Machine Learning

(BITS F464)

Dr.N.L.Bhanu Murthy
BITS Pilani
Hyderabad Campus
What is Learning?

“Gain knowledge or understanding of or skill in by study,

instruction or experience” - Webster

BITS Pilani, Hyderabad Campus

What is Learning?

“Learning is any process by which a system improves

performance from experience.” - Herbert Simon
Researcher in Professor @
 Artificial Intelligence  Carnegie Mellon University
 Cognitive psychology  University of California,
 Computer science Berkeley
 Economics  Illinois Institute of Technology
 Political science

Awards:
 Turing Award, 1975
 Nobel Prize in Economics1978
 National Medal of Science1986 1916 - 2001
 von Neumann Theory Prize1988

BITS Pilani, Hyderabad Campus

What is Machine Learning?

Machine Learning is study of

algorithms that
 improve their performance P
 at some task T
 with experience E
Tom Mitchell (1990)

Well-defined learning task: <P,T,E>

BITS Pilani, Hyderabad Campus

Example - Machine Learning
Handwritten Digit Recognition
T: Recognizing hand-written words
P: Percentage of words correctly classified
E: Database of human-labeled images of handwritten words

BITS Pilani, Hyderabad Campus

Example - Machine Learning
T: Driving on four-lane highways using vision sensors
P: Average distance traveled before a human-judged error
E: A sequence of images and steering commands recorded while
observing a human driver.

BITS Pilani, Hyderabad Campus

Example - Machine Learning

BITS Pilani, Hyderabad Campus

Example - Machine Learning

BITS Pilani, Hyderabad Campus

Example - Machine Learning

Learning to drive an autonomous

vehicle (Pomerleau, 1989).

BITS Pilani, Hyderabad Campus

Example - Machine Learning

BITS Pilani, Hyderabad Campus

Example - Machine Learning

BITS Pilani, Hyderabad Campus

Examples of Successful Applications of Machine Learning

 Learning to recognize spoken words

(Lee, 1989; Waibel, 1989).

 Learning to classify new astronomical structures

(Fayyad et al., 1995).

 Learning to play world-class backgammon

(Tesauro 1992, 1995).

 Categorize email messages as spam or legitimate.

BITS Pilani, Hyderabad Campus

Gary Kasparov on loss to Deep Blue

• Human uses 1% calculation, 99% understanding

– based on patterns, drawing information from experience
• Machine opposite: 99% calculation 1% understanding
– though this understanding is growing

BITS Pilani, Hyderabad Campus

Machine Learning, a Magic?

No, more like gardening

 Seeds = Algorithms
 Nutrients = Data
 Gardener = You
 Plants = Programs

BITS Pilani, Hyderabad Campus

Machine Learning in Computer Science

Speech/Au Robotics
dio
Processing Planning
Natural
Language
Processing
Machine Vision/Image
Processing
Biomedical/Chemed
ical Learning
Informatics

Human Financial Modeling

Computer Analytics
Interaction

BITS Pilani, Hyderabad Campus

They said it!!
 “A breakthrough in machine learning would be worth ten Microsofts”
- Bill Gates, Chairman, Microsoft

 Machine learning is the hot new thing”

- John Hennessy, President, Stanford

 “Web rankings today are mostly a matter of machine learning”

- Prabhakar Raghavan, Dir. Research, Yahoo

 “Machine learning is going to result in a real revolution”

- Greg Papadopoulos, CTO, Sun

 “Machine learning is today’s discontinuity” - Jerry Yang, CEO, Yahoo

 “Machine learning is the next Internet”

- Tony Tether, Director, DARPA
BITS Pilani, Hyderabad Campus
Future prospects..

BITS Pilani, Hyderabad Campus

History of Technology

BITS Pilani, Hyderabad Campus

12 IT skills that employers can't say no to

1) Machine learning
2) Mobilizing applications
3) Wireless networking
4) Human-computer interface
5) Project management
6) General networking skills
7) Network convergence technicians
8) Open-source programming
9) Business intelligence systems
10) Embedded security
11) Digital home technology integration
12) .Net, C #, C ++, Java -- with an edge

BITS Pilani, Hyderabad Campus

History of Machine Learning

• 1950s
– Samuel’s checker player
– Selfridge’s Pandemonium
• 1960s:
– Neural networks: Perceptron
– Pattern recognition
– Learning in the limit theory
– Minsky and Papert prove limitations of Perceptron
• 1970s:
– Symbolic concept induction
– Winston’s arch learner
– Expert systems and the knowledge acquisition bottleneck
– Quinlan’s ID3
– Michalski’s AQ and soybean diagnosis
– Scientific discovery with BACON
– Mathematical discovery with AM

BITS Pilani, Hyderabad Campus

History of Machine Learning (cont.)
• 1980s:
– Advanced decision tree and rule learning
– Explanation-based Learning (EBL)
– Learning and planning and problem solving
– Utility problem
– Analogy
– Cognitive architectures
– Resurgence of neural networks (connectionism,
backpropagation)
– Valiant’s PAC Learning Theory
– Focus on experimental methodology
• 1990s
– Data mining
– Adaptive software agents and web applications
– Text learning
– Reinforcement learning (RL)
– Inductive Logic Programming (ILP)
– Ensembles: Bagging, Boosting, and Stacking
– Bayes Net learning
BITS Pilani, Hyderabad Campus
History of Machine Learning (cont.)

• 2000s
– Support vector machines
– Kernel methods
– Graphical models
– Statistical relational learning
– Transfer learning
– Sequence labeling
– Collective classification and structured outputs
– Computer Systems Applications
• Compilers
• Debugging
• Graphics
• Security (intrusion, virus, and worm detection)
– Email management
– Personalized assistants that learn
– Learning in robotics and vision

BITS Pilani, Hyderabad Campus

Why does Machine Learning need math?
 Calculus
– We need to identify the maximum likelihood, or minimum risk.
Optimization
– Integration allows the marginalization of continuous probability
density functions
 Linear Algebra
– Many features leads to high dimensional spaces
– Vectors and matrices allow us to compactly describe and manipulate
high dimensional feature spaces.
 Vector Calculus
– All of the optimization needs to be performed in high dimensional
spaces
– Optimization of multiple variables simultaneously – Gradient Descent
– Want to take a marginal over high dimensional distributions like
Gaussians.
BITS Pilani, Hyderabad Campus
Teaching & Evaluation (BITS C464 – L P U – 3 0 3)

Evaluation Components & Criteria

Component Weightage Duration Date Mode
(out of 200)
Mid Test 50 90 minutes As per Closed Book
Timetable
Quiz(2) 30 (each quiz 30 minutes Closed Book
TBD
for 15)
Assignments 40 Open Book

Comprehensive 80 3 hours As per Closed Book

Timetable

Make-up Policy: Make-up for other tests will be granted on prior permission and on
justifiable grounds only.

Course Notices: All notices pertaining to this course will be displayed on the LTC
Notice Board as well as the CS & IS Notice Board.

Chamber Consultation: Friday 1600 Hrs – 1700 Hrs

BITS Pilani, Hyderabad Campus
Text Book

T1. Christopher Bishop: Pattern Recognition and Machine Learning,

Springer International Edition.

BITS Pilani, Hyderabad Campus

Reference Book

R1. Tom M. Mitchell: Machine Learning, The McGraw-Hill Companies, Inc.

BITS Pilani, Hyderabad Campus

Reference Book

BITS Pilani, Hyderabad Campus

Reference Book

BITS Pilani, Hyderabad Campus

Reference Book

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

BITS Pilani, Hyderabad Campus

Supervised Learning

Decision Tree Learning

 Target Concept
“Days on which my friend, yar, enjoys his favorite water sport”
(you may find it more intuitive to think of
“Days on which the beach will be crowded” concept)
 Task
Learn to predict the value of EnjoySport/Crowded for an arbitrary day

 Training Examples for the Target Concept

Example Sky Air Humidity Wind Water Forecast Enjoy

Temp Sport
0 Sunny Warm Normal Strong Warm Same Yes
1 Sunny Warm High Strong Warm Same Yes
2 Rainy Cold High Strong Warm Change No
3 Sunny Warm High Strong Cool Change Yes

BITS Pilani, Hyderabad Campus

Supervised Learning
Decision Tree Learning
 Hypothesis space search
 Occam’s razor
 Overfitting
 Measure for Selecting attributes –
Entropy, Gini Index etc.
 Issues in DT learning
Outlook

Sunny Overcast Rain

Humidity Yes Wind

High Normal Strong Weak

No Yes No Yes
BITS Pilani, Hyderabad Campus
Generative and Discriminative Models: An analogy

 Generative approach is to learn each language and determine

as to which language the speech belongs to

 Discriminative approach is determine the linguistic differences

without learning any language– a much easier task!

BITS Pilani, Hyderabad Campus

Taxonomy of ML Models

BITS Pilani, Hyderabad Campus

Supervised Learning
denotes +1
Support Vector Machine (SVM)
denotes -1
x2

V. Vapnik

x1

BITS Pilani, Hyderabad Campus

Supervised Learning
denotes +1
Support Vector Machine (SVM)
denotes -1
x2

V. Vapnik

x1

BITS Pilani, Hyderabad Campus

Supervised Learning denotes +1
denotes -1
x2
Support Vector Machine (SVM)

V. Vapnik

x1

BITS Pilani, Hyderabad Campus

Supervised Learning denotes +1
denotes -1
x2
Support Vector Machine (SVM)

V. Vapnik

x1

BITS Pilani, Hyderabad Campus

Supervised Learning denotes +1
denotes -1
x2
Support Vector Machine (SVM) Margin
“safe zone”

V. Vapnik

x1

BITS Pilani, Hyderabad Campus

Supervised Learning
Artificial Neural Networks (ANN)

 Networks of processing units (neurons) with connections

(synapses) between them

 Large number of neurons: 1014

 Large connectitivity: 104

 Parallel processing

 Distributed computation/memory

 Robust to noise, failures

BITS Pilani, Hyderabad Campus

Supervised Learning

Regression

 Polynomial Curve Fitting

 Model Selection
 Overfitting & Regularization
 Probabilistic interpretation
 Bayesian curve fitting
 Linear Basis Function Models
 Bias – Variance Decomposition
 Bayesian Linear Regression

BITS Pilani, Hyderabad Campus

Unsupervised Learning

 Clustering Algorithms

BITS Pilani, Hyderabad Campus

Model / Hypothesis Evaluation
t Tests

Precision

Recall

F-Measure

AuC for ROC curves

R2

Spearman / Pearson Correlation

BITS Pilani, Hyderabad Campus
Thank You!!

BITS Pilani, Hyderabad Campus