not

explicitly programmed, but instead these
algorithms learn patterns from data.
 Algorithms whose performance improve as they
are exposed to more data over time.

 DL: subset of ML, that uses the multilayered

neural networks to analyze different
patterns from vast amount of data with a
structure that is similar to the human
25
neural system.
 What is ML: summary
 Arthur Samuel (1959) defined machine learning as “a
sub-field of computer science that gives computers
the ability to learn without being explicitly
programmed.”

 It means that ML is able to perform a specified task

without being directly told how to do it.
 These programs learn from repeatedly seeing data,
rather than being explicitly programmed by humans.

 Example:
 Decide whether emails are spam or not spam.

26
 What is ML: summary
 Example:
 Decide whether emails are spam or not spam.
 We would start off with dataset where we have a bunch emails that
are going to be labeled spam vs not spam.
 These emails will be preprocessed and fed through a ML algorithm
that learns the patterns for spam vs not spam
 Once the ML algorithm is trained we can use it to predict as new
emails are coming in
 So we trained on this label dataset, and now we can run in
production as new emails come in we predict spam vs not spam

27
 ML terminoloygy
 In this example, we learn to classify wine quality from a set of
measurement features and wine type

Fixed Volatile Citric Density PH alcohol type Quality

acidity acidity acid
7.4 1.9 0 0.9978 3.51 9.4 red Poor
8.5 0.28 0.56 0.9969 3.3 10.5 red Excellent
6.7 0.24 0.3 0.9919 3.04 11.3 white Excellent
8.1 0.27 0.41 0.9908 2.99 12 white Poor
7.3 0.65 0 0.9946 3.39 10 red Excellent
7 0.27 0.36 1.001 3 8.8 white Excellent
7.8 0.88 0 0.9968 3.2 9.8 red poor

28
 ML terminoloygy
 In this example, we learn to classify wine quality from a set of
measurement features and wine type

Fixed Volatile Citric Density PH alcohol type Quality

acidity acidity acid
7.4 1.9 0 0.9978 3.51 9.4 red Poor
8.5 0.28 0.56 0.9969 3.3 10.5 red Excellent
6.7 0.24 0.3 0.9919 3.04 11.3 white Excellent
8.1 0.27 0.41 0.9908 2.99 12 white Poor
7.3 0.65 0 0.9946 3.39 10 red Excellent
7 0.27 0.36 1.001 3 8.8 white Excellent
7.8 0.88 0 0.9968 3.2 9.8 red poor

29
 ML terminology
Features (explanatory variables)
 attributes of the data used for prediction
 Number of features to predict target variable?

Fixed Volatile Citric Density PH alcohol type Quality

acidity acidity acid
7.4 1.9 0 0.9978 3.51 9.4 red Poor
8.5 0.28 0.56 0.9969 3.3 10.5 red Excellent
6.7 0.24 0.3 0.9919 3.1 11.3 white Excellent Target
8.1 0.27 0.41 0.9908 2.9 12 white Poor category or
value that we
7.3 0.65 0 0.9946 3.4 10 red Excellent are trying to
predict
7 0.27 0.36 1.001 3 8.8 white Excellent
7.8 0.88 0 0.9968 3.2 9.8 red poor

Example (Observation)
 A single data point within the data (one raw)
 Number of examples? 30
 ML terminology
Features (explanatory variables)
 attributes of the data used for prediction
 Number of features to predict target variable?

Fixed Volatile Citric Density PH alcohol type Quality

acidity acidity acid
7.4 1.9 0 0.9978 3.51 9.4 red Poor
8.5 0.28 0.56 0.9969 3.3 10.5 red Excellent
6.7 0.24 0.3 0.9919 3.1 11.3 white Excellent Target
category or
8.1 0.27 0.41 0.9908 2.9 12 white Poor
value that we
7.3 0.65 0 0.9946 3.4 10 red Excellent are trying to
predict
7 0.27 0.36 1.001 3 8.8 white Excellent
7.8 0.88 0 0.9968 3.2 9.8 red poor

Example (Observation)
 A single data point within the data (one raw)
 Number of examples? 31
 ML terminology
Features (explanatory variables)
 attributes of the data used for prediction
 Number of features to predict target variable?

Fixed Volatile Citric Density PH alcohol type Quality

acidity acidity acid
7.4 1.9 0 0.9978 3.51 9.4 red Poor
8.5 0.28 0.56 0.9969 3.3 10.5 red Excellent
6.7 0.24 0.3 0.9919 3.1 11.3 white Excellent Target
category or
8.1 0.27 0.41 0.9908 2.9 12 white Poor
value that we
7.3 0.65 0 0.9946 3.4 10 red Excellent are trying to
predict
7 0.27 0.36 1.001 3 8.8 white Excellent
7.8 0.88 0 0.9968 3.2 9.8 red poor

Label
Example (Observation) The target
value for a
 A single data point within the data (one raw) single data
 Number of examples? point 32
 What is ML: summary
 A widely accepted formal definition by Tom Mitchell
(1997, professor of Carnegie Mellon University):
 A computer program is said to learn from experience E with
respect to some class of tasks T and performance measure P,
if its performance at the tasks T , as measured by P,
improves with the experiences.

 According to this definition, we can reformulate the

previous email classification problem as
 the task of identifying spam messages (task T) using the data
of previously labeled email messages (experience E) through a
machine learning algorithm with the goal of improving the
future email spam labeling (performance measure P)

33
 Types of ML

 Supervised Learning
 Unsupervised Learning
 Semi-supervised Learning
 Reinforcement Learning

34
 Types of ML: Supervised vs Unsupervised

Dataset Goal Example

Supervised Has a target Make Fraud

Learning column predictions detection

Doesn’t have Customer

Unsupervised Find structure
a target segmentation
Learning in the data
column

35
Supervised Learning:
classification vs regression

36
 Unsupervised Learning

Clustering

Dimensionality reduction
37
 Classic ML example: fraud detection
 Suppose you wanted to identify fraudulent credit
card transaction
 detecting fraud is a common Machine Learning problem
 You can define your features to be
 transaction time,
 transaction amounts,
 transaction location,
 category of purchase.
 The algorithm could learn what feature combinations
suggest unusual activity
 This structured data with intuitive features are going to be a
good task for our traditional Machine Learning

38
 Classic ML limitations
 Suppose you wanted to determine if an image is
of a cat or a dog.
 What features would you use?
 For images, the data is taken as numerical
data to reference the coloring of each
individual pixel within our image
 So a pixel then could be used as a feature
 But if you imagine even a small image will have
256 by 256 pixels, which will come out to over
65,000 pixels.
 Sixty five thousand pixels mean 65,000 features
which is a huge amount of features to be working
with
 Another issue is that using each pixel as an
individual, you lose the spatial relationship to the
pixels around it
 In other words, the information of a pixel makes 39
sense relative to its surrounding pixels
 Classic ML limitations
 Suppose you wanted to determine if an image is of
a cat or a dog.
 What features would you use?
 This where deep learning can come in
 Deep Learning techniques will give you the
capability to learn these features on its own and
combine these pixels to define these spatial
relationships
 Deep learning is a ML that involves using very
complicated models called “deep neural networks”
 DL models determine best representation of
original data; in classic ML, humans must do this.

40
 DL vs Classic ML with Example

Feature
Classic Detection
Machine
Learning Step 1:
Determine
features

41
 DL vs Classic ML with Example

ML
Feature classifier John
Classic Detection algorithm
Machine
Learning Step 1: Step 2:
Determine Feed them
features through
model

Deep Learning
(Steps 1 and 2
are combine into
1 step)

42
Tasks that require Machine Learning:
What makes a 2?

43
Tasks that benefit from machine learning:
cooking

44
 Learning – a two step process

 Model construction
 A training set is used to create the model.
 The model is represented as classification rules, decision
trees, or mathematical formula

 Model usage
 the test set is used to see how well it works for classifying
future or unknown objects

45
 Learning – a two step process
 Model construction
 A training set is used to create the model.
 The model is represented as classification rules, decision
trees, or mathematical formula
Classification
Algorithms
Training
Data
Classifier
NAME RANK YEARS TENURED (Model)
M ike A ssistan t P ro f 3 no
M ary A ssistan t P ro f 7 yes
B ill P ro fesso r 2 yes IF rank = ‘professor’
Jim A sso ciate P ro f 7 yes
D ave A ssistan t P ro f 6 no
OR years > 6
Anne A sso ciate P ro f 3 no THEN tenured = ‘yes’ 46
 Learning – a two step process
 Model usage
 the test set is used to see how well it works for classifying
future or unknown objects

Classifier
model
Testing
Data Unseen Data

(Jeff, Professor, 4)
NAME RANK YEARS TENURED
Tom A ssistan t P ro f 2 no Tenured?
M erlisa A sso ciate P ro f 7 no
G eo rg e P ro fesso r 5 yes 47
Jo sep h A ssistan t P ro f 7 yes
Challenges in Machine Learning
 Data
 Getting huge data
 Unclear data acquisition and representation issues
 Protection of security, integrity, and privacy
 Handling high-dimensionality
 Handling noise, incomplete and imbalanced data

 Computational resources
 CPU, GPU, Cloud

 Algorithms
 Selection of algorithms
 Efficiency and scalability of machine learning algorithms
 Degree of interpretability 48
 Basic steps in Machine Learning
1. Problem statement
 What problem are you trying to solve
2. Data collection
 What data do you need to solve it?
3. Data pre-processing
 How should you clean data so you model can use it?
4. Feature engineering
 Select representatives features to improve performance
5. Modeling
 Build a model to solve your problem
6. Validation
 Did I solve the problem
7. Deployment
 Put it into production

49
Applications
 Association
 Supervised Learning
 Classification
 Regression
 Unsupervised Learning
 Reinforcement Learning

50
Learning Associations
 Basket analysis:
P (Y | X ) probability that somebody who buys X also
buys Y where X and Y are products/services.

Example: P ( chips | beer ) = 0.7

51
Classification
 Example: Credit
scoring
 Differentiating
between low-risk
and high-risk
customers from
their income and
savings

Discriminant: IF income > θ1 AND savings > θ2

THEN low-risk ELSE high-risk

52
Applications
 Transportation: autonomous cars, automated
tracking, Shipping, search and rescue
 Communication: language translation
 Healthcare: Enhancing diagnosis, Drug discovery
 Industry: Factory automation, precision agriculture
 Finance: Fraud detection, algorithmic trading,
 Energy : Oil and Gas exploration, conservation
 Government: Defense, safety and security, smarter
cities,
 and More...

53
Unsupervised Learning
 Learning “what normally happens”
 No output
 Clustering: Grouping similar instances
 Example applications
 Customer segmentation in CRM
 Image compression: Color quantization
 Bioinformatics: Learning motifs

54
Reinforcement Learning
 Learning a policy: A sequence of outputs
 No supervised output but delayed reward
 Credit assignment problem
 Game playing
 Robot in a maze
 Multiple agents, partial observability, ...

55
 Project : guideline
 Competition between ML algorithms.
 You will be given some data for training a ML system, and you
will try to develop the best method
 You should prepare slides for a 10 minute presentation, with 5
minutes for the question and answer session
 The final report of your project should consists of Abstract,
Introduction, Statement of the problem, Research Questions,
Objectives, Related Works, Results and Discussions,
Conclusion and Future works, and References.
 Final report submission and Presentation of you work is 1 week
before class end
 Projects which have well-designed experiments and a thorough
analysis of the results will score higher.
 The writing style and the clarity of the report is an asset to score
good grade
56
 Project: list of potential topics-
category 1

1. Drug Prediction System

2. Prediction of Heart Disease Presence
3. Prediction of Cervical Cancer
4. Water quality Prediction

57
 Project: list of potential topics-
category 2
1. Breast Cancer Classification from mammography
images
2. Brain tumor classification from MRI images
3. Brain tumor classification from Histology images
4. Covid-19 detection from X-ray images
5. Tuberculosis classification from CT images

58
 Assignment: topics
1. Support Vector Machine
2. Naïve Bayes and Random Forest
3. Deep Boltzmann Machine
4. Deep Belief Network
5. Deep Learning with Autoencoder
6. Density-based Clustering (DBSCAN)
7. Grid-based Clustering (STING)
8. Deep Reinforcement Learning
9. Semi-supervised Learning

59
 Provisional Calendar
 Week-1: June 6-12
 Chapter-1: Overview of ML
 Chapter-2: Regression algorithms
 Week-2: June 13-19
 Chapter-2: KNN, DT, Evaluation metrics
 Week-3: June 20-26
 Chapter-3: NVB, SVM, ANN,CNN
 Week-4 : June 27- July 3
 Chapter-3: RNN, LSTM, DBM, DBN, Autoencoder
 Week-5: July 4 – 10
 Chapter-4: K-means, K-medoids, AGNES, Density-based, Grid-
based
 WEEK-6: July 11-17
 Chapter-5: Reinforcement learning
 Chpater-6: Dimensionality reduction techniques, Ensemble
learning, Transfer learning, Interpretability, Handling Imbalanced
and multimodal inputs, Semi-supervised learning
 Week 7 : July 18-24
 Revision
 Week 8 : July 25 - August 6
 Project and Exam preparation 60
 Summary
 Machine Learning
 Study of algorithms that
 improve their performance
 at some task
 with experience

 Factors that have contributed to the current state

of Machine Learning:
 bigger data sets,
 faster computers,
 open source packages, and
 a wide range of neural network architectures.

61
Thank You

62