MLT unit 1- Introduction To Machine Learning And types Of ML , Cross Validation

Machine Learning Techniques Unit 1

1. CHAPTER 1:
Introduction

2. 1.Unit I:
1. Introduction
2.What is Machine learning,
3. why machine learning,
4. History of Machine Learning,
5.Fundamentals of Machine Learning,
6. Data needed,
7.ML Techniques overview,
8.Validation Techniques (Cross-Validations),
9.Supervised Learning,
10.unsupervised learning,
11.reinforcement learning

3. Learning
Learning is any process by which system
improves performance from experience-
Herbert A. Simon
Alpydin & Ch. Eick: ML Topic1 3

4. Why “Learn”?
Learning is used when:
Human expertise does not exist (navigating on
Mars),
Humans are unable to explain their expertise
(speech recognition)
Solution changes in time (routing on a computer
network)
Solution needs to be adapted to particular cases
(user biometrics)
4

5. What We Talk About When We
Talk About“Learning”
Learning general models from a data of particular
examples
Datais cheapand abundant (data warehouses, data
marts);
knowledgeis expensiveand scarce.
Example in retail: Customer transactions to
consumer behavior:
People who bought “Bread” also bought “Butter”
(www.amazon.com)
Build a model that is a good and useful
approximationto the data.
5

6. History
6

7. Alpydin & Ch. Eick: ML Topic1 7

8. Alpydin & Ch. Eick: ML Topic1 8

9. Machine Learning
It is something we encounter on a daily basis.
In its simplest form,
machine learning is a set of algorithms learned
from data and/or experiences, rather than being
explicitly programmed.
The goalof AI is to imitate and mimic human
behavior, and machine learning gives us the
mathematical toolsthat allow us to do that.
9

10. What is Machine Learning?
“Learning is any process by which a system
improves performance from experience
Machine Learning
Study of algorithms that
improve their performance
at some task
with experience
Optimize a performance criterion using example data
or past experience.
10

11. Alpydin & Ch. Eick: ML Topic1 11

12. When Do We Use Machine Learning?
ML is used when:
• Human expertisedoes not exist (navigating on Mars)
• Humans can’t explain their expertise (speech recognition)
• Models must be customized(personalized medicine)
• Models are based on huge amounts of data (genomics)
Alpydin & Ch. Eick: ML Topic1 12

13. Growth of Machine Learning
Machine learning is preferred approach to
Speech recognition, Natural language processing
Computer vision
Medical outcomes analysis
Robot control
This trend is accelerating
Improved machine learning algorithms
Improved data capture, networking, faster computers
Software too complex to write by hand
New sensors / IO devices
Demand for self-customization to user, environment
Alpydin & Ch. Eick: ML Topic1 13

14. Applications of Machine
Learning
Alpydin & Ch. Eick: ML Topic1 14

15. Machine learning finds applications across a wide range
of domains, including:
Finance: Predicting stock prices, credit risk assessment,
fraud detection.
Healthcare: Disease diagnosis, medical image analysis,
drug discovery.
Marketing: Customer segmentation, recommendation
systems, personalized advertising.
Natural Language Processing (NLP): Sentiment analysis,
language translation, chatbots.
Computer Vision: Object detection, facial recognition,
autonomous vehicles.
Gaming: Character behavior modeling, opponent AI,
procedural content generation.
Manufacturing: Quality control, predictive maintenance,
supply chain optimization.
Alpydin & Ch. Eick: ML Topic1 15

16. Types of Machine Learning
Alpydin & Ch. Eick: ML Topic1 16

17. Machine learning can be broadly categorized into three
main types:
Supervised Learning: In supervised learning, the
algorithm learns from labeled examples in the form of
input-output pairs. It aims to map inputs to correct
outputs by identifying patterns in the data. Common
algorithms include linear regression (for regression
tasks) and classification algorithms like decision trees,
support vector machines, and neural networks.
Unsupervised Learning: Unsupervised learning involves
learning patterns and structures from unlabeled data.
Clustering and dimensionality reduction are typical
tasks in unsupervised learning. K-means clustering and
principal component analysis (PCA) are popular
unsupervised algorithms.
Alpydin & Ch. Eick: ML Topic1 17

18. Reinforcement Learning: Reinforcement learning is
inspired by behavioral psychology, where an agent
learns how to interact with an environment to maximize
rewards. It involves taking actions in an environment to
achieve a specific goal. Reinforcement learning is
widely used in robotics, game playing, and autonomous
systems.
Alpydin & Ch. Eick: ML Topic1 18

19. Alpydin & Ch. Eick: ML Topic1 19

20. Alpydin & Ch. Eick: ML Topic1 20

21. Supervised-Types
Alpydin & Ch. Eick: ML Topic1 21

22. Alpydin & Ch. Eick: ML Topic1 22

23. Unsupervised -Types
23

24. Alpydin & Ch. Eick: ML Topic1 24

25. Alpydin & Ch. Eick: ML Topic1 25

26. Learning Associations
Basket analysis:
P (Y | X ) probability that somebody who buys Xalso buys Y where Xand Y
are products/services.
Example: P ( chips |coke) = 0.7
Market-Basket transactionsTID Items
1 Bread, Milk
2 Bread, Diaper, Beer, Eggs
3 Milk, Diaper, Beer, Coke
4 Bread, Milk, Diaper, Beer
5 Bread, Milk, Diaper, Coke

27. Classification
27
Example: Credit scoring
Differentiating between
low-riskand high-risk
customers from their
incomeand savings
Discriminant:IF income> θ
1AND savings> θ
2
THEN low-risk ELSE high-risk
Model

28. Classification: Applications
Pattern recognition
Face recognition: Pose, lighting, occlusion (glasses,
beard), make-up, hair style
Character recognition: Different handwriting styles.
Speech recognition: Temporal dependency.
Use of a dictionary or the syntax of the language.
Sensor fusion: Combine multiple modalities; eg, visual (lip
image) and acoustic for speech
Medical diagnosis: From symptoms to illnesses
Web Advertizing: Predict if a user clicks on an ad on the
Internet.
28

29. Face Recognition
29
Training examples of a person
Test images
AT&T Laboratories, Cambridge UK
http://www.uk.research.att.com/facedatabase.html

30. Example: Price of a used car
x : car attributes
y : price
y = g (x | θ)
g ( ) model,
θparameters
30
Prediction: Regression
y = wx+w
0

31. Regression Applications
Navigating a car: Angle of the steering wheel (CMU NavLab)
Kinematics of a robot arm
31
α
1= g
1(x,y)
α
2= g
2(x,y)
α
1
α
2
(x,y)

32. Supervised Learning: Uses
Prediction of future cases: Use the rule to predict the
output for future inputs
Knowledge extraction: The rule is easy to understand
Compression: The rule is simpler than the data it
explains
Outlier detection: Exceptions that are not covered by
the rule, e.g., fraud
32
Example: decision trees tools that create rules

33. Unsupervised Learning
Learning “what normally happens”
No output
Clustering: Grouping similar instances
Other applications: Summarization, Association Analysis
Example applications
Customer segmentation in CRM
Image compression: Color quantization
Bioinformatics: Learning motifs
33

34. Reinforcement Learning
Topics:
Policies: what actions should an agent take in a particular
situation
Utility estimation: how good is a state (used by policy)
No supervised output but delayed reward
Credit assignment problem(what was responsible for
the outcome)
Applications:
Game playing
Robot in a maze
Multiple agents, partial observability, ...
34

35. validation set
usesasubsetofthetrainingdatatoprovideanunbiased
evaluation of a model.
Thevalidationdatasetcontrastswithtrainingandtest
setsinthatitisanintermediatephaseusedforchoosing
thebestmodelandoptimizingit.
Alpydin & Ch. Eick: ML Topic1 35

36. Alpydin & Ch. Eick: ML Topic1 36

37. Alpydin & Ch. Eick: ML Topic1 37